[Fis] AI Discussion--Part 2 (by Eric Werner)

[Pedro C. Marijuán]

Dear FIS Colleagues,

And herewith the second part, corresponding to a new FIS colleague, 
*Eric Werner*, one of the pioneers in computer science, artificial 
intelligence, and distributed cognition.
The original file of the contribution is also attached (a pdf in a high 
quality format). The regular text below is needed for our archives, as 
the attachments are often scrubbed by the server.
Discussants remember please the limit of three weekly messages, except 
the two presenters (unlimited responses).

So, we have two pioneers of artificial intelligence, Yixin and Eric, 
pointing in very different but fundamental and complementary 
directions... let us enjoy the discussion!!


All the best--Pedro

-------------------------------------------------------------------------------------------------

*Social Information in**
**Artificial Social Intelligence*
Social information can make Large Language Models
and Large Behavioral Models cooperative

*
*

*The Minimal Social Brain*

What would a minimal model of an agent's neural network-mind-brain such as
human be like? What are its minimal characteristics? What is the global 
architecture?
How would we map the agent's neural network to our abstract model? What 
is the
minimal complexity of the architecture that captures the essence of what 
it is to be an
intelligent agent in the world?
Can a purely linguistic interaction between a human and a dialogue-prompter
(computational device) capture the mental capacities, the network, the 
network state
and information states of the human agent's brain?

*Short history of communication theory*

There is an inherent entropy (ambiguity) in the semantics and pragmatics 
of any
linguistic sentence. Hence, a purely linguistic interaction would 
inevitably generate an
entropic model of the agent's brain-network. However, the model may be 
functionally
equivalent to the agent's brain-network in many situations of use. 
Wittgenstein's
notion of meaning as use (see 'Blue and Brown Books', 'Philosophical 
Investigations')
(Wittgenstein 1958)is relevant here since it suggests functional 
equivalence is
sufficient for human communication.
John von Neumann mathematical theory of games suggests that there is a
minimal architecture for agents in games and, more generally, in the 
economic world
(von Neumann J. 1947). In (Werner 2023 forthcoming) I investigate the 
logic of
information in games, specifically the logic of information states in 
agents and how
they evolve in time. Later I extended that work to investigate the logic 
of intentions
and social action in agents with limited information about their world. 
This extended
von Neumann's concept of information-set to include agent strategic 
states (that give
a model of the agent's intentional states). This permitted the 
formalization and
understanding of multi-agent cooperation by way of interacting 
intentional states of
agents(Werner 1988).
Wittgenstein realized the inadequacy of a purely information based theory of
language and meaning. In his Philosophical Investigations(Wittgenstein 
1958) and the
Blue and Brown Books(Wittgenstein 1958) he viewed meaning a given by 
use, such as
in a language game. Searle followed with his work on speech acts. Later 
Habermas
(Habermas 2015, Habermas 2015) used these concepts for his work on social
communication. However, all of these theories of language and social 
action were
based on informal vague notions of meaning (semantics and pragmatics). 
Chomsky
theories of syntax were just that with no semantics, no pragmatics, no 
theory of
communication.
In (Werner 1999) (Werner 1991) (Werner 1996) (Werner 1988) I generalized
information-based communication theory going beyond (Shannon 1949), (von
Neumann J. 1947), logical foundationalism (e.g., (Whitehead and Russell 
1950) early
Wittgenstein’s Tractatus (Wittgenstein 1969) to include 
intention-strategic-based
communication and cooperation theory. On my view communication involves
linguistic intentional states formalized as linguistic strategic states 
linked to the world
via semantics and pragmatic meaning.
Moreover, I showed the intimate relationship between states of 
information and
what an agent can do and what the agent can intend in a complex social 
world of
other agents(Werner 1991). This work unified the logics of information 
(world-stateinformation),
intention (strategic information) and ability (the logic of can and social
ability). Add to that von Neumann's formalization of utility and you get the
foundational bedrock of the minimal architecture of mental states of 
communicating
social agents whether they be human, animal or robotic.
How does this relate to artificial intelligence (AI) and ChatGPT, 
general AI (GAI)
and autonomous robots (self driving cars, social robots, ethical robots)?

*The social brain*

How is the architecture of social agents and social information relate 
to the
architecture of the human brain, the connectome, cerebral cortex, visual 
cortex,
Wernicke area, Broca area, and the cerebellum?
Where is social information and intentional capacity located in the 
human brain?

*The architecture of a social brain*

I would like to discuss the overall architecture of the social brain . The
fundamental question is: What makes a society possible at all? What is 
the minimal
architecture of the brain to enable an agent to be social?
Communication by language is obviously central, but it requires more. It 
requires
a particular mental capability, and the parts of the agents that have 
social capacities.
The capacity of the social brain is the ability to represent intentions, 
not only the
intentions of self, but the intentions of others. Through the 
interaction of the
representation of intentions of others, and the representation of 
intentions of the self,
it makes coordination and cooperation possible.
Intentions can be about actions, but they can also be about linguistic 
actions.
Coordinated speech is one of the fundamental processes for achieving 
coordinated
intentions for multi-agent intentional states. Intentional states are 
fundamentally
different than knowledge about the world which we call information states.
Intentional states require informational states because the agent needs 
to know about
the world in order to be able to act. The agents abilities are direct 
result of the
information the agent has about the world.
The less information that the agent has about the world, the fewer the 
things
that the agent can do. Thus the logic of ability is directly related to 
the logic of
information. The logic of ability underpins the logic of intention.

