starting discussions: weak and strong IDSO

[Mikhail.Prokopenko]

Dear all,

 

Now that the list is operational, it would make sense to start our discussions (all the typical email-list rules apply) :)

 

There are at least two main motivations for Information-driven Self-organisation (IDSO).

 

a) A weak IDSO: many evolutionary/selection/self-organisation pressures can be approximated information-theoretically – and such approximations can serve as shortcuts in explaining natural phenomena as well as in designing biologically-inspired systems (epistemological view on IDSO?)

 

b) A strong IDSO: many optimal structures evolve/self-organise in nature when information transfer within certain channels is maximised – this view maintains that maximization of information transfer through selected channels is one of the main evolutionary pressures (ontological view on IDSO?)

 

There may be other views, but in this message we would like to elaborate on the strong notion, following our recent work [Piraveenan et al., 2007]: Although the evolutionary process involves a larger number of drives and constraints, information fidelity (i.e. preservation) is a consistent motif throughout biology: e.g., modern evolution operates close to the error threshold [Adami, 1998], and biological sensorimotor equipment typically exhausts the available informatory capacity (under given constraints) close to the limit [Laughlin et al., 1998]. Adami, in fact, argues that the evolutionary process extracts valuable information and stores it in the genes. Since this process is relatively slow [Zurek, 1990], it is a selective advantage to preserve this information, once captured.

 

In other words, constraints (like noise in the environment) reduce the channel's bandwidth, and the system evolves to preserve information by self-improving, re-structuring, and so on.

 

Comments?

 

Thanks,

Daniel Polani and Mikhail Prokopenko

 

Refs:

 

Adami, C., 1998: Introduction to Artificial Life. Springer.

 

Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998: The metabolic cost of neural information. Nature Neuroscience 1(1), 36–41.

 

Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of Genetic Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M. Rocha, E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial Life: 9th European Conference on Artificial Life (ECAL-2007), Lisbon, Portugal, September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648, pp. 42-52, Springer, Berlin.

 

Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.: Complexity, Entropy and the Physics of Information. Santa Fe Studies in the Sciences of Complexity, Reading, Mass., Addison-Wesley.
==============================================

 

Please use "Reply to All" if you intend your message to go to the list. Thanks.

 

-------------------------------------------------------
Dr Mikhail Prokopenko
Senior Research Scientist, Adaptive Systems Team Leader
Autonomous Systems Lab, CSIRO ICT Centre, Australia
http://www.ict.csiro.au/asl/
phone [phone]
http://www.ict.csiro.au/staff/Mikhail.Prokopenko/

 

[Antonio Lafusa]

Hi,

I have a sort of counter-example about IDSO,
and the fact that information exchange is used in order to predict
self-organisation.

In a system called sodaplay (www.sodaplay.com), structures made of
bodies connected by springs can produce a coordinated behaviour (e.g.
walk). In this case, there is no information exchange beween bodies.
They actually coordinate the movement through the physics.

Even a snakebot can be equipped with joints that oscillate with a
given frequency, amplitude and phase, such that its behaviour
corresponds to walk straight, with no "channels" for communication.

I went through this because I am trying to implement a minimal control
effort (MCE) for my modular robots. The joints (between modules)
oscillate with a fixed phase and
amplitude. Therefore, the  behaviour depends just on the topology of
the structure. There is no exchange of information, except for the
growth and division mechanism.
The growth (i.e. aggregation of a module) and division (cut of a
physical link) are determined by a local rule (CA-like) and depends on
the internal state.

In this case, to have an efficient design, we aim to a minimal control
effort, which in turn minimise the exchange of information (instead of
maximising it), without loosing the abilities selected by the
environment, in order to survive.

Antonio


On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:

>
>
> Dear all,
>
> Now that the list is operational, it would make sense to start our
> discussions (all the typical email-list rules apply) :)
>
> There are at least two main motivations for Information-driven
> Self-organisation (IDSO).
>
> a) A weak IDSO: many evolutionary/selection/self-organisation pressures can
> be approximated information-theoretically – and such approximations can
> serve as shortcuts in explaining natural phenomena as well as in designing
> biologically-inspired systems (epistemological view on IDSO?)
>
> b) A strong IDSO: many optimal structures evolve/self-organise in nature
> when information transfer within certain channels is maximised – this view
> maintains that maximization of information transfer through selected
> channels is one of the main evolutionary pressures (ontological view on
> IDSO?)
>
> There may be other views, but in this message we would like to elaborate on
> the strong notion, following our recent work [Piraveenan et al., 2007]:
> Although the evolutionary process involves a larger number of drives and
> constraints, information fidelity (i.e. preservation) is a consistent motif
> throughout biology: e.g., modern evolution operates close to the error
> threshold [Adami, 1998], and biological sensorimotor equipment typically
> exhausts the available informatory capacity (under given constraints) close
> to the limit [Laughlin et al., 1998]. Adami, in fact, argues that the
> evolutionary process extracts valuable information and stores it in the
> genes. Since this process is relatively slow [Zurek, 1990], it is a
> selective advantage to preserve this information, once captured.
>
> In other words, constraints (like noise in the environment) reduce the
> channel's bandwidth, and the system evolves to preserve information by
> self-improving, re-structuring, and so on.
>
> Comments?
>
> Thanks,
> Daniel Polani and Mikhail Prokopenko
>
> Refs:
>
> Adami, C., 1998: Introduction to Artificial Life. Springer.
>
> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998: The
> metabolic cost of neural information. Nature Neuroscience 1(1), 36–41.
>
> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of Genetic
> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M. Rocha,
> E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial Life: 9th
> European Conference on Artificial Life (ECAL-2007), Lisbon, Portugal,
> September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648, pp.
> 42-52, Springer, Berlin.
>
> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
> Complexity, Entropy and the Physics of Information. Santa Fe Studies in the
> Sciences of Complexity, Reading, Mass., Addison-Wesley.
> ==============================================
>
> Please use "Reply to All" if you intend your message to go to the list.
> Thanks.
>
> -------------------------------------------------------
> Dr Mikhail Prokopenko
> Senior Research Scientist, Adaptive Systems Team Leader
> Autonomous Systems Lab, CSIRO ICT Centre, Australia
> http://www.ict.csiro.au/asl/
> phone [phone]
> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>
>

[Juergen Pahle]

Dear Antonio,

I am not really sure why this would be a counter-example of IDSO.

There is no need to have specific control systems in order to have
information transfer. For example in animals, the nervous system and the
hormonal system are used to transfer information (using quite different
strategies). Surely, there is also information transfer involved when a
cell just physically touches a surface (and there are cellular systems
to transduce that information). In your case, I guess, the bodies just
exchange information physically. This information transfer is then
implicitly optimized, e.g. in terms of delay of transmission, by
reconfiguring the system.
Maybe, I missed some important point. It would be nice if you could
elaborate a little more. Thank you.

Best wishes,
Juergen


Antonio Lafusa wrote:

> Hi,
>
> I have a sort of counter-example about IDSO,
> and the fact that information exchange is used in order to predict
> self-organisation.
>
> In a system called sodaplay (www.sodaplay.com), structures made of
> bodies connected by springs can produce a coordinated behaviour (e.g.
> walk). In this case, there is no information exchange beween bodies.
> They actually coordinate the movement through the physics.
>
> Even a snakebot can be equipped with joints that oscillate with a
> given frequency, amplitude and phase, such that its behaviour
> corresponds to walk straight, with no "channels" for communication.
>
> I went through this because I am trying to implement a minimal control
> effort (MCE) for my modular robots. The joints (between modules)
> oscillate with a fixed phase and
> amplitude. Therefore, the  behaviour depends just on the topology of
> the structure. There is no exchange of information, except for the
> growth and division mechanism.
> The growth (i.e. aggregation of a module) and division (cut of a
> physical link) are determined by a local rule (CA-like) and depends on
> the internal state.
>
> In this case, to have an efficient design, we aim to a minimal control
> effort, which in turn minimise the exchange of information (instead of
> maximising it), without loosing the abilities selected by the
> environment, in order to survive.
>
> Antonio

[Oliver Obst]

Hi all,

On 21/11/2007, at 9:49 PM, Antonio Lafusa wrote:

> I have a sort of counter-example about IDSO,
> and the fact that information exchange is used in order to predict
> self-organisation.
>
> In a system called sodaplay (www.sodaplay.com), structures made of
> bodies connected by springs can produce a coordinated behaviour (e.g.
> walk). In this case, there is no information exchange beween bodies.
> They actually coordinate the movement through the physics.

the system is nice.
I'm not sure if it is a counter example.
What does it mean that connected structures can produce a coordinated  
behavior? They can be designed like this, in a similar way coupled  
neural oscillators can produce stable oscillations with no external  
input.
I think the absence of external input doesn't mean there's no exchange  
of information: the impact of the environment on the oscillation is  
already some information to the system.

> Even a snakebot can be equipped with joints that oscillate with a
> given frequency, amplitude and phase, such that its behaviour
> corresponds to walk straight, with no "channels" for communication.

true. But to stable evolve it to walk straight (or to learn a straight  
walk) there would be some information necessary (?).

> In this case, to have an efficient design, we aim to a minimal control
> effort, which in turn minimise the exchange of information (instead of
> maximising it), without loosing the abilities selected by the
> environment, in order to survive.


Mmh, doesn't that mean you want to maximize the (use of) information  
on the available channels?
How do you measure your control effort?

cheers
Oliver

--
Oliver Obst                     form follows function (Louis Sullivan).
Fon: [phone]                             http://oliver.obst.eu/
Autonomous Systems Lab  CSIRO ICT Centre   http://www.ict.csiro.au/asl/

[Mikhail.Prokopenko]

Hi all,
 
I think Antonio raised an interesting point.
 
It seems to me that information is still exchanged in systems like these (e.g., sodaplay or snakebots) - however, it doesn't require explicit module-to-module communication channels with packets being sent. Rather, the physical coupling creates the channel (like Juergen and Oliver mentioned): if the states of the remote modules become synchronised then this means that information has been exchanged - via "stigmergy" of physical constraints.
 
In other words, a stigmergic information flow may minimise the control effort in individual modules - but this would need to be studied precisely.
 
Thanks!
Mikhail
 

        -----Original Message-----
        From: Lafusa, Antonio
        Sent: Wed 21/11/2007 9:49 PM
        To: Prokopenko, Mikhail (ICT Centre, North Ryde)
        Cc: IDSO-CSIRO
        Subject: Re: starting IDSO discussions: weak and strong IDSO
       
       

        Hi,
       
        I have a sort of counter-example about IDSO,
        and the fact that information exchange is used in order to predict
        self-organisation.
       
        In a system called sodaplay (www.sodaplay.com), structures made of
        bodies connected by springs can produce a coordinated behaviour (e.g.
        walk). In this case, there is no information exchange beween bodies.
        They actually coordinate the movement through the physics.
       
        Even a snakebot can be equipped with joints that oscillate with a
        given frequency, amplitude and phase, such that its behaviour
        corresponds to walk straight, with no "channels" for communication.
       
        I went through this because I am trying to implement a minimal control
        effort (MCE) for my modular robots. The joints (between modules)
        oscillate with a fixed phase and
        amplitude. Therefore, the  behaviour depends just on the topology of
        the structure. There is no exchange of information, except for the
        growth and division mechanism.
        The growth (i.e. aggregation of a module) and division (cut of a
        physical link) are determined by a local rule (CA-like) and depends on
        the internal state.
       
        In this case, to have an efficient design, we aim to a minimal control
        effort, which in turn minimise the exchange of information (instead of
        maximising it), without loosing the abilities selected by the
        environment, in order to survive.
       
