Connectionists: practopoiesis

Tue Jun 10 11:01:53 EDT 2014

Dear John,

Apologies for a delayed reply. I had a busy few days.

I will gladly try answering your question on why your system using a 
Markov decision process is a T2-system:

Simply put, your system interacts with the environment at two different 
levels:

1) At one level, the system deals with the current states and the 
decision of the next actions. At this level, the surrounding world 
affects the system in the form of states that the system takes.

In other words, a Markov process with fixed probability distributions 
interacts with the world only at that one level and thus, has one 
traverse and this is what makes it a T1.

2) However, there is another important component of the system--the 
learning mechanism for the probability distributions. This part also 
interacts with the environment, but at a different level of generality. 
This part estimates the probability distributions from the past and 
therefore, makes the entire system to interact with the environment also 
at a more general level.

As a result, the system interacts, in total, at two levels with its 
environment. And these two levels are organized into a "poietic" 
hierarchy. The processes at one level (learning; probability estimates) 
determine the properties of the other (states to actions), but not the 
other way around (e.g., Markov decision process does not act on its own 
learning rules, or on its own probability distributions).

These two levels of interaction make the system a T2. I think that this 
is quite simple and does not require a formal mathematical proof. I 
suggest reading the manuscript on practopoiesis to get more insight.

In general, all the intelligent algorithms that I know of, and all of 
the brain theories that I know of, are based on T2-systems. There is 
always one level that acts in response to sensory inputs, and there is 
another level that is responsible for learning of how to properly act 
(e.g., neural network + the plasticity mechanism).

In contrast to those theories, I argue that a biological organism with a 
nervous system is in fact a T3-system. It has one more (third) level of 
interaction with the environment, and all the three levels are organized 
into a poietic hierarchy. This brings a lot of additional flexibility, 
which a T2-system cannot achieve. T3-systems are more adaptive than T-2 
systems, much like T2-systems are more adaptive than T1-systems. See 
here for some accounts: 
http://www.danko-nikolic.com/distinction-between-t2-and-t3-systems/
Also, I argued that abductive reasoning can be achieved only by a T3-system.

To illustrate what a third adaptive level could bring to a Markov 
process, imagine a system that can adaptively change its probability 
matrices, from situation to situation, as needed--not by representing 
the recent statistics of states, but by having a large depository of 
probability functions (e.g., thousands) that have been already acquired 
by some third mechanism and have been extracted from a much broader 
sample of states (lifetime long). Then, by looking at the history of the 
recent few states, the system decides which of the probability functions 
it will use now (which may be quite different than the recent statistics).

This system would be a T3 because it would interact with the environment 
at the following three levels:

1) current state -> next action
2) recent history of states -> choice of probability functions from the 
depository
3) long-term statistics -> creation of the probability function depository

Notice the strict poietic hierarchy among the three.

I hope that this is clear enough and also enough of a teaser to learn 
more about practopoietic theory.