*Information, Intention and Utility*

There is one more component and that is value or utility. Utility drives the
formation of intentions of the agent. Therefore, to get a minimal social 
agent we need
at least three components: Intention S, information I, and utility or 
value (Werner
1988)r.
I called is three components, the Representational Capacity of the agent
R = ( S, I, V ).
A fourth element may be called planning and logical inference i.e., 
Planing-Logic
PL.
It may be that each one of these components ( S, I, V, PL ) will require 
different
types of training for a ChatGPT-like Large Language Model (LLM) or Large 
Behavior
Model (LBM).
They correspond to different areas of the human brain (e.g., the visual 
cortex,
Wernicke area (language and visual understanding), Broca area (speech 
generation),
and more, of course. The connectome located in the White Matter of the 
brain links
the different areas. Its role is fundamental for brain coordinated 
processing
(Geschwind 1974). For a nice overview of the connectome research see 
(Catani,
Sandrone et al. 2015)

*Artificial Social Intelligence*

Social information is different than information about the state of the 
world.
Intentional strategic states are implemented through social information
representations. The whole system of human, linguistic communication is 
an example
of functional social information exchange. An intentional state is the 
prime example
of social information.

*The social genome and the social brain*

We can view the genome as a social structure, which contains both the two
parental genomes. There's an interaction between the parental genomes, 
by means of
a protocol, a meta-protocol that determines which genome is in control 
at which time
in the development of the embryo. This enables the development of the 
brain to also
reflect both of the parents and their social capacities. It explains the 
fact that the
thinking of different children, maybe reflect that a one parent or that 
of another at
different stages of development. The brain is sectioned into different 
areas that
reflect the control of one parents genome or that of the other just as 
the body is
sectioned into different areas as it is being controlled by one parental 
genome or the
other parental genome.

*Social Information*

Thus, the very development of the brain is controlled by social information
processes see my “Brain meta genomics: Genome mind mapping:
Network protocols partition the developing brain” (Werner 2023).
The social information is embedded at all levels of the human body. It is
fundamental to the interaction between parental genomes. It is embedded 
at the level
of cell interactions, through cell-cell communication. It is embedded in 
the brain in
different areas of the brain that specialize in various processes. Many 
areas of the
brain process social information such as intentions and values, 
utilities or emotions.
Entire communication system of the brain from output to input from speech
generation to speech understanding is fundamentally a social information 
processing
system. The human brain is fundamentally social in nature.
If we are going to create cooperative LLMs or LBMs then their 
architecture has
to reflect the architecture of the human social brain. These artificial 
intelligent social
systems, will have to process social information much like human brains 
process
social information.

*
*

*Bibliography*

Catani, M., S. Sandrone and A. Vesalius (2015). Brain renaissance : from 
Vesalius to
contemporary neuroscience. New York, NY, Oxford University Press New 
York, NY.
Geschwind, N. (1974). Selected Papers on Language and the Brain. 
Dordrecht, Springer
Netherlands Dordrecht.
Habermas, J. ê. (2015). The theory of communicative action : 1. Reason 
and the
rationalization of society. [Place of publication not identified], 
Polity Press [Place of
publication not identified].
Habermas, J. ê. (2015). The theory of communicative action : 2. 
Lifeworld and systems, a
critique of functionalist reason. [Place of publication not identified], 
Polity Press [Place of
publication not identified].
Shannon, C., Weaver, W. (1949). The Mathematical Theory of 
Communication. Urbana,
University of Illinois Press.
von Neumann J., M. O. (1947). The Theory of Games and Economic Behavior. 
Princeton, NJ,
Princeton University Press.
Werner, E. (1988). "Toward a theory of communication and cooperation for 
multiagent
planning." Theoretical Aspects of Reasoning About Knowledge: Proceedings 
of the Second
Conference: 129-143.
Werner, E. (1991). A Unified View of Information, Intention and Ability. 
Decentralized AI. Y. M.
Demazeau, J.P., Elsevier Science Publishers. 2.
Werner, E. (1996). Logical foundations of distributed artificial 
intelligence. Foundations of
Distributed Artificial Intelligence. G. O'Hare, and N. Jennings, Wiley.
Werner, E. (1999). The Ontogeny of the Social Self. Towards a Formal 
Computational Theory.
Human Cognition and Social Agent Technology. K. Dautenhahn, John 
Benjamins: 263-300.
Werner, E. (2023). Brain meta genomics Genome mind mapping: Network 
protocols partition
the developing brain. Internet of Life. To be publshed.
Whitehead, A. N. and B. Russell (1950). Principia mathematica. 
Cambridge, Cambridge
University Press Cambridge.
Wittgenstein, L. (1958). The Blue and Brown Books. Oxford, Blackwell.
Wittgenstein, L. (1958). Philosophical Investigations. Oxford, Blackwell.
Wittgenstein, L. (1969). Schriften. 1 : Tractatus logico- philosophicus, 
Tagebücher
1914-1916, Philosophische Untersuchungen.




---------------------------------------------------------------




-- 
Este correo electrónico ha sido analizado en busca de virus por el software antivirus de Avast.
https://urldefense.com/v3/__http://www.avast.com__;!!D9dNQwwGXtA!VeH2Eob1yXFT1f4kPK6abL5zd-oLgx5iToa3_0ZbpxK3BM_cULGtxRsKztu2PObLEJfXb8U_iatNa1JbtJMTqV0_nrMK$ 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listas.unizar.es/pipermail/fis/attachments/20231012/a9bae7be/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Social Information in Artificial Social Intelligence V3.pdf
Type: application/pdf
Size: 225596 bytes
Desc: not available
URL: <http://listas.unizar.es/pipermail/fis/attachments/20231012/a9bae7be/attachment-0001.pdf>