        Antonio
       
       
        On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
        >
        >
        > Dear all,
        >
        > Now that the list is operational, it would make sense to start our
        > discussions (all the typical email-list rules apply) :)
        >
        > There are at least two main motivations for Information-driven
        > Self-organisation (IDSO).
        >
        > a) A weak IDSO: many evolutionary/selection/self-organisation pressures can
        > be approximated information-theoretically – and such approximations can
        > serve as shortcuts in explaining natural phenomena as well as in designing
        > biologically-inspired systems (epistemological view on IDSO?)
        >
        > b) A strong IDSO: many optimal structures evolve/self-organise in nature
        > when information transfer within certain channels is maximised – this view
        > maintains that maximization of information transfer through selected
        > channels is one of the main evolutionary pressures (ontological view on
        > IDSO?)
        >
        > There may be other views, but in this message we would like to elaborate on
        > the strong notion, following our recent work [Piraveenan et al., 2007]:
        > Although the evolutionary process involves a larger number of drives and
        > constraints, information fidelity (i.e. preservation) is a consistent motif
        > throughout biology: e.g., modern evolution operates close to the error
        > threshold [Adami, 1998], and biological sensorimotor equipment typically
        > exhausts the available informatory capacity (under given constraints) close
        > to the limit [Laughlin et al., 1998]. Adami, in fact, argues that the
        > evolutionary process extracts valuable information and stores it in the
        > genes. Since this process is relatively slow [Zurek, 1990], it is a
        > selective advantage to preserve this information, once captured.
        >
        > In other words, constraints (like noise in the environment) reduce the
        > channel's bandwidth, and the system evolves to preserve information by
        > self-improving, re-structuring, and so on.
        >
        > Comments?
        >
        > Thanks,
        > Daniel Polani and Mikhail Prokopenko
        >
        > Refs:
        >
        > Adami, C., 1998: Introduction to Artificial Life. Springer.
        >
        > Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998: The
        > metabolic cost of neural information. Nature Neuroscience 1(1), 36–41.
        >
        > Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of Genetic
        > Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M. Rocha,
        > E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial Life: 9th
        > European Conference on Artificial Life (ECAL-2007), Lisbon, Portugal,
        > September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648, pp.
        > 42-52, Springer, Berlin.
        >
        > Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
        > Complexity, Entropy and the Physics of Information. Santa Fe Studies in the
        > Sciences of Complexity, Reading, Mass., Addison-Wesley.
        > ==============================================
        >
        > Please use "Reply to All" if you intend your message to go to the list.
        > Thanks.
        >
        > -------------------------------------------------------
        > Dr Mikhail Prokopenko
        > Senior Research Scientist, Adaptive Systems Team Leader
        > Autonomous Systems Lab, CSIRO ICT Centre, Australia
        > http://www.ict.csiro.au/asl/
        > phone [phone]
        > http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
        >
        >
       

[Carlos Gershenson-2]

Hi all,

I agree with Mikhail, Juergen, and Olivier: I think you can describe information transfer in systems where there is no explicit information transfer, i.e. the designer didn't put it there. An example is with self-organizing traffic lights:

http://homepages.vub.ac.be/~cgershen/sos/SOTL/SOTL.html

(just press the GO button and wait a bit...)

There is no direct communication between traffic lights: they just sense how many cars are approaching. Still, they are able to synchronize. This is because the cars can be seen as information that is sent from one traffic light to the next.

I am quite sympathetic with weak IDSO: information seems to be a good formalism (along many others) to describe, understand, and build systems.

About strong IDSO, it makes sense, but I am not so sure about it. It seems problematic to me, because having more information does not always give advantages. It has to be "useful" information, for the current context/niche. An example with koalas: 30% of their cranium is empty. In their ecological niche, they do not need to process much information, as there were no predators, and they basically eat and sleep all day. What for to waste energy in brainpower when food is scarce? In this case, evolution seems to have favored less information processing, as this came with an energetic cost that didn't pay off...

Other example can be seen with parasites that exploit the information of their host (even at the genetic level).

If we wanted to speak about strong IDSO, we need to speak already about the "usefulness" of the information. I think it is possible, but we need to consider the observer who judges the usefulness or purpose of the information (and we'll enter the teleological debates...)

Still, for engineering this is not a problem: the engineer decides the purpose of the system, so IDSO can be a useful principle to use. But then this might be seen more as weak IDSO...

Best regards,

Carlos Gershenson New England Complex Systems Institute 24 Mt. Auburn St. Cambridge MA 02138, USA

[Stanley N. Salthe]

Antonio --  Physics does not escape being informed.  Following Howard
Pattee's thinking, we can distinguish between dynamics and information
contstraining the dynamics.  Its as general as, say, Y = aXpowerb.  The
relation between X and Y describe the dynamical results (implicitly the
dynamics), but the values of a and b are information.

STAN

>Hi,
>
>I have a sort of counter-example about IDSO,
>and the fact that information exchange is used in order to predict
>self-organisation.
>
>In a system called sodaplay (www.sodaplay.com), structures made of
>bodies connected by springs can produce a coordinated behaviour (e.g.
>walk). In this case, there is no information exchange beween bodies.
>They actually coordinate the movement through the physics.
>
>Even a snakebot can be equipped with joints that oscillate with a
>given frequency, amplitude and phase, such that its behaviour
>corresponds to walk straight, with no "channels" for communication.
>
>I went through this because I am trying to implement a minimal control
>effort (MCE) for my modular robots. The joints (between modules)
>oscillate with a fixed phase and
>amplitude. Therefore, the  behaviour depends just on the topology of
>the structure. There is no exchange of information, except for the
>growth and division mechanism.
>The growth (i.e. aggregation of a module) and division (cut of a
>physical link) are determined by a local rule (CA-like) and depends on
>the internal state.
>
>In this case, to have an efficient design, we aim to a minimal control
>effort, which in turn minimise the exchange of information (instead of
>maximising it), without loosing the abilities selected by the
>environment, in order to survive.
>
>Antonio
>
>
>On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
>>
>>
>> Dear all,
>>
>> Now that the list is operational, it would make sense to start our
>> discussions (all the typical email-list rules apply) :)
>>
>> There are at least two main motivations for Information-driven
>> Self-organisation (IDSO).
>>
>> a) A weak IDSO: many evolutionary/selection/self-organisation pressures can
>> be approximated information-theoretically ñ and such approximations can
>> serve as shortcuts in explaining natural phenomena as well as in designing
>> biologically-inspired systems (epistemological view on IDSO?)
>>
>> b) A strong IDSO: many optimal structures evolve/self-organise in nature
>> when information transfer within certain channels is maximised ñ this view
>> maintains that maximization of information transfer through selected
>> channels is one of the main evolutionary pressures (ontological view on
>> IDSO?)
>>
>> There may be other views, but in this message we would like to elaborate on
>> the strong notion, following our recent work [Piraveenan et al., 2007]:
>> Although the evolutionary process involves a larger number of drives and
>> constraints, information fidelity (i.e. preservation) is a consistent motif
>> throughout biology: e.g., modern evolution operates close to the error
>> threshold [Adami, 1998], and biological sensorimotor equipment typically
>> exhausts the available informatory capacity (under given constraints) close
>> to the limit [Laughlin et al., 1998]. Adami, in fact, argues that the
>> evolutionary process extracts valuable information and stores it in the
>> genes. Since this process is relatively slow [Zurek, 1990], it is a
>> selective advantage to preserve this information, once captured.
>>
>> In other words, constraints (like noise in the environment) reduce the
>> channel's bandwidth, and the system evolves to preserve information by
>> self-improving, re-structuring, and so on.
>>
>> Comments?
>>
>> Thanks,
>> Daniel Polani and Mikhail Prokopenko
>>
>> Refs:
>>
>> Adami, C., 1998: Introduction to Artificial Life. Springer.
>>
>> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998: The
>> metabolic cost of neural information. Nature Neuroscience 1(1), 36ñ41.
>>
>> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of Genetic
>> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M. Rocha,
>> E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial Life: 9th
>> European Conference on Artificial Life (ECAL-2007), Lisbon, Portugal,
>> September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648, pp.
>> 42-52, Springer, Berlin.
>>
>> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
>> Complexity, Entropy and the Physics of Information. Santa Fe Studies in the
>> Sciences of Complexity, Reading, Mass., Addison-Wesley.
>> ==============================================
>>
>> Please use "Reply to All" if you intend your message to go to the list.
>> Thanks.
>>
>> -------------------------------------------------------
>> Dr Mikhail Prokopenko
>> Senior Research Scientist, Adaptive Systems Team Leader
>> Autonomous Systems Lab, CSIRO ICT Centre, Australia
>> http://www.ict.csiro.au/asl/
>> phone [phone]
>> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>>
>>

[Hussein Abbass]

I think Antonio raised an important issue, what do we mean with information.

The purpose of this email is (1) to share with you some of my humble views; and (2) to provoke a discussion about this question. In my opinion, without clarifying what we mean with information, it is hard to understand what IDSO really mean. Is IDSO an acronym for "everything" driven self-organisation or it denotes a subset of "everything" called "information", in which case what do mean with "information".

Avoiding the traditional debate about the distinction between data, information, knowledge and wisdom, I think about information when modelling a problem in three categories:

*) messages being explicitly exchanged between system components or between the components and the environment. I define the environment to be anything external to the system including the physical environment. So for an agent X, other agents are part of the environment were X is situated. This also includes any information that X does not know about, which form the information environment of X. So we have an information environment, a social environment, a physical environment, etc.

*) constraints among the system components (internal) and constraints set by the environment on the system (external). I see constraints as meta-information which are either (a) evolved over time such as traditions and ethics; (b) enforced on the system by the designer such as the way we connect parts to each other and how many degrees of freedom we allow when we design a joint; (b) enforced on the system by the environment such as physics (I do not think the laws of physics have evolved but we can debate about this). Constraints can be represented in many forms: by physical connection of parts, mathematically, using symbolic knowledge, etc. Constraints can be hard (the agent has no choice) or soft (the agent can break them with a cost).

*) objectives which represent the internal evaluation mechanism of the components and the overall system. An objective is a function that evaluates the performance of the agent. The agent can use it to eliminate states, and evaluate its performance in the environment. If the evaluation of the objective occurs internal to the agent, it becomes like an agent utility function. If the evaluation of the objective occurs by the environment, it becomes like a reinforcement function. Wherever the evaluation takes place, it does not change the fact that the agent uses this information to correct its actions and update its internal states (i.e. knowledge which becomes constraints on future agents' behaviour).

In this way, I can define an agent behaviour as a function of the messages, constraints, and objectives. This behaviour can take one or a mix of three forms: (1) actions that affect other agents (2) actions that affect the environment (3) actions that affect the agent itself. Each of these actions can change the messages, constraints and objectives of an agent. The way this change occurs is by a subset of the constraints and objectives responsible for handling the dynamics.

This might be an engineering oriented view of the world, but it would be nice to hear people views on that issue and examples of when this taxnomy breaks.

I also agree with Carlos that we need to define "usefulness" of information. But I would like to generalize "useful" information for an agent to also include the cognitive ability of the agent to sense and process this information; otherwise the agent will be faced with information overload. Therefore, maximising information should be subject to two constraints: "relevance" and agent's capacity to use the information.

Cheers
Hussein



-----Original Message-----
From: Stanley N. Salthe [mailto:ssalthe@...]
Sent: Thursday, 22 November 2007 2:46 AM
To: Antonio Lafusa
Cc: idso@...
Subject: Re: starting IDSO discussions: weak and strong IDSO

Antonio --  Physics does not escape being informed.  Following Howard Pattee's thinking, we can distinguish between dynamics and information contstraining the dynamics.  Its as general as, say, Y = aXpowerb.  The relation between X and Y describe the dynamical results (implicitly the dynamics), but the values of a and b are information.

