Invited talk @ICAR meet 2020 at the CNR.

1. Commonsense reasoning as a key feature for dynamic
knowledge invention and computational creativity
Antonio Lieto
Università di Torino, Dipartimento di Informatica, IT
ICAR-CNR, Palermo, IT
ICAR Meet 2020
https://speakers.acm.org/speakers/lieto_12489

2. Commonsense Reasoning
Concerns all the non standard (or non monotonic)
forms of reasoning
- induction
- abduction
- default reasoning
- …
2

3. Commonsense Reasoning
Concerns all the non standard (or non monotonic)
forms of reasoning
- induction
- abduction
- default reasoning
- …
3
TIPICALITY

4. ….
AI and CogSci Approaches to Commonsense
Reasoning (partial overview)
Semantic Networks
(Collins and Quillians, 1969)
Fuzzy Logic
Zadeh, 1966
Frames
(Minsky, 1975)
Scripts
(Shank & Abelson,
1977)
Circumscription
(Mc Carthy,
1980,86)
Default Logic
Reiter (1980)
Newell Simon, GPS
(1962)
…

5. ….
AI and CogSci Approaches to Commonsense
Reasoning (partial overview)
Semantic Networks
(Collins and Quillians, 1969)
Fuzzy Logic
Zadeh, 1966
Frames
(Minsky, 1975)
Scripts
(Shank & Abelson,
1977)
Circumscription
(Mc Carthy,
1980,86)
Default Logic
Reiter (1980)
Newell Simon, GPS
(1962)
…
Cognitive Heuristics

6. ….
AI and CogSci Approaches to Commonsense
Reasoning (partial overview)
Semantic Networks
(Collins and Quillians, 1969)
Fuzzy Logic
Zadeh, 1966
Frames
(Minsky, 1975)
Scripts
(Shank & Abelson,
1977)
Circumscription
(Mc Carthy,
1980,86)
Default Logic
Reiter (1980)
Newell Simon, GPS
(1962)
…
Cognitive Heuristics
Machine-oriented Heuristics

7. Cognitive AI/Computational CogSci

8. Lieto, 2021, Cognitive Design for Artificial Minds, Taylor & Francis.

9. Goal-Directed Systems
Goal-directed problem solving is a crucial activity for natural and
artificial intelligent systems (Aha 2018).
Classical strategies for achieving a goal (if the goal cannot be
directly reached) concern knowledge expansion/augmentation
activities:
- e.g. via the communication with another agents
- via a learning process
- via an external injection of novel knowledge in the declarative
memory of an artificial system
9

10. Outline
• Main idea: some goal-reasoning tasks can be solved by a
process of creative commonsense knowledge
invention based on a recombination of the knowledge
already available
– Knowledge invention as a process of creative
commonsense combinatorics
– This generative phenomenon represents an open problem
in current research in AI and automated reasoning: lack of
heuristic frameworks able to automatically deal with this
faculty.
10

11. Guppy Effect (PET FISH problem)
Prototypes are not compositional (Osherson and Smith, 1981).
The resulting PET FISH concept is not merely composed by the additive inclusion
of the typical features of the two composing concepts (i.e. PET and FISH).
Fish = {Greyish, Lives-in Water, not Warm.. }
PET = {hasFur, Warm, not Lives-in Water… }

12. Guppy Effect (PET FISH problem)
Prototypes are not compositional (Osherson and Smith, 1981).
The resulting PET FISH concept is not merely composed by the additive inclusion
of the typical features of the two composing concepts (i.e. PET and FISH).
PET = {hasFur, Warm, not Lives-in Water… }
PET Fish =
{Lives-in Water, not Warm,
Red.. }
Fish = {Greyish, Lives-in Water, not Warm.. }

13. TCL
A non monotonic Description Logic of typicality (TCL), for typicality-
based concept combination based on 3 ingredients
• Description Logics with Typicality (ALC + T)
• Probabilities and Distributed Semantics (Disponte)
• Heuristics from Cognitive Semantics (HEAD-MODIFER)
Lieto & Pozzato, "A Description Logic Framework for Commonsense Conceptual
Combination Integrating Typicality, Probabilities and Cognitive Heuristics", in Journal of
Experimental & Theoretical Artificial Intelligence, 2020. https://arxiv.org/pdf/1811.02366.pdf

14. DL with Typicality (ALC+T)
Non-monotonic extensions of Description Logics for reasoning about
prototypical properties and inheritance with exceptions.
Basic idea: to extend DLs with a typicality operator T (Giordano et al.
2015)
- A KB comprises assertions T(C) ⊑ D
- T(Student) ⊑ FacebookUsers means “normally, students use Facebook”
T is nonmonotonic

