RPE Presentation 1

Joshua Schwartz

Kansas State University

14 September 2021
The purpose of this presentation

I want to convince you that:

I Studying artificial general intelligence, with the goal of producing a machine with
human-level intelligence, is a worthy pursuit
I The specific, narrow problem identified herein will be useful in that line of inquiry,
and will produce interesting results within a semester
Watterson, 1993
Watterson, 1995
Watterson, 1995
Turing, 1950

“Computing machinery and intelligence”

I “I propose to consider the question, ‘Can machines think?’”
I “Imitation game,” a.k.a. “Turing test”
I “Are there imaginable digital computers which would do well in the imitation
game?”
I “I believe that at the end of the century. . . one will be able to speak of machines
thinking without expecting to be contradicted”
I “Learning machine”/“child machine”
I “We can only see a short distance ahead, but we can see plenty there that needs
to be done”
Hillis, 1985

The Connection Machine

I Section 1.1: “We would like to make a thinking machine”
I Thinking Machines Corp.
Goertzel/Pennachin, 2007
Artificial General Intelligence
I “What is meant by AGI is, loosely speaking, AI systems that possess a reasonable
degree of self-understanding and autonomous self-control, and have the ability to
solve a variety of complex problems in a variety of contexts, and to learn to solve
new problems that they [didn’t] know about at the time of their creation”
I “The grand goal of AGI remains mostly unrealized, and how long it will be until
this situation is remedied remains uncertain”
I “[AGI] was the original focus of the AI field, but due to the demonstrated difficulty
of the problem, not many AI researchers are directly concerned with it anymore”
I “Like nanotechnology, it is ‘merely an engineering problem,’ though certainly a
very difficult one”
I “Given the diverse and inter-contradictory nature of the different AGI approaches
[here], it stands to reason that a good percentage of the authors have got to be
significantly wrong on significant points!”
Hitzler/Kühnberger, 2010

“Facets of artificial general intelligence”

I “We argue that [the] time has come for a serious endeavor to work towards
artificial general intelligence”
I Learning: neural networks, MDPs
I Integration: “integrate. . . different forms of reasoning into one uniform
framework”; neural-symbolic integration; cognitive architectures
I “The future will tell whether there will be light at the end of the tunnel”
Lighthill, 1972
“Artificial intelligence: A general survey”
I “[T]here is a vast difference of approach between the practical, technological aims
of category A (Advanced Automation of human activities) and the fundamental,
biological aims of category C (Computer-based CNS studies)”
I “[T]he whole case for the existence of a continuous, coherent field of [AI research]
depends critically on whether between categories A and C there exists a significant
category of research that may be described as a ‘Bridge’ category, B”
I “[A] general-purpose program, the coveted long-term goal of AI activity, seems as
remote as ever”
I “[O]ne rather general cause for the disappointments that have been experienced:
failure to recognise the implications of the ‘combinatorial explosion’”
I “[P]ure mathematical-logic methods suffer defeat at the hands of the
combinatorial explosion, and have to be replaced by ‘heuristic’ methods”
I From the debate with McCarthy/Michie/Gregory: “The general-purpose robot is
a mirage”
Dreyfus, 1972
What Computers Can’t Do: A Critique of Artificial Reason
I “[A]n overall pattern has emerged: success with simple mechanical forms of
information processing, great expectations, and then failure when confronted with
more complicated forms of behavior”
I “[T]he game seems to be to see how far one can get with the appearance of
complexity before the real problem of complexity has to be faced”
I “Judging from their behavior, human beings avoid rather than resolve the
difficulties confronting workers in [AI] by avoiding the discrete
information-processing techniques from which these difficulties arise”
I “[T]hat the world can be exhaustively analyzed in terms of determinate data or
atomic facts is the deepest assumption underlying work in AI”
I “It turns out that it is the sort of intelligence which we share with animals. . . that
has resisted machine simulation”
I “[A] ‘machine’ which could use natural language and recognize complex patterns
would have to have a body so that it could be at home in the world”
Fjelland, 2020

“Why general artificial intelligence will not be realized”

I “One of Dreyfus’ main arguments was that human knowledge is partly tacit, and
therefore cannot be articulated and incorporated in a computer program”
I “Computer power will never develop into human reason, because the two are
fundamentally different”
I “The main thesis of this paper is that we will not be able to realize AGI because
computers are not in the world”
I “My conclusion is very simple: Hubert Dreyfus’ arguments against general AI are
still valid” (in spite of the “deep learning revolution”)
Landgrebe/Smith, 2019

“There is no artificial general intelligence”

I “[T]he ability of a machine to conduct convincing dialogues with human
beings. . . should be accepted. . . as a necessary condition of AGI”
I “A dialogue is a complex stochastic temporal process. . . that lacks the Markov
property.. . . All such systems. . . elude mathematical modeling”
I “[T]here are so many different sources and dimensions of variance involved in a
communicative act that the possibilities of forming an utterance are practically
infinite”
I “[I]t is for mathematical reasons impossible to program a machine in such a way
that it could master human dialogue behaviour in its full generality”
I Fundamentally circular: machines cannot have general intelligence because
machines cannot have general intelligence
Wolfram, 2002
A New Kind of Science
I “I believe that what is by far the most likely is that at the lowest level space is in
effect a giant network of nodes”
I “Just as for space, it is my strong belief that time is fundamentally discrete”
I “[T]he universe might work. . . like a mobile automaton or Turing machine, in
which just a single cell gets updated at each step”
I “[M]y strong belief [is] that. . . every detail of our universe does indeed follow rules
that can be represented by a very simple program—and that everything we see
will ultimately emerge just from running this program”
I “I strongly suspect that there is a definite core to the phenomenon of human
thinking. . . based on rules that are rather simple”
I “[V]ery simple rules are quite sufficient to produce highly complex behavior”
I “[A]ll processes, whether they are produced by human effort or occur
spontaneously in nature, can be viewed as computations”
Adams/. . . /Goertzel/. . . , 2012