STAN

>Hi,
>
>I have a sort of counter-example about IDSO, and the fact that
>information exchange is used in order to predict self-organisation.
>
>In a system called sodaplay (www.sodaplay.com), structures made of
>bodies connected by springs can produce a coordinated behaviour (e.g.
>walk). In this case, there is no information exchange beween bodies.
>They actually coordinate the movement through the physics.
>
>Even a snakebot can be equipped with joints that oscillate with a given
>frequency, amplitude and phase, such that its behaviour corresponds to
>walk straight, with no "channels" for communication.
>
>I went through this because I am trying to implement a minimal control
>effort (MCE) for my modular robots. The joints (between modules)
>oscillate with a fixed phase and amplitude. Therefore, the  behaviour
>depends just on the topology of the structure. There is no exchange of
>information, except for the growth and division mechanism.
>The growth (i.e. aggregation of a module) and division (cut of a
>physical link) are determined by a local rule (CA-like) and depends on
>the internal state.
>
>In this case, to have an efficient design, we aim to a minimal control
>effort, which in turn minimise the exchange of information (instead of
>maximising it), without loosing the abilities selected by the
>environment, in order to survive.
>
>Antonio
>
>
>On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
>>
>>
>> Dear all,
>>
>> Now that the list is operational, it would make sense to start our
>> discussions (all the typical email-list rules apply) :)
>>
>> There are at least two main motivations for Information-driven
>> Self-organisation (IDSO).
>>
>> a) A weak IDSO: many evolutionary/selection/self-organisation
>> pressures can be approximated information-theoretically ñ and such
>> approximations can serve as shortcuts in explaining natural phenomena
>> as well as in designing biologically-inspired systems
>> (epistemological view on IDSO?)
>>
>> b) A strong IDSO: many optimal structures evolve/self-organise in
>> nature when information transfer within certain channels is maximised
>> ñ this view maintains that maximization of information transfer
>> through selected channels is one of the main evolutionary pressures
>> (ontological view on
>> IDSO?)
>>
>> There may be other views, but in this message we would like to
>> elaborate on the strong notion, following our recent work [Piraveenan et al., 2007]:
>> Although the evolutionary process involves a larger number of drives
>> and constraints, information fidelity (i.e. preservation) is a
>> consistent motif throughout biology: e.g., modern evolution operates
>> close to the error threshold [Adami, 1998], and biological
>> sensorimotor equipment typically exhausts the available informatory
>> capacity (under given constraints) close to the limit [Laughlin et
>> al., 1998]. Adami, in fact, argues that the evolutionary process
>> extracts valuable information and stores it in the genes. Since this
>> process is relatively slow [Zurek, 1990], it is a selective advantage to preserve this information, once captured.
>>
>> In other words, constraints (like noise in the environment) reduce
>> the channel's bandwidth, and the system evolves to preserve
>> information by self-improving, re-structuring, and so on.
>>
>> Comments?
>>
>> Thanks,
>> Daniel Polani and Mikhail Prokopenko
>>
>> Refs:
>>
>> Adami, C., 1998: Introduction to Artificial Life. Springer.
>>
>> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998:
>> The metabolic cost of neural information. Nature Neuroscience 1(1), 36ñ41.
>>
>> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of
>> Genetic
>> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M.
>> Rocha, E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial
>> Life: 9th European Conference on Artificial Life (ECAL-2007), Lisbon,
>> Portugal, September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648, pp.
>> 42-52, Springer, Berlin.
>>
>> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
>> Complexity, Entropy and the Physics of Information. Santa Fe Studies
>> in the Sciences of Complexity, Reading, Mass., Addison-Wesley.
>> ==============================================
>>
>> Please use "Reply to All" if you intend your message to go to the list.
>> Thanks.
>>
>> -------------------------------------------------------
>> Dr Mikhail Prokopenko
>> Senior Research Scientist, Adaptive Systems Team Leader Autonomous
>> Systems Lab, CSIRO ICT Centre, Australia http://www.ict.csiro.au/asl/ 
>> phone [phone]
>> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>>
>>

Prof. (Chair of IT) Dr. Hussein A. Abbass, SMIEEE, SMACS

School of Information Technology and Electrical Engineering,

UNSW@ADFA, Australian Defence Force Academy Campus,

Northcott Drive, Campbell, Canberra, ACT 2600, Australia.

Web: http://www.itee.adfa.edu.au/~abbass 

Lab Web: http://www.itee.adfa.edu.au/~alar 

Email: h.abbass@...

Tel. (+61) (2) 62688158

Mob. [phone]

Fax (+61) (2) 62688581

PS: if you send an email to me and it does not get through our system, you can re-send it to hussein.abbass@...

This message is intended for the addressee named and may contain confidential information. If you are not the intended recipient, please delete it and notify the sender. Views expressed in this message are those of the individual sender and are not necessarily the views of the University of New South Wales Campus at the Australian Defence Force Academy in Canberra.

CRICOS Provider Number 00100G

[Berryman, Matthew]

Re the sodaplay, it could be said that the physical dynamics is
transferring information through the springs. The information on the
state (position, momentum) of an atom is conveyed to its neighbouring
atoms via photons. It's been a while since I studied physics though ;)

I agree with Hussein's email, which makes sense in the case of
engineering / engineered systems, but it does break down when one
considers biology: what is the objective, where does one draw the system
boundary (e.g. in the case of obligate symbiotes) etc.
 
"it is a selective advantage to preserve this information, once
captured." <- well it all depends on context, once the context changes
(the organism shifts to a new environment) then the information may end
up under no selectionary pressure. Parasites are an interesting case,
since they may lose information, but the system (parasite + host)
increases in the complexity and information (though the relationship
between the two isn't easily summed up by mathematics :)

I'm wary of strong IDSO / any teleological arguments. The distribution
of complexity amongst life forms spreads out to the right (ie the
maximum tends to increase) because there is a minimal complexity, and
evolution can build complexity (including efficient information
channels), though it can just as easily remove it. There's always
increases /and/ decreases in complexity and in information capacity
(consider moles' eyesight and hearing). Measures of any of this
(biological complexity, biological information channel capacity) are not
easy and depend highly on the context and how one defines the system.

"In other words, constraints (like noise in the environment) reduce the
channel's bandwidth"
Noise in the environment can increase a nonlinear channels bandwidth:
http://www.eleceng.adelaide.edu.au/personal/mmcdonne/Publications/index.
html

Regards,
Matthew Berryman
BH: [phone]
Fax: [phone]
Mobile: [phone]
Web: http://www.berrymanconsulting.com/


-----Original Message-----
From: Antonio Lafusa [mailto:alafusa@...]
Sent: Wednesday, 21 November 2007 9:19 PM
To: Mikhail.Prokopenko@...
Cc: idso@...
Subject: Re: starting IDSO discussions: weak and strong IDSO

Hi,

I have a sort of counter-example about IDSO, and the fact that
information exchange is used in order to predict self-organisation.

In a system called sodaplay (www.sodaplay.com), structures made of
bodies connected by springs can produce a coordinated behaviour (e.g.
walk). In this case, there is no information exchange beween bodies.
They actually coordinate the movement through the physics.

Even a snakebot can be equipped with joints that oscillate with a given
frequency, amplitude and phase, such that its behaviour corresponds to
walk straight, with no "channels" for communication.

I went through this because I am trying to implement a minimal control
effort (MCE) for my modular robots. The joints (between modules)
oscillate with a fixed phase and amplitude. Therefore, the  behaviour
depends just on the topology of the structure. There is no exchange of
information, except for the growth and division mechanism.
The growth (i.e. aggregation of a module) and division (cut of a
physical link) are determined by a local rule (CA-like) and depends on
the internal state.

In this case, to have an efficient design, we aim to a minimal control
effort, which in turn minimise the exchange of information (instead of
maximising it), without loosing the abilities selected by the
environment, in order to survive.

Antonio


On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:

>
>
> Dear all,
>
> Now that the list is operational, it would make sense to start our
> discussions (all the typical email-list rules apply) :)
>
> There are at least two main motivations for Information-driven
> Self-organisation (IDSO).
>
> a) A weak IDSO: many evolutionary/selection/self-organisation
> pressures can be approximated information-theoretically - and such
> approximations can serve as shortcuts in explaining natural phenomena
> as well as in designing biologically-inspired systems (epistemological

> view on IDSO?)
>
> b) A strong IDSO: many optimal structures evolve/self-organise in
> nature when information transfer within certain channels is maximised
> - this view maintains that maximization of information transfer
> through selected channels is one of the main evolutionary pressures
> (ontological view on
> IDSO?)
>
> There may be other views, but in this message we would like to
> elaborate on the strong notion, following our recent work [Piraveenan

et al., 2007]:
> Although the evolutionary process involves a larger number of drives
> and constraints, information fidelity (i.e. preservation) is a
> consistent motif throughout biology: e.g., modern evolution operates
> close to the error threshold [Adami, 1998], and biological
> sensorimotor equipment typically exhausts the available informatory
> capacity (under given constraints) close to the limit [Laughlin et
> al., 1998]. Adami, in fact, argues that the evolutionary process
> extracts valuable information and stores it in the genes. Since this
> process is relatively slow [Zurek, 1990], it is a selective advantage
to preserve this information, once captured.
>
> In other words, constraints (like noise in the environment) reduce the

> channel's bandwidth, and the system evolves to preserve information by

> self-improving, re-structuring, and so on.
>
> Comments?
>
> Thanks,
> Daniel Polani and Mikhail Prokopenko
>
> Refs:
>
> Adami, C., 1998: Introduction to Artificial Life. Springer.
>
> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998:
> The metabolic cost of neural information. Nature Neuroscience 1(1),

36-41.
>
> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of Genetic
> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M.
> Rocha, E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial
> Life: 9th European Conference on Artificial Life (ECAL-2007), Lisbon,
> Portugal, September 10-14, Lecture Notes in Artificial Intelligence,
vol. 4648, pp.
> 42-52, Springer, Berlin.
>
> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
> Complexity, Entropy and the Physics of Information. Santa Fe Studies
> in the Sciences of Complexity, Reading, Mass., Addison-Wesley.
> ==============================================
>
> Please use "Reply to All" if you intend your message to go to the
list.

> Thanks.
>
> -------------------------------------------------------
> Dr Mikhail Prokopenko
> Senior Research Scientist, Adaptive Systems Team Leader Autonomous
> Systems Lab, CSIRO ICT Centre, Australia http://www.ict.csiro.au/asl/ 
> phone [phone]
> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>
>

IMPORTANT: This email remains the property of the Australian Defence Organisation and is subject to the jurisdiction of section 70 of the CRIMES ACT 1914.  If you have received this email in error, you are requested to contact the sender and delete the email.

[High Performance Coder]

I don't think that information is a property of a physical system per
se, but rather a property of a model of a physical system.

For instance in thermodynamics, information is useful because it
applies to the thermodynamic model of the system, not the microscopic
dynamics.

In models employing self-organisation (a type of emergence) as a
concept, information is a highly relevant concept, and it makes sense
to ask questions about extremum principles of information flow about
such systems.

Note I have yet to read the ECAL07 paper, so I haven't chimed in until
now.

I have also registered this mailing list with nabble.com so as to
maintain an archive of postings. When this is online, I'll repost
everything that's been posted so far to it to complete the record.

Cheers

On Thu, Nov 22, 2007 at 10:50:11AM +1100, Hussein Abbass wrote:
>
> I think Antonio raised an important issue, what do we mean with information.
>

--

----------------------------------------------------------------------------
A/Prof Russell Standish                  Phone [phone] (mobile)
Mathematics                        
UNSW SYDNEY 2052                 hpcoder@...
Australia                                http://www.hpcoders.com.au
----------------------------------------------------------------------------

[Berryman, Matthew]

Ref from Adami on complexity:
http://www.pnas.org/cgi/content/full/97/9/4463

Note that they assume: "within a fixed environment"

Cheers,
Matt.

-----Original Message-----
From: Mikhail.Prokopenko@... [mailto:Mikhail.Prokopenko@...]
Sent: Wednesday, 21 November 2007 11:15 PM
To: alafusa@...
Cc: idso@...
Subject: RE: starting IDSO discussions: weak and strong IDSO

Hi all,
 
I think Antonio raised an interesting point.
 
It seems to me that information is still exchanged in systems like these
(e.g., sodaplay or snakebots) - however, it doesn't require explicit
module-to-module communication channels with packets being sent. Rather,
the physical coupling creates the channel (like Juergen and Oliver
mentioned): if the states of the remote modules become synchronised then
this means that information has been exchanged - via "stigmergy" of
physical constraints.
 