With best regards,

Danko

On 31/05/14 15:41, Juyang Weng wrote:
>
> On 5/30/14 5:01 PM, Danko Nikolic wrote:
>> Dear John,
>>
>>   I just read your SASE paper to make a comparison to practopoiesis, 
>> as you asked. Interesting paper. Nice work.
>>
>>   I like your succinct stile, formal and accurate. Your theorems make 
>> it clear what you mean by awareness, etc.
>>
>>   Here is what I can conclude about "A theory of developmental 
>> architecture" by Juyang Weng:
>>
>> Markov decision process is a T2-system. It does not make a difference 
>> whether you have sensors that collect only outside information or you 
>> also have sensors collecting internal information. It remains a T2 
>> system. The same holds for actions. Internal actions do not change a 
>> T2 into T3 either.
> Then, you must have rigorous and precise definition so that you are 
> not the only God to say T2 or T3.
> If your work is scientific, your detailed, rigorous and precise 
> definition but be able to be verified to be true or false by other 
> researchers.
> Otherwise, your work is religion like.
>
> For example, in the above your writing, you do not say "why" SASE is 
> T2 this is necessary for a scientific work.
>
>>
>>   Therefore, according to practopoietic theory, your theory falls 
>> into the category of T2-systems.
> You do not seem to have a verifiable theory yet.   It appears to be 
> religion like to me.
>>
>>   As you know, I laid out arguments that this is not enough to 
>> produce intelligent adaptive behavior that matches biological 
>> behaving systems. My suggestion would be to find ways to expand it 
>> into a T3.
> Your arguments are too vague to be qualified as scientific work. In 
> religion, the God says "if you believe it, it is there. Otherwise, it 
> is not."
>
> -John
>> With best regards,
>>
>>
>> Danko
>>
>>
>>
>>
>> On May 28, 2014, at 2:30 PM, Juyang Weng wrote:
>>
>>> Danko, thank you for the links.  You might want to take a look at my 
>>> Self-Aware and Self-Effecting (SASE)
>>> architecture.  Your three-traverse idea has some similarity to it, 
>>> but your theory is not (yet?) supported by fully computational detail
>>> as SASE.
>>>
>>> -John
>>>
>>> On 5/28/14 6:58 PM, Danko Nikolic wrote:
>>>> Dear all,
>>>>
>>>> I made an effort to make practopoiesis more approachable to a wider 
>>>> audience:
>>>>
>>>>
>>>> One: I wrote a short popular article on the implications of 
>>>> practopoiesis for artificial intelligence:
>>>>
>>>> http://www.singularityweblog.com/practopoiesis/
>>>>
>>>>
>>>>
>>>> Two: I made a list of the key concepts with a brief explanation of 
>>>> each:
>>>>
>>>> http://www.danko-nikolic.com/practopoiesis#Concepts
>>>>
>>>>
>>>>
>>>> I hope that this will be helpful.
>>>>
>>>> With warm greetings from Germany,
>>>>
>>>> Danko Nikolic
>>>>
>>>>
>>>
>>> -- 
>>> --
>>> Juyang (John) Weng, Professor
>>> Department of Computer Science and Engineering
>>> MSU Cognitive Science Program and MSU Neuroscience Program
>>> 428 S Shaw Ln Rm 3115
>>> Michigan State University
>>> East Lansing, MI 48824 USA
>>> Tel: 517-353-4388
>>> Fax: 517-432-1061
>>> Email: weng at cse.msu.edu <mailto:weng at cse.msu.edu>
>>> URL: http://www.cse.msu.edu/~weng/ <http://www.cse.msu.edu/%7Eweng/>
>>> ----------------------------------------------
>>>
>>
>> -- 
>>
>> Prof. Dr. Danko Nikolić
>>
>>
>> Web:http://www.danko-nikolic.com <http://www.danko-nikolic.com/>
>>
>> Mail address 1:
>> Group Leader
>> Department of Neurophysiology
>> Max Planck Institute for Brain Research
>> Deutschordenstr. 46
>> 60528 Frankfurt am Main
>> GERMANY
>>
>> Mail address 2:
>> Research Fellow
>> Frankfurt Institute for Advanced Studies
>> Wolfgang Goethe University
>> Ruth-Moufang-Str. 1
>> 60433 Frankfurt am Main
>> GERMANY
>>
>> ----------------------------
>> Office: (..49-69) 96769-736
>> Lab: (..49-69) 96769-209
>> Fax: (..49-69) 96769-327
>> danko.nikolic at gmail.com <mailto:danko.nikolic at gmail.com>
>> ----------------------------
>>
>>
>>
>>
>>
>
> -- 
> --
> Juyang (John) Weng, Professor
> Department of Computer Science and Engineering
> MSU Cognitive Science Program and MSU Neuroscience Program
> 428 S Shaw Ln Rm 3115
> Michigan State University
> East Lansing, MI 48824 USA
> Tel: 517-353-4388
> Fax: 517-432-1061
> Email:weng at cse.msu.edu
> URL:http://www.cse.msu.edu/~weng/
> ----------------------------------------------
>

-- 
Prof. Dr. Danko Nikolić

Web:
http://www.danko-nikolic.com

Mail address 1:
Department of Neurophysiology
Max Planck Institut for Brain Research
Deutschordenstr. 46
60528 Frankfurt am Main
GERMANY

Mail address 2:
Frankfurt Institute for Advanced Studies
Wolfgang Goethe University
Ruth-Moufang-Str. 1
60433 Frankfurt am Main
GERMANY

----------------------------
Office: (..49-69) 96769-736
Lab: (..49-69) 96769-209
Fax: (..49-69) 96769-327
danko.nikolic at gmail.com
----------------------------

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/connectionists/attachments/20140610/9fda38fd/attachment.html>