“Mapping the landscape of human-level artificial intelligence”

I “The heterogeneity of general intelligence in humans makes it practically
impossible to develop a comprehensive, fine-grained measurement system for AGI”
I “There are nearly as many different AGI architectures as there are researchers in
the field”
I “[S]even scenarios will be presented as milestones suggesting a roadmap across
the AGI landscape”: arbitrary job automation, general video-game learning,
preschool learning, reading comprehension, story or scene comprehension, school
learning, Wozniak test
Goertzel, 2014
“Artificial general intelligence: Concept, state of the art, and future prospects”
I “Core AGI hypothesis: the creation and study of synthetic intelligences with
sufficiently broad (e.g. human-level) scope and strong generalization capability, is
at bottom qualitatively different from the creation and study of synthetic
intelligences with significantly narrower scope and weaker generalization
capability”
I “[M]ost contemporary approaches to designing AGI systems fall into four toplevel
categories: symbolic, emergentist, hybrid and universalist”
I “The problem of defining agreed-upon metrics for incremental progress remains
largely open”
I “Current AGI theory is a patchwork of overlapping concepts, frameworks and
hypotheses, often synergetic and sometimes mutually contradictory”
I “What might a general theory of general intelligence look like?”
Goertzel, 2021

“The general theory of general intelligence: A pragmatic patternist perspective”

I “Patternist philosophy of mind”: “understanding of a mind as the set of patterns
associated with an intelligent system”
I “[D]istinction metagraphs, with typed nodes and links”: “foundational knowledge
representation and meta-representation scheme for AGI theory and practice”
I “Achievement of reasonably high degrees of general intelligence under conditions
of constrained resources relies heavily on ‘cognitive synergy’”
Hypothesis A

We can construct a generally intelligent agent.

. . . but how?
One particular approach (hereafter “the book”)

Toward Human-Level Artificial Intelligence:

Representation and Computation of
Meaning in Natural Language
Philip Jackson
Dover Publications, 2019
9780486833002
Definitions

Adapted from the book’s glossary.

(Linguistic) Concept
Any thought, percept, action, belief, idea, or similar representable as a natural
language expression.

Executable concept (xconcept)

A concept describing a process or behavior for an agent to perform.

Conceptual language
A language for expressing concepts internally within an intelligent system.

Tala
A conceptual language based on English syntax, developed preliminarily in the book.
Definitions (cont.)

TalaMind
The theoretical approach of the book and its hypotheses, and an architecture the book
discusses for design of systems according to the hypotheses.

Tala agent
A system that implements the TalaMind architecture, to act as an agent within an
environment.

Intelligence kernel
A system of concepts that can create and modify concepts to behave intelligently
within an environment; a way of describing a “baby machine” approach to AGI, as a
self-extending system of concepts.
TalaMind hypotheses

1. Intelligent systems can be designed as intelligence kernels.

2. The concepts of an intelligence kernel may be expressed in an open, extensible
conceptual language, providing a representation of natural language semantics
based very largely on the syntax of a particular natural language such as English,
which serves as a language of thought for the system.
3. Methods from cognitive linguistics, e.g., constructions, mental spaces, and
conceptual blends, may be used for multiple levels of mental representation and
computation.
TalaMind responses to objections

I Dreyfus
I “This thesis does not assume the mind operates with a fixed set of formal
rules. . . [xconcepts] may be modified by other [xconcepts], or accepted as input from
the outside environment”
I “The TalaMind approach allows specifying executable concepts very generally, and
using or adapting such concepts to support actual, specific contexts, without having
to specify all possible conditions of a context”
I Lighthill
I . . . ???
Jackson, 2020

“Understanding understanding and ambiguity in natural language”

I “[I]t does not appear there is any valid theoretical reason why the syntax and
semantics of a natural language like English cannot be used directly by an AI
system as its language of thought, without translation into formal languages, to
help achieve human-level AI”
I “[T]here is not a valid theoretical reason why ambiguity is a barrier to the use of a
natural language of thought by an artificial cognitive system”
I “Given the universal, extensible explanatory semantics of natural language, the
potential scope of human-level intelligence and artificial intelligence with natural
language is a ‘beginning of infinity’”
Hypothesis B

We can construct a general intelligence kernel.

Hypothesis C

We can construct a general Tala agent.

Genesereth/Thielscher, 2014

General Game Playing

I “General game players are computer systems able to play strategy games based
solely on formal game descriptions supplied at ‘runtime’”
I “[GGP] raises questions about the nature of intelligence and serves as a laboratory
in which to evaluate competing approaches to artificial intelligence”
I “Unfortunately, building programs to play specific games has limited value in AI”
Hypothesis D

We can construct a general game-playing Tala agent.

Hypothesis E

We can construct a Tala agent to play any particular game optimally.

Hypothesis F

We can construct an efficient Tala agent to play any particular game optimally.
Hypothesis G

We can construct an efficient Tala agent to play tic-tac-toe optimally.

A sketch of the plan

1. Re-implement a variant of the TalaMind architecture for use here and possible
future experimentation, using my own flavor of Tala based on what’s in the book
2. Teach a Tala agent to play tic-tac-toe
2.1 Hard-coded
2.2 No initial knowledge of the game; rules and strategy provided as input
2.3 Rules provided as input, strategy deduced?