In other words, a stigmergic information flow may minimise the control
effort in individual modules - but this would need to be studied
precisely.
 
Thanks!
Mikhail
 

        -----Original Message-----
        From: Lafusa, Antonio
        Sent: Wed 21/11/2007 9:49 PM
        To: Prokopenko, Mikhail (ICT Centre, North Ryde)
        Cc: IDSO-CSIRO
        Subject: Re: starting IDSO discussions: weak and strong IDSO
       
       

        Hi,
       
        I have a sort of counter-example about IDSO,
        and the fact that information exchange is used in order to
predict
        self-organisation.
       
        In a system called sodaplay (www.sodaplay.com), structures made
of
        bodies connected by springs can produce a coordinated behaviour
(e.g.
        walk). In this case, there is no information exchange beween
bodies.
        They actually coordinate the movement through the physics.
       
        Even a snakebot can be equipped with joints that oscillate with
a
        given frequency, amplitude and phase, such that its behaviour
        corresponds to walk straight, with no "channels" for
communication.
       
        I went through this because I am trying to implement a minimal
control
        effort (MCE) for my modular robots. The joints (between modules)
        oscillate with a fixed phase and
        amplitude. Therefore, the  behaviour depends just on the
topology of
        the structure. There is no exchange of information, except for
the
        growth and division mechanism.
        The growth (i.e. aggregation of a module) and division (cut of a
        physical link) are determined by a local rule (CA-like) and
depends on
        the internal state.
       
        In this case, to have an efficient design, we aim to a minimal
control
        effort, which in turn minimise the exchange of information
(instead of
        maximising it), without loosing the abilities selected by the
        environment, in order to survive.
       
        Antonio
       
       
        On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
        >
        >
        > Dear all,
        >
        > Now that the list is operational, it would make sense to start
our
        > discussions (all the typical email-list rules apply) :)
        >
        > There are at least two main motivations for Information-driven
        > Self-organisation (IDSO).
        >
        > a) A weak IDSO: many evolutionary/selection/self-organisation
pressures can
        > be approximated information-theoretically - and such
approximations can
        > serve as shortcuts in explaining natural phenomena as well as
in designing
        > biologically-inspired systems (epistemological view on IDSO?)
        >
        > b) A strong IDSO: many optimal structures evolve/self-organise
in nature
        > when information transfer within certain channels is maximised
- this view
        > maintains that maximization of information transfer through
selected
        > channels is one of the main evolutionary pressures
(ontological view on
        > IDSO?)
        >
        > There may be other views, but in this message we would like to
elaborate on
        > the strong notion, following our recent work [Piraveenan et
al., 2007]:
        > Although the evolutionary process involves a larger number of
drives and
        > constraints, information fidelity (i.e. preservation) is a
consistent motif
        > throughout biology: e.g., modern evolution operates close to
the error
        > threshold [Adami, 1998], and biological sensorimotor equipment
typically
        > exhausts the available informatory capacity (under given
constraints) close
        > to the limit [Laughlin et al., 1998]. Adami, in fact, argues
that the
        > evolutionary process extracts valuable information and stores
it in the
        > genes. Since this process is relatively slow [Zurek, 1990], it
is a
        > selective advantage to preserve this information, once
captured.
        >
        > In other words, constraints (like noise in the environment)
reduce the
        > channel's bandwidth, and the system evolves to preserve
information by
        > self-improving, re-structuring, and so on.
        >
        > Comments?
        >
        > Thanks,
        > Daniel Polani and Mikhail Prokopenko
        >
        > Refs:
        >
        > Adami, C., 1998: Introduction to Artificial Life. Springer.
        >
        > Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C,
1998: The
        > metabolic cost of neural information. Nature Neuroscience
1(1), 36-41.
        >
        > Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of
Genetic
        > Coding: an Information-theoretic Model, in F. Almeida e Costa,
L. M. Rocha,
        > E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial
Life: 9th
        > European Conference on Artificial Life (ECAL-2007), Lisbon,
Portugal,
        > September 10-14, Lecture Notes in Artificial Intelligence,
vol. 4648, pp.
        > 42-52, Springer, Berlin.
        >
        > Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H.,
ed.:
        > Complexity, Entropy and the Physics of Information. Santa Fe
Studies in the
        > Sciences of Complexity, Reading, Mass., Addison-Wesley.
        > ==============================================
        >
        > Please use "Reply to All" if you intend your message to go to
the list.
        > Thanks.
        >
        > -------------------------------------------------------
        > Dr Mikhail Prokopenko
        > Senior Research Scientist, Adaptive Systems Team Leader
        > Autonomous Systems Lab, CSIRO ICT Centre, Australia
        > http://www.ict.csiro.au/asl/
        > phone [phone]
        > http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
        >
        >
       


IMPORTANT: This email remains the property of the Australian Defence Organisation and is subject to the jurisdiction of section 70 of the CRIMES ACT 1914.  If you have received this email in error, you are requested to contact the sender and delete the email.

[Fabio.Boschetti]

Some parts of this message have been removed. Learn more about Nabble's security policy.

i also like Antonio’s point. A question could be “what benefit do we get by seeing Antonio’s system in term of information, which we would not get by seeing it via ‘traditional physics’ only?”

 

 

 

-----Original Message-----
From: Prokopenko, Mikhail (ICT Centre, North Ryde)
Sent: Wednesday, 21 November 2007 9:45 PM
To: Lafusa, Antonio
Cc: IDSO-CSIRO
Subject: RE: starting IDSO discussions: weak and strong IDSO

 

Hi all,

 

I think Antonio raised an interesting point.

 

It seems to me that information is still exchanged in systems like these (e.g., sodaplay or snakebots) - however, it doesn't require explicit module-to-module communication channels with packets being sent. Rather, the physical coupling creates the channel (like Juergen and Oliver mentioned): if the states of the remote modules become synchronised then this means that information has been exchanged - via "stigmergy" of physical constraints.

 

In other words, a stigmergic information flow may minimise the control effort in individual modules - but this would need to be studied precisely.

 

Thanks!

Mikhail

 

-----Original Message-----
From: Lafusa, Antonio
Sent: Wed 21/11/2007 9:49 PM
To: Prokopenko, Mikhail (ICT Centre, North Ryde)
Cc: IDSO-CSIRO
Subject: Re: starting IDSO discussions: weak and strong IDSO

Hi,

I have a sort of counter-example about IDSO,
and the fact that information exchange is used in order to predict
self-organisation.

In a system called sodaplay (www.sodaplay.com), structures made of
bodies connected by springs can produce a coordinated behaviour (e.g.
walk). In this case, there is no information exchange beween bodies.
They actually coordinate the movement through the physics.

Even a snakebot can be equipped with joints that oscillate with a
given frequency, amplitude and phase, such that its behaviour
corresponds to walk straight, with no "channels" for communication.

I went through this because I am trying to implement a minimal control
effort (MCE) for my modular robots. The joints (between modules)
oscillate with a fixed phase and
amplitude. Therefore, the  behaviour depends just on the topology of
the structure. There is no exchange of information, except for the
growth and division mechanism.
The growth (i.e. aggregation of a module) and division (cut of a
physical link) are determined by a local rule (CA-like) and depends on
the internal state.

In this case, to have an efficient design, we aim to a minimal control
effort, which in turn minimise the exchange of information (instead of
maximising it), without loosing the abilities selected by the
environment, in order to survive.

Antonio


On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
>
>
> Dear all,
>
> Now that the list is operational, it would make sense to start our
> discussions (all the typical email-list rules apply) :)
>
> There are at least two main motivations for Information-driven
> Self-organisation (IDSO).
>
> a) A weak IDSO: many evolutionary/selection/self-organisation pressures can
> be approximated information-theoretically – and such approximations can
> serve as shortcuts in explaining natural phenomena as well as in designing
> biologically-inspired systems (epistemological view on IDSO?)
>
> b) A strong IDSO: many optimal structures evolve/self-organise in nature
> when information transfer within certain channels is maximised – this view
> maintains that maximization of information transfer through selected
> channels is one of the main evolutionary pressures (ontological view on
> IDSO?)
>
> There may be other views, but in this message we would like to elaborate on
> the strong notion, following our recent work [Piraveenan et al., 2007]:
> Although the evolutionary process involves a larger number of drives and
> constraints, information fidelity (i.e. preservation) is a consistent motif
> throughout biology: e.g., modern evolution operates close to the error
> threshold [Adami, 1998], and biological sensorimotor equipment typically
> exhausts the available informatory capacity (under given constraints) close
> to the limit [Laughlin et al., 1998]. Adami, in fact, argues that the
> evolutionary process extracts valuable information and stores it in the
> genes. Since this process is relatively slow [Zurek, 1990], it is a
> selective advantage to preserve this information, once captured.
>
> In other words, constraints (like noise in the environment) reduce the
> channel's bandwidth, and the system evolves to preserve information by
> self-improving, re-structuring, and so on.
>
> Comments?
>
> Thanks,
> Daniel Polani and Mikhail Prokopenko
>
> Refs:
>
> Adami, C., 1998: Introduction to Artificial Life. Springer.
>
> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998: The
> metabolic cost of neural information. Nature Neuroscience 1(1), 36–41.
>
> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of Genetic
> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M. Rocha,
> E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial Life: 9th
> European Conference on Artificial Life (ECAL-2007), Lisbon, Portugal,
> September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648, pp.
> 42-52, Springer, Berlin.
>
> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
> Complexity, Entropy and the Physics of Information. Santa Fe Studies in the
> Sciences of Complexity, Reading, Mass., Addison-Wesley.
> ==============================================
>
> Please use "Reply to All" if you intend your message to go to the list.
> Thanks.
>
> -------------------------------------------------------
> Dr Mikhail Prokopenko
> Senior Research Scientist, Adaptive Systems Team Leader
> Autonomous Systems Lab, CSIRO ICT Centre, Australia
> http://www.ict.csiro.au/asl/
> phone [phone]
> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>
>

[Russ Abbott]

Hi,

I can't resist jumping in here.

Mikhail's original statement was, I gather, in support, for example, of Adami's claim that "the evolutionary process extracts valuable information and stores it in the genes." That seems quite true. Genes do seem to do that sort of thing. That claim, though doesn't say that everything that happens in nature (or in man-made systems like sodaplay) involves the explicit encoding and passing of information.

Furthermore, it seems that even strong IDSO is an example of weak IDSO in the sense that genes do not store information about the world. Yes, one can say that certain genes evolved to correspond to certain regularities in nature. But the genes themselves generally don't encode those regularities. Genes for, say, light skin that evolved in higher latitudes don't say anything about the length of the day or the strength of the sun. They just generate proteins that result in light skin, which correspond to certain facts about nature where those genes evolved.

So I'd like to request that we attempt to clarify a bit what the issue is that's being discussed.

-- Russ

On Nov 21, 2007 7:01 PM, <Fabio.Boschetti@...> wrote:

i also like Antonio 's point. A question could be "what benefit do we get by seeing Antonio 's system in term of information, which we would not get by seeing it via 'traditional physics' only?"

 

 

 

-----Original Message-----
From: Prokopenko, Mikhail (ICT Centre, North Ryde)
Sent: Wednesday, 21 November 2007 9:45 PM
To: Lafusa, Antonio
Cc: IDSO-CSIRO
Subject: RE: starting IDSO discussions: weak and strong IDSO

 

Hi all,

 

I think Antonio raised an interesting point.

 

It seems to me that information is still exchanged in systems like these (e.g., sodaplay or snakebots) - however, it doesn't require explicit module-to-module communication channels with packets being sent. Rather, the physical coupling creates the channel (like Juergen and Oliver mentioned): if the states of the remote modules become synchronised then this means that information has been exchanged - via "stigmergy" of physical constraints.

 

In other words, a stigmergic information flow may minimise the control effort in individual modules - but this would need to be studied precisely.

 

Thanks!

