1. The Cognitive Paradigm in the
Artificial Intelligence Research
Antonio Lieto
Università di Torino, Dipartimento di Informatica, IT
ICAR-CNR, Palermo, IT
June 10, 2020, Kyiv-Mohyla National University
https://www.antoniolieto.net
2. Driving Questions
- What characterize cognitively inspired AI systems?
- What are examples of cognitively inspired AI systems?
- How do they differ from standard AI systems?
- How can cognitively inspired AI systems be used?
2
4. From Human to Artificial Cognition
4
Inspiration
Why?
Humans (and/or other natural systems) are still,
by far, the best unmatched systems able to solve
a wide-range of problems
5. From Human to Artificial Cognition (and
back)
5
Inspiration
Explanation
6. “Natural/Cognitive” Inspiration and AI
Early AI
Cognitive or Biological Inspiration
for the Design of “Intelligent Systems”
M. Minsky
R. Shank
Modern AI
“Intelligence” in terms of
optimality of a performance
(narrow tasks)
mid‘80s
A. Newell
H. Simon
D. Rumhelart
J. McClelland
N. Wiener
Nowadays:
Renewed attention
“The gap between natural
and artificial
systems is still enormous”
(A. Sloman, AIC 2014).
7. 7
Cognitivism Nouvelle AI
Focus on high level cognitive functions Mainly focused on perception
Assuming structured representations
(physical symbol system, Simon and
Newell, 1976)
Assuming unstructured representation
(e.g. such as neural networks etc.) and
also integration with symbolic
approaches.
Architectural Perspective (integration
and interaction of all cognitive functions)
System perspective (not necessary to
consider a whole architectural
perspective).
Inspiration from human cognition
(heuristic-driven approach)
Bio-inspired computing, bottom-up
approach (for learning etc.).
2 Main Perspectives
8. 10
‘Cognition is a type of computation’
Intelligence as “symbolic manipulation”
Cognitivism & Symbolic Representations
High level of abstraction
9. Nouvelle AI and Connectionism
- Sub-symbolic representations (including deep nets) —-> LEARNING,
PERCEPTION, CATEGORIZATION.
Low-level of abstraction
10. Modern successful AI systems
10
IBM Watson
(symbolic)
Alpha Go (Deep Mind)
(connectionist)
11. A Matter of Levels
Both the “cognitivist” and “nouvelle” approach can realize,
in principio, “cognitive artificial systems” or “artificial
models of cognition” provided that their models operate
at the “right” level of description.
When a biologically/cognitively inspired computational system/
architecture has an explanatory power w.r.t. the natural system
taken as source of inspiration ?
Which are the requirements to consider in order to design a
computational model of cognition with an explanatory power?
Functionalist vs Structuralist Models 11
12. Functionalism
• Functionalism (introduced by H. Putnam) postulates a weak
equivalence between cognitive processes and AI procedures.
• AI procedures have the functional role (“function as”) of human
cognitive procedures.
• Equivalence on the functional macroscopic properties of a given
intelligent behaviour (based on the same input-output specification).
• Multiple realizability (cognitive functions can be implemented in
different ways).
• This should produce predictive models (given an input and a set of
procedures functionally equivalent to what is performed by cognitive
processes then one can predict a given output).
12
13. Problems with Functionalism
• If the equivalence is so weak it is not possible to
interpret the results of a system (e.g. interpretation of
the system failures…).
• A pure functionalist model (posed without structural
constraints) is a black box where a predictive model
with the same output of a cognitive process can be
obtained with no explanatory power.
13
14. Birds and Jets
- Both Birds and Jets can fly but a jet is not a good explanatory model
of a bird since its flights mechanisms are different from the mechanism
of bird.
- Purely functional models/systems are not “computational models of
cognition” (they have no explanatory power w.r.t. the natural system
taken as source of inspiration). 14
15. Modern “Functional” Systems in AI
They are very good artificial systems but they have no explanatory
power with respect to how humans solve/face the same problems.
In this sense they are not cognitive ! (e.g. despite IBM claims) 15
16. 16
They are NOT Cognitive Systems
Extended version of 2D Space of Cognitive Modelling (from Vernon, 2014)
22. Structuralism
• Strong equivalence between cognitive
processes and AI procedures (Cordeschi 2002,
Milkowski, 2013).