Mikhail

 

-----Original Message-----
From: Lafusa, Antonio
Sent: Wed 21/11/2007 9:49 PM
To: Prokopenko, Mikhail (ICT Centre, North Ryde)
Cc: IDSO-CSIRO
Subject: Re: starting IDSO discussions: weak and strong IDSO

Hi,

I have a sort of counter-example about IDSO,
and the fact that information exchange is used in order to predict
self-organisation.

In a system called sodaplay (www.sodaplay.com), structures made of
bodies connected by springs can produce a coordinated behaviour (e.g.
walk). In this case, there is no information exchange beween bodies.
They actually coordinate the movement through the physics.

Even a snakebot can be equipped with joints that oscillate with a
given frequency, amplitude and phase, such that its behaviour
corresponds to walk straight, with no "channels" for communication.

I went through this because I am trying to implement a minimal control
effort (MCE) for my modular robots. The joints (between modules)
oscillate with a fixed phase and
amplitude. Therefore, the  behaviour depends just on the topology of
the structure. There is no exchange of information, except for the
growth and division mechanism.
The growth (i.e. aggregation of a module) and division (cut of a
physical link) are determined by a local rule (CA-like) and depends on
the internal state.

In this case, to have an efficient design, we aim to a minimal control
effort, which in turn minimise the exchange of information (instead of
maximising it), without loosing the abilities selected by the
environment, in order to survive.

Antonio


On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
>
>
> Dear all,
>
> Now that the list is operational, it would make sense to start our
> discussions (all the typical email-list rules apply) :)
>
> There are at least two main motivations for Information-driven
> Self-organisation (IDSO).
>
> a) A weak IDSO: many evolutionary/selection/self-organisation pressures can
> be approximated information-theoretically – and such approximations can
> serve as shortcuts in explaining natural phenomena as well as in designing
> biologically-inspired systems (epistemological view on IDSO?)
>
> b) A strong IDSO: many optimal structures evolve/self-organise in nature
> when information transfer within certain channels is maximised – this view
> maintains that maximization of information transfer through selected
> channels is one of the main evolutionary pressures (ontological view on
> IDSO?)
>
> There may be other views, but in this message we would like to elaborate on
> the strong notion, following our recent work [Piraveenan et al., 2007]:
> Although the evolutionary process involves a larger number of drives and
> constraints, information fidelity (i.e. preservation) is a consistent motif
> throughout biology: e.g., modern evolution operates close to the error
> threshold [Adami, 1998], and biological sensorimotor equipment typically
> exhausts the available informatory capacity (under given constraints) close
> to the limit [Laughlin et al., 1998]. Adami, in fact, argues that the
> evolutionary process extracts valuable information and stores it in the
> genes. Since this process is relatively slow [Zurek, 1990], it is a
> selective advantage to preserve this information, once captured.
>
> In other words, constraints (like noise in the environment) reduce the
> channel's bandwidth, and the system evolves to preserve information by
> self-improving, re-structuring, and so on.
>
> Comments?
>
> Thanks,
> Daniel Polani and Mikhail Prokopenko
>
> Refs:
>
> Adami, C., 1998: Introduction to Artificial Life. Springer.
>
> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998: The
> metabolic cost of neural information. Nature Neuroscience 1(1), 36–41.
>
> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of Genetic
> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M. Rocha,
> E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial Life: 9th
> European Conference on Artificial Life (ECAL-2007), Lisbon, Portugal,
> September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648, pp.
> 42-52, Springer, Berlin.
>
> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
> Complexity, Entropy and the Physics of Information. Santa Fe Studies in the
> Sciences of Complexity, Reading, Mass., Addison-Wesley.
> ==============================================
>
> Please use "Reply to All" if you intend your message to go to the list.
> Thanks.
>
> -------------------------------------------------------
> Dr Mikhail Prokopenko
> Senior Research Scientist, Adaptive Systems Team Leader
> Autonomous Systems Lab, CSIRO ICT Centre, Australia
> http://www.ict.csiro.au/asl/
> phone [phone]
> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>
>

--
-- Russ Abbott
_____________________________________________
Professor, Computer Science
California State University, Los Angeles
o Check out my blog at http://russabbott.blogspot.com/

[Stanley N. Salthe]

Responding to Carlos' posting --
 There is a general systems principle that I have been touting since 1993
(and which is supported by detailed models of ecosystems), which is that
there is for all dissipative structures at any scalea a canonical
developmental trajectory, immature -> mature -> senescent (see my book
Development & Evolution, MIT Press for the criteria).   The basic cause of
senescene seems to be informational overload.   Regarding

>If we wanted to speak about strong IDSO, we need to speak already about
>the "usefulness" of the information. I think it is possible, but we need
>to consider the observer who judges the usefulness or purpose of the
>information (and we'll enter the teleological debates...)

Informational overload occurs when an already definitive system continues
uploading information (no material system can avoid this).  This
information begins to cause conflicting constraints, resulting in delayed
and less flexible system responses to perturbations, as well as being used
to bolster system habitual tendenies, resulting in system rigidity.  In
models some of you might be interested in the 'bias/variance dilemma' in
neural networks .

STAN

> Hi all,
>I agree with Mikhail, Juergen, and Olivier: I think you can describe
>information transfer in systems where there is no explicit information
>transfer, i.e. the designer didn't put it there. An example is with
>self-organizing traffic
>lights:<http://homepages.vub.ac.be/~cgershen/sos/SOTL/SOTL.html>http://homepages
>.vub.ac.be/~cgershen/sos/SOTL/SOTL.html(just press the GO button and wait
>a bit...)There is no direct communication between traffic lights: they
>just sense how many cars are approaching. Still, they are able to
>synchronize. This is because the cars can be seen as information that is
>sent from one traffic light to the next.
>I am quite sympathetic with weak IDSO: information seems to be a good
>formalism (along many others) to describe, understand, and build systems.
>About strong IDSO, it makes sense, but I am not so sure about it. It seems
>problematic to me, because having more information does not always give
>advantages. It has to be "useful" information, for the current
>context/niche. An example with koalas: 30% of their cranium is empty. In
>their ecological niche, they do not need to process much information, as
>there were no predators, and they basically eat and sleep all day. What
>for to waste energy in brainpower when food is scarce? In this case,
>evolution seems to have favored less information processing, as this came
>with an energetic cost that didn't pay off...Other example can be seen
>with parasites that exploit the information of their host (even at the
>genetic level).
>If we wanted to speak about strong IDSO, we need to speak already about
>the "usefulness" of the information. I think it is possible, but we need
>to consider the observer who judges the usefulness or purpose of the
>information (and we'll enter the teleological debates...) Still, for
>engineering this is not a problem: the engineer decides the purpose of the
>system, so IDSO can be a useful principle to use. But then this might be
>seen more as weak IDSO...
>Best regards,        <http://homepages.vub.ac.be/%7Ecgershen/>
><http://homepages.vub.ac.be/%7ecgershen/>Carlos Gershenson
> New England Complex Systems Institute
> 24 Mt. Auburn St. Cambridge MA 02138, USA
><http://complexes.blogspot.com/>
>

[Stanley N. Salthe]

Reacting to Hussein's, I present my 'definition' of information.  It is the
'intersection' between
(1) a reduction in uncertainty (Shannon)
(2) any constraint on entropy production (that is, on anything that happens
in our disequilibrium world)
(3) a difference that makes a difference (to a system of interpretance --
this involves semiotics) (from Bateson)
This seems to cover all aspects, and it easy to see that all three would be
pesent where one of them occurs.

STAN

>I think Antonio raised an important issue, what do we mean with information.
>
>The purpose of this email is (1) to share with you some of my humble
>views; and (2) to provoke a discussion about this question. In my opinion,
>without clarifying what we mean with information, it is hard to understand
>what IDSO really mean. Is IDSO an acronym for "everything" driven
>self-organisation or it denotes a subset of "everything" called
>"information", in which case what do mean with "information".
>
>Avoiding the traditional debate about the distinction between data,
>information, knowledge and wisdom, I think about information when
>modelling a problem in three categories:
>
>*) messages being explicitly exchanged between system components or
>between the components and the environment. I define the environment to be
>anything external to the system including the physical environment. So for
>an agent X, other agents are part of the environment were X is situated.
>This also includes any information that X does not know about, which form
>the information environment of X. So we have an information environment, a
>social environment, a physical environment, etc.
>
>*) constraints among the system components (internal) and constraints set
>by the environment on the system (external). I see constraints as
>meta-information which are either (a) evolved over time such as traditions
>and ethics; (b) enforced on the system by the designer such as the way we
>connect parts to each other and how many degrees of freedom we allow when
>we design a joint; (b) enforced on the system by the environment such as
>physics (I do not think the laws of physics have evolved but we can debate
>about this). Constraints can be represented in many forms: by physical
>connection of parts, mathematically, using symbolic knowledge, etc.
>Constraints can be hard (the agent has no choice) or soft (the agent can
>break them with a cost).
>
>*) objectives which represent the internal evaluation mechanism of the
>components and the overall system. An objective is a function that
>evaluates the performance of the agent. The agent can use it to eliminate
>states, and evaluate its performance in the environment. If the evaluation
>of the objective occurs internal to the agent, it becomes like an agent
>utility function. If the evaluation of the objective occurs by the
>environment, it becomes like a reinforcement function. Wherever the
>evaluation takes place, it does not change the fact that the agent uses
>this information to correct its actions and update its internal states
>(i.e. knowledge which becomes constraints on future agents' behaviour).
>
>In this way, I can define an agent behaviour as a function of the
>messages, constraints, and objectives. This behaviour can take one or a
>mix of three forms: (1) actions that affect other agents (2) actions that
>affect the environment (3) actions that affect the agent itself. Each of
>these actions can change the messages, constraints and objectives of an
>agent. The way this change occurs is by a subset of the constraints and
>objectives responsible for handling the dynamics.
>
>This might be an engineering oriented view of the world, but it would be
>nice to hear people views on that issue and examples of when this taxnomy
>breaks.
>
>I also agree with Carlos that we need to define "usefulness" of
>information. But I would like to generalize "useful" information for an
>agent to also include the cognitive ability of the agent to sense and
>process this information; otherwise the agent will be faced with
>information overload. Therefore, maximising information should be subject
>to two constraints: "relevance" and agent's capacity to use the
>information.
>
>Cheers
>Hussein
>
>
>
>-----Original Message-----
>From: Stanley N. Salthe [mailto:ssalthe@...]
>Sent: Thursday, 22 November 2007 2:46 AM
>To: Antonio Lafusa
>Cc: idso@...
>Subject: Re: starting IDSO discussions: weak and strong IDSO
>
>Antonio --  Physics does not escape being informed.  Following Howard
>Pattee's thinking, we can distinguish between dynamics and information
>contstraining the dynamics.  Its as general as, say, Y = aXpowerb.  The
>relation between X and Y describe the dynamical results (implicitly the
>dynamics), but the values of a and b are information.
>
>STAN
>
>>Hi,
>>
>>I have a sort of counter-example about IDSO, and the fact that
>>information exchange is used in order to predict self-organisation.
>>
>>In a system called sodaplay (www.sodaplay.com), structures made of
>>bodies connected by springs can produce a coordinated behaviour (e.g.
>>walk). In this case, there is no information exchange beween bodies.
>>They actually coordinate the movement through the physics.
>>
>>Even a snakebot can be equipped with joints that oscillate with a given
>>frequency, amplitude and phase, such that its behaviour corresponds to
>>walk straight, with no "channels" for communication.
>>
>>I went through this because I am trying to implement a minimal control
>>effort (MCE) for my modular robots. The joints (between modules)
>>oscillate with a fixed phase and amplitude. Therefore, the  behaviour
>>depends just on the topology of the structure. There is no exchange of
>>information, except for the growth and division mechanism.
>>The growth (i.e. aggregation of a module) and division (cut of a
>>physical link) are determined by a local rule (CA-like) and depends on
>>the internal state.
>>
>>In this case, to have an efficient design, we aim to a minimal control
>>effort, which in turn minimise the exchange of information (instead of
>>maximising it), without loosing the abilities selected by the
>>environment, in order to survive.
>>
>>Antonio
>>
>>
>>On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
>>>
>>>
>>> Dear all,
>>>
>>> Now that the list is operational, it would make sense to start our
>>> discussions (all the typical email-list rules apply) :)
>>>
>>> There are at least two main motivations for Information-driven
>>> Self-organisation (IDSO).
>>>
>>> a) A weak IDSO: many evolutionary/selection/self-organisation
>>> pressures can be approximated information-theoretically ñ and such
>>> approximations can serve as shortcuts in explaining natural phenomena
>>> as well as in designing biologically-inspired systems
>>> (epistemological view on IDSO?)
>>>
>>> b) A strong IDSO: many optimal structures evolve/self-organise in
>>> nature when information transfer within certain channels is maximised
>>> ñ this view maintains that maximization of information transfer
>>> through selected channels is one of the main evolutionary pressures
>>> (ontological view on
>>> IDSO?)
>>>
>>> There may be other views, but in this message we would like to
>>> elaborate on the strong notion, following our recent work [Piraveenan
>>>et al., 2007]:
>>> Although the evolutionary process involves a larger number of drives
>>> and constraints, information fidelity (i.e. preservation) is a
>>> consistent motif throughout biology: e.g., modern evolution operates
>>> close to the error threshold [Adami, 1998], and biological
>>> sensorimotor equipment typically exhausts the available informatory
>>> capacity (under given constraints) close to the limit [Laughlin et
>>> al., 1998]. Adami, in fact, argues that the evolutionary process
>>> extracts valuable information and stores it in the genes. Since this
>>> process is relatively slow [Zurek, 1990], it is a selective advantage
>>>to preserve this information, once captured.
>>>
>>> In other words, constraints (like noise in the environment) reduce
>>> the channel's bandwidth, and the system evolves to preserve
>>> information by self-improving, re-structuring, and so on.
>>>
>>> Comments?
>>>
>>> Thanks,
>>> Daniel Polani and Mikhail Prokopenko
>>>
>>> Refs:
>>>
>>> Adami, C., 1998: Introduction to Artificial Life. Springer.
>>>
>>> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998:
>>> The metabolic cost of neural information. Nature Neuroscience 1(1), 36ñ41.
>>>
>>> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of
>>> Genetic
>>> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M.
>>> Rocha, E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial
>>> Life: 9th European Conference on Artificial Life (ECAL-2007), Lisbon,
>>> Portugal, September 10-14, Lecture Notes in Artificial Intelligence,
>>>vol. 4648, pp.
>>> 42-52, Springer, Berlin.
>>>
>>> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
>>> Complexity, Entropy and the Physics of Information. Santa Fe Studies
>>> in the Sciences of Complexity, Reading, Mass., Addison-Wesley.
>>> ==============================================
>>>
>>> Please use "Reply to All" if you intend your message to go to the list.
>>> Thanks.
>>>
>>> -------------------------------------------------------
>>> Dr Mikhail Prokopenko
>>> Senior Research Scientist, Adaptive Systems Team Leader Autonomous
>>> Systems Lab, CSIRO ICT Centre, Australia http://www.ict.csiro.au/asl/
>>> phone [phone]
>>> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>>>
>>>
>
>
>
>Prof. (Chair of IT) Dr. Hussein A. Abbass, SMIEEE, SMACS
>
>School of Information Technology and Electrical Engineering,
>
>UNSW@ADFA, Australian Defence Force Academy Campus,
>
>Northcott Drive, Campbell, Canberra, ACT 2600, Australia.
>
>Web: http://www.itee.adfa.edu.au/~abbass
>
>Lab Web: http://www.itee.adfa.edu.au/~alar
>
>Email: h.abbass@...
>
>Tel. (+61) (2) 62688158
>
>Mob. [phone]
>
>Fax (+61) (2) 62688581
>
>PS: if you send an email to me and it does not get through our system, you
>can re-send it to hussein.abbass@...
>
>This message is intended for the addressee named and may contain
>confidential information. If you are not the intended recipient, please
>delete it and notify the sender. Views expressed in this message are those
>of the individual sender and are not necessarily the views of the
>University of New South Wales Campus at the Australian Defence Force
>Academy in Canberra.
>
>CRICOS Provider Number 00100G