• Focus not only on the functional organization
of the processes but also on the human-
likeliness of a model (bio/psychologically
plausibility).
22
23. Wiener’s “Paradox”
“The best material model of a cat is another or possibly the same cat” (Rosenblueth &
Wiener45)
Z.Pylyshyn (’79): “if we do not formulate any restriction about a model we obtain the functionalism of a
Turing machine. If we apply all the possible restrictions we reproduce a whole human being”
- Also for complete simulation of complete models (e.g. very simple organisms like
the Caenorhabditis elegans, Kitano et al. 98) it is problematic a full understanding and
testing of biological hypotheses.
24. A Design Problem
• Need for looking at a descriptive level on which to enforce
the constraints in order to carry out a human-like
computation.
• A design perspective: between the explanatory level of
functionalism (based on the macroscopic stimulus-response
relationship) and the mycroscopic one of fully structured
models (reductionist materialism) we have, in the middle, a
lot of possible structural models.
24
25. “Cognition” as Design Constraint
• We need a function-structure coupling for the design of
cognitive artificial systems.
• The interpretations of the experimental results coming
from cognitive psychology/neuroscience indicate us the
algorithm procedures (the heuristics/design constraints)
that we can implement in our system in a functional-
structural way.
• I.e. the implementation can be done in different ways
(multiple realizability account of the functionalism) but
the model needs to be constrained to the target system
(needs to be structurally valid). 25
26. Many Structural Models
Both the presented AI approaches may build structural
models of cognition at different levels of details (having an
empirical adequacy).
26
Cognitive Function
(NL Understanding)
Cognitive Processes Neural Structures
Sintax
Morphology
Lexical
Processing…
Bio-Physical Plausibility
of the Processes
Cognitive Plausibility
of the Processes
Cognitivism Emergent AI
27. First take home messages
• Cognitive Artificial Models have an explanatory power only if
they are structurally valid models (realizable in different
ways and empirically adequate).
• Neural models are not necessarily more plausible than hybrid
or symbolic models (their explanatory role depends on the
way in which structural constraints are enforced on them)
• Cognitive Artificial Systems built with this design perspective
can be used as “computational experiment” and provide
results useful for refining of rethinking theoretical aspects of
the natural inspiring system.
30. GPS (General Problem Solver)
A system able to demonstrate simple logic theorems whose decision strategies were
explicitly inspired by human verbal protocols (Simon, Shaw, Newell, 1959).
Idea: the computer system had to approximate the decision operations described by the
humans in their verbal descriptions as closely as possible.
In particular, the GPS system was able to implement a key mechanism in human problem
solving: the so called means-ends analysis (or M-E heuristic). In M-E analysis the problem
solver compares the current situation with the goal situation and reduces the difference
(HEURISTIC SEARCH).
30
Nobel Prize
“bounded rationality"
31. Semantic Networks
Ross Quillian (1968) developed a a psychologically plausible model of human semantic
memory implemented in a computer system. The idea of Quillian was that human memory
was associative in nature and that concepts were represented as sort of nodes in graphs and
activated through a mechanism of “spreading activation” allowing to propagate information
through the network to determine relationships between objects.
31
32. RM Model of Past-Tense Acquisition
• Shows how emergentist approach can explain some features of
language acquisition without any predefined rule
• Training of the network corresponds to the development of mental
skills in humans. 32
33. Cognitive Architectures
33
Allen Newell (1990)
Unified Theory of Cognition
A cognitive architecture (Newell, 1990) implements the
invariant structure of the cognitive system.
The work on such systems started in the ‘80s (SOAR
(Newell, Laird and Rosenbloom, 1982)
It captures the underlying commonality between different
intelligent agents and provides a framework from which
intelligent behavior arises.
The architectural approach emphasizes the role of
memory in the cognitive process.
34. DUAL PECCS: DUAL- Prototype and Exemplars
Conceptual Categorization System
Lieto, Radicioni, Rho (IJCAI 2015, JETAI 2017)
36. 36
1) Multiple representations for the same concept
2) On such diverse, but connected, representation are executed
different types of reasoning (System 1/ System 2) to integrate.
2 Cognitive Assumptions
Type 1 Processes Type 2 Processes
Automatic Controllable
Parallel, Fast Sequential, Slow
Pragmatic/contextualized
…
Logical/Abstract
…
37. Heterogeneous Proxytypes Hypothesis
The different proposals that have been advanced can be grouped in three main classes: a) fuzzy approaches, b) probabilistic and Bayesan approaches, c) approaches based on
non-monotonic formalisms.