[Stanley N. Salthe]

Matthew said:
>I'm wary of strong IDSO / any teleological arguments. The distribution
>of complexity amongst life forms spreads out to the right (ie the
>maximum tends to increase) because there is a minimal complexity, and
>evolution can build complexity (including efficient information
>channels), though it can just as easily remove it. There's always
>increases /and/ decreases in complexity and in information capacity
>(consider moles' eyesight and hearing). Measures of any of this
>(biological complexity, biological information channel capacity) are not
>easy and depend highly on the context and how one defines the system.

S: Teleology can be generalized throughout nature, thus:

{teleomaty {teleonomy {teleology}}} (brackets as in set theory). (read
-ology is an example of -onomy is an example of -omaty)  This is a
subsumptive hierarchy ({a subsumes {b}})

that is:

{physical tendency, as in variational principles {function, as in biology
{purpose}}}

with an example:

{entropy production {male agressive action {warfare}}}

The justification is a combined materialist and evolutionary argument --
nothing comes from nothing; everything has a precursor in ancestral systems.

STAN

[Kostas.Alexandridis]

Dear a

While I don't necessarily disagree with all the previous contributions, I would like to add another dimension to the discussion that does touches the issues discussed from a different perspective.

a. Beyond any mechanistic and mathematic interpretation of information (including at a system level of interaction) lays the issue of induction: how we assess information flows that operates simultaneously in multiple spatial, cognitive and temporal scales? The problem is obvious for example in quantum computation, but is as equally present in many natural and social sciences (see the micro-to-macro problem in sociology and social psychology).
b. There is an ongoing debate, especially in the resilience community (see Brian Walkers' work) whether or not one can actually foresee a change in a systems' state threshold before actually crossing it. Many argue that you can never know if a change (or regime shift) happens until you crossed the change threshold. Others argue that you can heuristically assess such a probability.
c. We do not have methods that reliably assess the value of information under conditions of deep uncertainty and extensively incomplete information (this is, unfortunately the case in many natural and cognitive processes observed in real world settings). There are some methodological remedies, e.g., see the work of Lempert, Popper and Bankes (2003) that propose seeking for robustness instead of optimization strategies in a large ensemble of runs, but these remedies are highly computationally intensive, and it turns out that lead to different conclusions (i.e., a robust strategy identifies solutions in parameter spaces that do not necessarily overlap or are similar to the optimal strategies in terms of both solutions as well as parameter spaces).
d. If stochasticity and heuristics are necessary to assess value and quality of information, then traditional interpretive scientific methods are not adequate to describe the system behavior. We need methods that are "scale-free" and are able to bride fundamental differences pertaining the semantics of the subjectivist versus the frequentist interpretation of probabilistic (and probabilogic) relationships.

Regards,
Kostas.


-----Original Message-----
From: N. Salthe, Stanley
Sent: Friday, 23 November 2007 6:20 AM
To: IDSO-CSIRO
Subject: RE: starting IDSO discussions: weak and strong IDSO

Reacting to Hussein's, I present my 'definition' of information.  It is the
'intersection' between
(1) a reduction in uncertainty (Shannon)
(2) any constraint on entropy production (that is, on anything that happens
in our disequilibrium world)
(3) a difference that makes a difference (to a system of interpretance --
this involves semiotics) (from Bateson)
This seems to cover all aspects, and it easy to see that all three would be
pesent where one of them occurs.

STAN

>I think Antonio raised an important issue, what do we mean with information.
>
>The purpose of this email is (1) to share with you some of my humble
>views; and (2) to provoke a discussion about this question. In my opinion,
>without clarifying what we mean with information, it is hard to understand
>what IDSO really mean. Is IDSO an acronym for "everything" driven
>self-organisation or it denotes a subset of "everything" called
>"information", in which case what do mean with "information".
>
>Avoiding the traditional debate about the distinction between data,
>information, knowledge and wisdom, I think about information when
>modelling a problem in three categories:
>
>*) messages being explicitly exchanged between system components or
>between the components and the environment. I define the environment to be
>anything external to the system including the physical environment. So for
>an agent X, other agents are part of the environment were X is situated.
>This also includes any information that X does not know about, which form
>the information environment of X. So we have an information environment, a
>social environment, a physical environment, etc.
>
>*) constraints among the system components (internal) and constraints set
>by the environment on the system (external). I see constraints as
>meta-information which are either (a) evolved over time such as traditions
>and ethics; (b) enforced on the system by the designer such as the way we
>connect parts to each other and how many degrees of freedom we allow when
>we design a joint; (b) enforced on the system by the environment such as
>physics (I do not think the laws of physics have evolved but we can debate
>about this). Constraints can be represented in many forms: by physical
>connection of parts, mathematically, using symbolic knowledge, etc.
>Constraints can be hard (the agent has no choice) or soft (the agent can
>break them with a cost).
>
>*) objectives which represent the internal evaluation mechanism of the
>components and the overall system. An objective is a function that
>evaluates the performance of the agent. The agent can use it to eliminate
>states, and evaluate its performance in the environment. If the evaluation
>of the objective occurs internal to the agent, it becomes like an agent
>utility function. If the evaluation of the objective occurs by the
>environment, it becomes like a reinforcement function. Wherever the
>evaluation takes place, it does not change the fact that the agent uses
>this information to correct its actions and update its internal states
>(i.e. knowledge which becomes constraints on future agents' behaviour).
>
>In this way, I can define an agent behaviour as a function of the
>messages, constraints, and objectives. This behaviour can take one or a
>mix of three forms: (1) actions that affect other agents (2) actions that
>affect the environment (3) actions that affect the agent itself. Each of
>these actions can change the messages, constraints and objectives of an
>agent. The way this change occurs is by a subset of the constraints and
>objectives responsible for handling the dynamics.
>
>This might be an engineering oriented view of the world, but it would be
>nice to hear people views on that issue and examples of when this taxnomy
>breaks.
>
>I also agree with Carlos that we need to define "usefulness" of
>information. But I would like to generalize "useful" information for an
>agent to also include the cognitive ability of the agent to sense and
>process this information; otherwise the agent will be faced with
>information overload. Therefore, maximising information should be subject
>to two constraints: "relevance" and agent's capacity to use the
>information.
>
>Cheers
>Hussein
>
>
>
>-----Original Message-----
>From: Stanley N. Salthe [mailto:ssalthe@...]
>Sent: Thursday, 22 November 2007 2:46 AM
>To: Antonio Lafusa
>Cc: idso@...
>Subject: Re: starting IDSO discussions: weak and strong IDSO
>
>Antonio --  Physics does not escape being informed.  Following Howard
>Pattee's thinking, we can distinguish between dynamics and information
>contstraining the dynamics.  Its as general as, say, Y = aXpowerb.  The
>relation between X and Y describe the dynamical results (implicitly the
>dynamics), but the values of a and b are information.
>
>STAN
>
>>Hi,
>>
>>I have a sort of counter-example about IDSO, and the fact that
>>information exchange is used in order to predict self-organisation.
>>
>>In a system called sodaplay (www.sodaplay.com), structures made of
>>bodies connected by springs can produce a coordinated behaviour (e.g.
>>walk). In this case, there is no information exchange beween bodies.
>>They actually coordinate the movement through the physics.
>>
>>Even a snakebot can be equipped with joints that oscillate with a given
>>frequency, amplitude and phase, such that its behaviour corresponds to
>>walk straight, with no "channels" for communication.
>>
>>I went through this because I am trying to implement a minimal control
>>effort (MCE) for my modular robots. The joints (between modules)
>>oscillate with a fixed phase and amplitude. Therefore, the  behaviour
>>depends just on the topology of the structure. There is no exchange of
>>information, except for the growth and division mechanism.
>>The growth (i.e. aggregation of a module) and division (cut of a
>>physical link) are determined by a local rule (CA-like) and depends on
>>the internal state.
>>
>>In this case, to have an efficient design, we aim to a minimal control
>>effort, which in turn minimise the exchange of information (instead of
>>maximising it), without loosing the abilities selected by the
>>environment, in order to survive.
>>
>>Antonio
>>
>>
>>On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
>>>
>>>
>>> Dear all,
>>>
>>> Now that the list is operational, it would make sense to start our
>>> discussions (all the typical email-list rules apply) :)
>>>
>>> There are at least two main motivations for Information-driven
>>> Self-organisation (IDSO).
>>>
>>> a) A weak IDSO: many evolutionary/selection/self-organisation
>>> pressures can be approximated information-theoretically ñ and such
>>> approximations can serve as shortcuts in explaining natural phenomena
>>> as well as in designing biologically-inspired systems
>>> (epistemological view on IDSO?)
>>>
>>> b) A strong IDSO: many optimal structures evolve/self-organise in
>>> nature when information transfer within certain channels is maximised
>>> ñ this view maintains that maximization of information transfer
>>> through selected channels is one of the main evolutionary pressures
>>> (ontological view on
>>> IDSO?)
>>>
>>> There may be other views, but in this message we would like to
>>> elaborate on the strong notion, following our recent work [Piraveenan
>>>et al., 2007]:
>>> Although the evolutionary process involves a larger number of drives
>>> and constraints, information fidelity (i.e. preservation) is a
>>> consistent motif throughout biology: e.g., modern evolution operates
>>> close to the error threshold [Adami, 1998], and biological
>>> sensorimotor equipment typically exhausts the available informatory
>>> capacity (under given constraints) close to the limit [Laughlin et
>>> al., 1998]. Adami, in fact, argues that the evolutionary process
>>> extracts valuable information and stores it in the genes. Since this
>>> process is relatively slow [Zurek, 1990], it is a selective advantage
>>>to preserve this information, once captured.
>>>
>>> In other words, constraints (like noise in the environment) reduce
>>> the channel's bandwidth, and the system evolves to preserve
>>> information by self-improving, re-structuring, and so on.
>>>
>>> Comments?
>>>
>>> Thanks,
>>> Daniel Polani and Mikhail Prokopenko
>>>
>>> Refs:
>>>
>>> Adami, C., 1998: Introduction to Artificial Life. Springer.
>>>
>>> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998:
>>> The metabolic cost of neural information. Nature Neuroscience 1(1), 36ñ41.
>>>
>>> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of
>>> Genetic
>>> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M.
>>> Rocha, E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial
>>> Life: 9th European Conference on Artificial Life (ECAL-2007), Lisbon,
>>> Portugal, September 10-14, Lecture Notes in Artificial Intelligence,
>>>vol. 4648, pp.
>>> 42-52, Springer, Berlin.
>>>
>>> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
>>> Complexity, Entropy and the Physics of Information. Santa Fe Studies
>>> in the Sciences of Complexity, Reading, Mass., Addison-Wesley.
>>> ==============================================
>>>
>>> Please use "Reply to All" if you intend your message to go to the list.
>>> Thanks.
>>>
>>> -------------------------------------------------------
>>> Dr Mikhail Prokopenko
>>> Senior Research Scientist, Adaptive Systems Team Leader Autonomous
>>> Systems Lab, CSIRO ICT Centre, Australia http://www.ict.csiro.au/asl/
>>> phone [phone]
>>> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>>>
>>>
>
>
>
>Prof. (Chair of IT) Dr. Hussein A. Abbass, SMIEEE, SMACS
>
>School of Information Technology and Electrical Engineering,
>
>UNSW@ADFA, Australian Defence Force Academy Campus,
>
>Northcott Drive, Campbell, Canberra, ACT 2600, Australia.
>
>Web: http://www.itee.adfa.edu.au/~abbass
>
>Lab Web: http://www.itee.adfa.edu.au/~alar
>
>Email: h.abbass@...
>
>Tel. (+61) (2) 62688158
>
>Mob. [phone]
>
>Fax (+61) (2) 62688581
>
>PS: if you send an email to me and it does not get through our system, you
>can re-send it to hussein.abbass@...
>
>This message is intended for the addressee named and may contain
>confidential information. If you are not the intended recipient, please
>delete it and notify the sender. Views expressed in this message are those
>of the individual sender and are not necessarily the views of the
>University of New South Wales Campus at the Australian Defence Force
>Academy in Canberra.
>
>CRICOS Provider Number 00100G

[Russ Abbott]

Mikhail, Is the paper you referred to in your opening message (Piraveenan, M., Polani, D., Prokopenko, M., "Emergence of Genetic Coding: an Information-theoretic Model") available online?

-- Russ

On Nov 22, 2007 1:38 PM, <Kostas.Alexandridis@...> wrote:

Dear a

While I don't necessarily disagree with all the previous contributions, I would like to add another dimension to the discussion that does touches the issues discussed from a different perspective.

a. Beyond any mechanistic and mathematic interpretation of information (including at a system level of interaction) lays the issue of induction: how we assess information flows that operates simultaneously in multiple spatial, cognitive and temporal scales? The problem is obvious for example in quantum computation, but is as equally present in many natural and social sciences (see the micro-to-macro problem in sociology and social psychology).
b. There is an ongoing debate, especially in the resilience community (see Brian Walkers' work) whether or not one can actually foresee a change in a systems' state threshold before actually crossing it. Many argue that you can never know if a change (or regime shift) happens until you crossed the change threshold. Others argue that you can heuristically assess such a probability.
c. We do not have methods that reliably assess the value of information under conditions of deep uncertainty and extensively incomplete information (this is, unfortunately the case in many natural and cognitive processes observed in real world settings). There are some methodological remedies, e.g., see the work of Lempert, Popper and Bankes (2003) that propose seeking for robustness instead of optimization strategies in a large ensemble of runs, but these remedies are highly computationally intensive, and it turns out that lead to different conclusions ( i.e., a robust strategy identifies solutions in parameter spaces that do not necessarily overlap or are similar to the optimal strategies in terms of both solutions as well as parameter spaces).
d. If stochasticity and heuristics are necessary to assess value and quality of information, then traditional interpretive scientific methods are not adequate to describe the system behavior. We need methods that are "scale-free" and are able to bride fundamental differences pertaining the semantics of the subjectivist versus the frequentist interpretation of probabilistic (and probabilogic) relationships.

Regards,
Kostas.

-----Original Message-----
From: N. Salthe, Stanley
Sent: Friday, 23 November 2007 6:20 AM
To: IDSO-CSIRO
Subject: RE: starting IDSO discussions: weak and strong IDSO

Reacting to Hussein's, I present my 'definition' of information.  It is the
'intersection' between
(1) a reduction in uncertainty (Shannon)
(2) any constraint on entropy production (that is, on anything that happens
in our disequilibrium world)
(3) a difference that makes a difference (to a system of interpretance --
this involves semiotics) (from Bateson)
This seems to cover all aspects, and it easy to see that all three would be
pesent where one of them occurs.

STAN

>I think Antonio raised an important issue, what do we mean with information.
>
>The purpose of this email is (1) to share with you some of my humble
>views; and (2) to provoke a discussion about this question. In my opinion,

>without clarifying what we mean with information, it is hard to understand
>what IDSO really mean. Is IDSO an acronym for "everything" driven
>self-organisation or it denotes a subset of "everything" called
>"information", in which case what do mean with "information".
>
>Avoiding the traditional debate about the distinction between data,
>information, knowledge and wisdom, I think about information when
>modelling a problem in three categories:
>
>*) messages being explicitly exchanged between system components or
>between the components and the environment. I define the environment to be
>anything external to the system including the physical environment. So for
>an agent X, other agents are part of the environment were X is situated.
>This also includes any information that X does not know about, which form
>the information environment of X. So we have an information environment, a
>social environment, a physical environment, etc.
>
>*) constraints among the system components (internal) and constraints set
>by the environment on the system (external). I see constraints as
>meta-information which are either (a) evolved over time such as traditions
>and ethics; (b) enforced on the system by the designer such as the way we
>connect parts to each other and how many degrees of freedom we allow when
>we design a joint; (b) enforced on the system by the environment such as
>physics (I do not think the laws of physics have evolved but we can debate
>about this). Constraints can be represented in many forms: by physical
>connection of parts, mathematically, using symbolic knowledge, etc.
>Constraints can be hard (the agent has no choice) or soft (the agent can
>break them with a cost).
>
>*) objectives which represent the internal evaluation mechanism of the
>components and the overall system. An objective is a function that
>evaluates the performance of the agent. The agent can use it to eliminate
>states, and evaluate its performance in the environment. If the evaluation
>of the objective occurs internal to the agent, it becomes like an agent
>utility function. If the evaluation of the objective occurs by the
>environment, it becomes like a reinforcement function. Wherever the
>evaluation takes place, it does not change the fact that the agent uses
>this information to correct its actions and update its internal states
>(i.e. knowledge which becomes constraints on future agents' behaviour).
>
>In this way, I can define an agent behaviour as a function of the
>messages, constraints, and objectives. This behaviour can take one or a
>mix of three forms: (1) actions that affect other agents (2) actions that
>affect the environment (3) actions that affect the agent itself. Each of
>these actions can change the messages, constraints and objectives of an
>agent. The way this change occurs is by a subset of the constraints and
>objectives responsible for handling the dynamics.
>
>This might be an engineering oriented view of the world, but it would be
>nice to hear people views on that issue and examples of when this taxnomy
>breaks.
>
>I also agree with Carlos that we need to define "usefulness" of
>information. But I would like to generalize "useful" information for an
>agent to also include the cognitive ability of the agent to sense and
>process this information; otherwise the agent will be faced with
>information overload. Therefore, maximising information should be subject
>to two constraints: "relevance" and agent's capacity to use the
>information.
>
>Cheers
>Hussein
>
>
>
>-----Original Message-----
>From: Stanley N. Salthe [mailto: ssalthe@...]
>Sent: Thursday, 22 November 2007 2:46 AM
>To: Antonio Lafusa
>Cc: idso@...
>Subject: Re: starting IDSO discussions: weak and strong IDSO
>
>Antonio --  Physics does not escape being informed.  Following Howard
>Pattee's thinking, we can distinguish between dynamics and information
>contstraining the dynamics.  Its as general as, say, Y = aXpowerb.  The
>relation between X and Y describe the dynamical results (implicitly the
>dynamics), but the values of a and b are information.
>
>STAN
>
>>Hi,
>>
>>I have a sort of counter-example about IDSO, and the fact that
>>information exchange is used in order to predict self-organisation.
>>
>>In a system called sodaplay (www.sodaplay.com), structures made of
>>bodies connected by springs can produce a coordinated behaviour (e.g.
>>walk). In this case, there is no information exchange beween bodies.
>>They actually coordinate the movement through the physics.
>>
>>Even a snakebot can be equipped with joints that oscillate with a given
>>frequency, amplitude and phase, such that its behaviour corresponds to
>>walk straight, with no "channels" for communication.
>>
>>I went through this because I am trying to implement a minimal control
>>effort (MCE) for my modular robots. The joints (between modules)
>>oscillate with a fixed phase and amplitude. Therefore, the  behaviour
>>depends just on the topology of the structure. There is no exchange of
>>information, except for the growth and division mechanism.
>>The growth (i.e. aggregation of a module) and division (cut of a
>>physical link) are determined by a local rule (CA-like) and depends on
>>the internal state.
>>
>>In this case, to have an efficient design, we aim to a minimal control
>>effort, which in turn minimise the exchange of information (instead of
>>maximising it), without loosing the abilities selected by the
>>environment, in order to survive.
>>
>>Antonio
>>
>>
>>On Nov 20, 2007 6:14 AM,  <Mikhail.Prokopenko@...> wrote:
>>>
>>>
>>> Dear all,
>>>
>>> Now that the list is operational, it would make sense to start our
>>> discussions (all the typical email-list rules apply) :)
>>>
>>> There are at least two main motivations for Information-driven
>>> Self-organisation (IDSO).
>>>
>>> a) A weak IDSO: many evolutionary/selection/self-organisation
>>> pressures can be approximated information-theoretically ñ and such
>>> approximations can serve as shortcuts in explaining natural phenomena
>>> as well as in designing biologically-inspired systems
>>> (epistemological view on IDSO?)
>>>
>>> b) A strong IDSO: many optimal structures evolve/self-organise in
>>> nature when information transfer within certain channels is maximised
>>> ñ this view maintains that maximization of information transfer
>>> through selected channels is one of the main evolutionary pressures
>>> (ontological view on
>>> IDSO?)
>>>
>>> There may be other views, but in this message we would like to
>>> elaborate on the strong notion, following our recent work [Piraveenan
>>>et al., 2007]:
>>> Although the evolutionary process involves a larger number of drives
>>> and constraints, information fidelity (i.e. preservation) is a
>>> consistent motif throughout biology: e.g., modern evolution operates
>>> close to the error threshold [Adami, 1998], and biological
>>> sensorimotor equipment typically exhausts the available informatory
>>> capacity (under given constraints) close to the limit [Laughlin et
>>> al., 1998]. Adami, in fact, argues that the evolutionary process
>>> extracts valuable information and stores it in the genes. Since this
>>> process is relatively slow [Zurek, 1990], it is a selective advantage
>>>to preserve this information, once captured.
>>>
>>> In other words, constraints (like noise in the environment) reduce
>>> the channel's bandwidth, and the system evolves to preserve
>>> information by self-improving, re-structuring, and so on.
>>>
>>> Comments?
>>>
>>> Thanks,
>>> Daniel Polani and Mikhail Prokopenko
>>>
>>> Refs:
>>>
>>> Adami, C., 1998: Introduction to Artificial Life. Springer.
>>>
>>> Laughlin, S.B., de Ruyter van Steveninck, R.R., Anderson, J.C, 1998:
>>> The metabolic cost of neural information. Nature Neuroscience 1(1), 36ñ41.
>>>
>>> Piraveenan, M., Polani, D., Prokopenko, M., 2007: Emergence of
>>> Genetic
>>> Coding: an Information-theoretic Model, in F. Almeida e Costa, L. M.
>>> Rocha, E. Costa, I. Harvey, A. Coutinho (eds). Advances in Artificial
>>> Life: 9th European Conference on Artificial Life (ECAL-2007), Lisbon,
>>> Portugal, September 10-14, Lecture Notes in Artificial Intelligence,
>>>vol. 4648, pp.
>>> 42-52, Springer, Berlin.
>>>
>>> Zurek, W.H., ed., 1990: Valuable Information. In Zurek, W.H., ed.:
>>> Complexity, Entropy and the Physics of Information. Santa Fe Studies
>>> in the Sciences of Complexity, Reading, Mass., Addison-Wesley.
>>> ==============================================
>>>
>>> Please use "Reply to All" if you intend your message to go to the list.
>>> Thanks.
>>>
>>> -------------------------------------------------------
>>> Dr Mikhail Prokopenko
>>> Senior Research Scientist, Adaptive Systems Team Leader Autonomous
>>> Systems Lab, CSIRO ICT Centre, Australia http://www.ict.csiro.au/asl/
>>> phone [phone]
>>> http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>>>
>>>
>
>
>
>Prof. (Chair of IT) Dr. Hussein A. Abbass, SMIEEE, SMACS
>
>School of Information Technology and Electrical Engineering,
>
>UNSW@ADFA, Australian Defence Force Academy Campus,
>
>Northcott Drive, Campbell, Canberra, ACT 2600, Australia.
>
>Web: http://www.itee.adfa.edu.au/~abbass
>
>Lab Web: http://www.itee.adfa.edu.au/~alar
>
>Email: h.abbass@...
>
>Tel. (+61) (2) 62688158
>
>Mob. [phone]
>
>Fax (+61) (2) 62688581
>
>PS: if you send an email to me and it does not get through our system, you
>can re-send it to hussein.abbass@...
>
>This message is intended for the addressee named and may contain
>confidential information. If you are not the intended recipient, please
>delete it and notify the sender. Views expressed in this message are those
>of the individual sender and are not necessarily the views of the
>University of New South Wales Campus at the Australian Defence Force
>Academy in Canberra.
>
>CRICOS Provider Number 00100G