TIPICALITY
The diverse types of connected representations can coexist and point to
the same conceptual entity. Each representation can be activated as a proxy
(for the entire concept) from the long term memory to the working memory of
a cognitive agent.
(Lieto, A. A Computational Framework for Concept Representation in Cognitive Systems and
Architectures: Concepts as Heterogeneous Proxytypes, Proc. of BICA 2014)
CLASSICAL
38. Ex. Heterogeneous Proxytypes at work
The different proposals that have been advanced can be grouped in three main classes: a) fuzzy approaches, b) probabilistic and Bayesan approaches, c) approaches based on
non-monotonic formalisms.
(Lieto, A. A Computational Framework for Concept Representation in Cognitive Systems and
Architectures: Concepts as Heterogeneous Proxytypes, Proc. of BICA 2014)
39. 39
Co-referring representational Structures via Wordnet
Lieto, A., Radicioni, D., Rho, V, (2017). Dual PECCS: a cognitive system for conceptual representation
and categorization, JETAI, 29 (2), 433-452, Taylor and Francis.
41. Overview
NL Description
-The big fish eating plankton
Typical
Representations
IE step and
mapping
List of Concepts :
-Whale 0.1
-Shark 0.5
-…
Output S1
(Prototype or
Exemplar)
Check on S2
Ontological Repr.
-Whale NOT Fish
-Whale Shark OK
Output S2 (CYC)
Output S1 + S2
Whale
Whale Shark
42. ACT-R, SOAR, CLARION and LIDA Extended Declarative Memories with
DUAL-PECCS
Salvucci et al. 2014 (DbPedia)
Lieto A., Lebiere C., Oltramari A. (2018), The Knowledge Level in Cognitive Architectures: Current Limitations and Possibile
Developments. Cognitive Systems Research (48): 39-55, Elsevier.
45. Evaluation
The different proposals that have been advanced can be grouped in three main classes: a) fuzzy approaches, b) probabilistic and Bayesan approaches, c) approaches based on non-monotonic
formalisms.
Query on commonsense questions
Gold standard: for each description recorded the human answers
for the categorization task.
Stimulus Expected
Concept
Expected Proxy-
Representation
Type of Proxy-
Representation
… … … …
The primate
with red nose
Monkey Mandrill EX
The feline with
black fur that
hunts mice
Cat Black cat EX
The big feline
with yellow fur
Tiger Prototypical
Tiger
PR
46. 46
- Concept Categorization Accuracy
- Proxyfication Accuracy
Evaluation Accuracy
Lieto, A., Radicioni, D., Rho, V, (2017). Dual PECCS: a cognitive system for conceptual
representation and categorization, JETAI, 29 (2), 433-452, Taylor and Francis.
47. Analysis
The different proposals that have been advanced can be grouped in three main classes: a) fuzzy approaches, b) probabilistic and Bayesan approaches, c) approaches based on non-monotonic
formalisms.
- The comparison of the obtained results with human
categorization is encouraging 77-89% (results of other
AI systems for such reasoning tasks are by far lower).
- The analysis of the results revealed that it is not true
that exemplars (if similar enough to the stimulus to
categorise) are always preferred w.r.t. the
prototypes.
- Need of a more fine-grained theory explaining more in
the details the interaction bewteen co-existing
representations in the heterogeneous hypothesis.
48. Upshots
The different proposals that have been advanced can be grouped in three main classes: a) fuzzy approaches, b) probabilistic and Bayesan approaches, c) approaches based on non-monotonic
formalisms.
Cognitively inspired systems have played and play an
important role in AI research.
In general, simple tasks for humans are the most
complicated ones for AI systems (in such problems the
cognitive heuristic approach can play an important role for
the development of better AI systems)
Notable areas: few/one shot learning, commonsense
reasoning, transfer learning, computation creativity,
narrative understanding, heuristic integration of
planning, action and goal-oriented reasoning,
computational models of emotion, analogy and methaphor
based reasoning, cognitive and social robotics,
explainable AI..(many others).
49. Cognitive Design for Artificial Minds
49
Forthcoming in 2021 !!
Taylor and Francis