--
-- Russ Abbott
_____________________________________________
Professor, Computer Science
California State University, Los Angeles
o Check out my blog at http://russabbott.blogspot.com/

[Stanley N. Salthe]

Folks -- It sometimes takes a while for things to 'sink in'.  I hate to
impugn the very title of our undertaking, but it might be worth pointing
out that there can be doubts about information 'driving' anything.  Using
the basic dichotomy of information / dynamics, information clearly
functions as constraints on dynamics.  It could be said to mediate initial
conditions into local results.  In this sense it is the 'formal cause' of
Aristotle's classification.  What does 'drive' dynamics then? Efficient
cause is the trigger or forcing, material cause is the local readiness,
while final cause is the 'purpose' (see my recent posting on finalities),
most generally just the tendency of any variational principle.  Final
causes 'pull' rather than 'drive', so that excludes them from driving.  I
think driving is best associated with efficient cause, which determines
when, e.g., self-organization will get going.  Information is carried by
the local 'set-up', and embodies the results of historically generated
situations / occasions.  Information informs, contextualizes, and then
mediates.

STAN

>    Dear all,   Now that the list is operational, it would  make sense to
>start our discussions (all the typical email-list rules apply)  :)   There
>are at least two main motivations for  Information-driven
>Self-organisation (IDSO).    a) A weak IDSO: many
>evolutionary/selection/self-organisation pressures can be approximated
>information-theoretically   and such approximations can serve as
>shortcuts in  explaining natural phenomena as well as in designing
>biologically-inspired  systems (epistemological view on IDSO?)   b) A
>strong IDSO: many optimal structures  evolve/self-organise in nature when
>information transfer within certain channels  is maximised   this view
>maintains that maximization of information transfer  through selected
>channels is one of the main evolutionary pressures (ontological  view on
>IDSO?)   There may be other views, but in this  message we would like to
>elaborate on the strong notion, following our recent  work [Piraveenan et
>al., 2007]: Although the evolutionary process involves a  larger number of
>drives and constraints, information fidelity (i.e.  preservation) is a
>consistent motif throughout biology: e.g., modern evolution  operates
>close to the error threshold [Adami, 1998], and biological sensorimotor
>equipment typically exhausts the available informatory capacity (under
>given  constraints) close to the limit [Laughlin et al., 1998]. Adami, in
>fact, argues  that the evolutionary process extracts valuable information
>and stores it in the  genes. Since this process is relatively slow [Zurek,
>1990], it is a selective  advantage to preserve this information, once
>captured.   In other words, constraints (like noise in  the environment)
>reduce the channel's bandwidth, and the system evolves to  preserve
>information by self-improving, re-structuring, and so on.    Comments?  
>Thanks, Daniel Polani and Mikhail Prokopenko   Refs:   Adami, C., 1998:
>Introduction to Artificial  Life. Springer.   Laughlin, S.B., de Ruyter
>van Steveninck,  R.R., Anderson, J.C, 1998: The metabolic cost of neural
>information. Nature  Neuroscience 1(1), 36 41.   Piraveenan, M., Polani,
>D., Prokopenko, M.,  2007: Emergence of Genetic Coding: an
>Information-theoretic Model, in F. Almeida  e Costa, L. M. Rocha, E.
>Costa, I. Harvey, A. Coutinho (eds). Advances in  Artificial Life: 9th
>European Conference on Artificial Life (ECAL-2007), Lisbon,  Portugal,
>September 10-14, Lecture Notes in Artificial Intelligence, vol. 4648,  pp.
>42-52, Springer, Berlin.   Zurek, W.H., ed., 1990: Valuable  Information.
>In Zurek, W.H., ed.: Complexity, Entropy and the Physics of  Information.
>Santa Fe Studies in the Sciences of Complexity, Reading, Mass.,
>Addison-Wesley.
>==============================================   Please  use "Reply to
>All" if you intend your message to go to the list.  Thanks.  
>-------------------------------------------------------
>Dr Mikhail  Prokopenko
>Senior Research Scientist, Adaptive Systems Team  Leader
>Autonomous Systems Lab, CSIRO ICT Centre, Australia
>http://www.ict.csiro.au/asl/
>phone  [phone]
>http://www.ict.csiro.au/staff/Mikhail.Prokopenko/
>  

[Antonio Lafusa]

Dear all,

Mikhail wrote:
> A strong IDSO: many optimal structures evolve/self-organise in nature
> when information transfer within certain channels is maximised – this view
> maintains that maximization of information transfer through selected
> channels is one of the main evolutionary pressures (ontological view on
> IDSO?)
>

in my opinon, in the definition of strong IDSO, the words "certain
channels" and "selected channels" make the meaning unclear.
Who decide which channels are selected/considered when calculating
information transfer (e.g. using mutual information)?
Do we mean that is evolution that select those channels, such that
other potential channels are not considered? Or that we know a priori
which channels are "useful"?

It would be convenient to have a general criterion to identify/achieve
self-organisation, referring to all the way in which information can
be transferred between a self-organising agent X and its environment.
Also assuming that selection pressure determines the amount and nature
of the information that flows ("useful" to survive).

To avoid confusion, we could reformulate strong IDSO, saying that the
information flow between an agent and the environment - including all
channels - is maximised when optimal structure evolve/self-organise.

In nature, the words "optimal structures", can be referred to
proteins, cells, organisms, species, etc. that survive in the
environment.

However, an important question is: Why self-organisation requires
information to be transferred from the environment to the agent? It
seems to me that this assumption is unnecessary if the environment
does not change (an example have been cited before, about Koalas
loosing information about the environment). If the environment does
change, the species learn something new about the environment all the
times that a mutation in the genotype is "succesful" (the gene
mutation corresponds to a proliferation of the organisms).

What the organism learn (i.e. the reduction of uncertainty about the
environment) is that a particular gene is useful in order to survive
in the current environment and interact with other genes. So a gene
contains some information about the combination environment-genotype.
Since the gene is part of the genotype, and the organism is part of
the environment, it is clear that all the single parts are likely to
contain information about the whole system.
In other words, it might be convenient to see the "channels" between
different scales (What a single gene "knows" about the
environment+genotype? what a genotype know about the environment?
etc...)

An implication of this point of view (strong IDSO) is that:

IF there is a transfer of information between the environment and the
agent AND the agent is a sub-system made of simpler elementary
components, THEN there must be a transfer of information within the
agent's parts (because a single elementary component cannot support
all the agent-environment information flow, and information is
distributed among the parts).

As an experiment, we could maximise just the mutual information
between the agent parts, to see if there is an increase in the mutual
information between the organism and environment.

This could be generalised considering other hierarchies:
gene-genotype, genotype-species, species-ecosystem, etc. (e.g. in an
ever changing environment species exchange more information during
co-evolution - they know more about eachother. IDSO implies that this
is true, so the above example could be studied to see if IDSO is a
valid approach).

In an ever changing environment, a sort of "real-time computation" is
required in order to survive (e.g. sensory-motor conrol, DNA
recombination). Thus, IDSO could be a valid criterion when a
continuous re-organisation is needed to respond to an ever changing
environment.

To summarise, I think that IDSO should be reformulated in order to
include two main aspects:

1) the channels used to transfer information are created/selected by
evolution (artificial or natural)

2) the absolute validity of IDSO is limited to open systems (e.g.
organisms, robots, societies) living in an fast, ever changing
environment (since the need for information to be captured and used by
the agent depends on how fast the environment change) when the cost of
storing/processing information is ZERO. When the cost of
storing/processing information is not zero, then evolution might find
other driving forces that contrast the increase of information
transfer.
If the information transfer is maximised, and the memory (i.e.
capacity to store raw information) is limited or costly, then the
capacity of an agent to use the experience from the past is limited.
If by chance, a certain environmental conditions re-appears after a
long time, then the organism cannot react using the information
accumulated in the past.

In my opinion, even in the genotype there is a long-term memory and
short-term memory, whose proportion are established by evolution. IDSO
account for the short-term memory, where "fresh" data coming from the
environment is overwritten as fast as the environment changes.

sorry to put so many things together...

- Antonio