15. 15

16. 16

17. 17

18. Distributed Semantics
We extended the ALC+T Logic with typicality inclusions equipped by real
numbers representing probabilities/degrees of belief.
We adopted the DISPONTE semantics (Riguzzi et al 2015) restricted to typicality
inclusions:
extension of ALC by inclusions p :: T (C ) ⊑ D
epistemic interpretation: “we believe p that typical Cs are Ds”
The result of this integration allowed us to reason on typical probabilistic scenarios
18

19. DISPONTE semantics
19
8 possible worlds
Query: NatureLover (kevin)?
P(NatureLover (kevin)) = 0.3 × 0.6 × 0.9 = 0.162

20. Cognitive Heuristics
Heuristics from cognitive semantics for the identification of plausible
mechanisms for blocking-inheritance.
HEAD-MODIFIER heuristics (Hampton, 2011):
- HEAD: stronger element of the combination
- MODIFIER weaker element
where C ⊑ CH ⊓ CM
The compound concept C as the combination of the HEAD (CH) and the
MODIFIER (CM)

21. (TCL) at work - Pipeline
21
1. KB with real data 2.Probabilistic
Scenarios
3. Selection of the most
appropriate scenarios
in TCL we assume a hybrid KB (Rigid and Typical Roles)

22. Selection Criteria
The typical properties of the form T(C) ⊑ D to ascribe to the
concept C are obtained in the set of scenarios, obtained by
applying the DISPONTE semantics, that are:
• consistent with respect to KB;
• not trivial, e.g. those ascribing all properties of the HEAD
are discarded;
• giving preference to CH w.r.t. CM with the highest
probability

23. (TCL) at work: PET FISH
23

24. (TCL) at work: PET FISH
24

25. (TCL) at work: PET FISH
25

26. (TCL) at work: PET FISH
26

27. (TCL) at work: PET FISH
27

28. (TCL) at work: PET FISH
28

29. Applications
- Computational Creativity
- Characters Generation
- Novel Genre Generation
- Recommender Systems
(Chiodino et al, ECAI 2020)
with
Centro Ricerche RAI
Cognitive modelling
Linda problem; Lieto & Pozzato, JETAI 20)

30. Goal oriented Knowledge Generation
Definition 1. Given a knowledge base K in the logic T
CL
, let G be a set of
concepts {D1, D2, . . . , Dn} called goal.
G = {Property1, Property2, Property3…}.
We say that a concept C is a solution to the goal G if either:
– for all Di ∈ G, either K |= C ⊑ D or K0 |= T(C) ⊑ D in the logic T
CL
or:
– C corresponds to the combination of at least two concepts C1 and C2
occurring in K, i.e.
C ≡ C1 ⊓ C2, and the C-revised knowledge base Kc provided by the logic
T
CL
is such that, for all Di ∈ G, either Kc |= C ⊑ D or Kc |= T(C) ⊑ D in T
CL

31. Concept composition
We tested our system on a task of concept composition for a
KB of objects.
GOALS
KB TCL

32. GOCCIOLA http://di.unito.it/gocciola
G = {Object, Graspable, Launching objects at distance}
32

33. Evaluation (30 subjects)

34. Cognitive Architectures
34
Allen Newell (1990)
Unified Theory of Cognition
A cognitive architecture (Newell, 1990) implements the
invariant structure of the cognitive system.
It captures the underlying commonality between different
intelligent agents and provides a framework from which
intelligent behavior arises.

35. SOAR Integration
Lieto et al. 2019, Cognitive Systems Research, Beyond Subgoaling, A dynamic knowledge generation framework
for creative problem solving in cognitive architectures.

36. DENOTER
Multimedia
Content
Crawling
…
…
Metadata
Extraction
& Selection
Extraction
Frequentist Data-Driven
Genre Attribution
DRAMA
0.85 :: T(Drama) ⊑
Homicide
0.83 :: T(Drama) ⊑ Life
TCL KB population
…
…
…
THRILLER
KIDS
TCLExploitation
KIDS_DRAMA
…
Reclassification
KIDS_DRAMA
Multimedia
Content
…
Recommendation

37. Threefold Evaluation
1) AUTOMATIC: % of reclassified content

38. Threefold Evaluation
1) AUTOMATIC: % of reclassified content
2) USER STUDY (involving 20 persons evaluated a total of 122
recommendations generated by the system).

39. Threefold Evaluation
1) AUTOMATIC: % of reclassified content
2) USER STUDY (involving 20 persons evaluated a total of 122
recommendations generated by the system).
3) EXPERTS INTERVIEW

40. Experts Interview (in RAI)
Some problematic elements emerged, that affected the overall
quality (and therefore the assigned ratings) of the recommended
items.
- Recommendation of very old episodes (e.g. episodes or
movies done before the ’60s).
- The experts pointed out the importance of considering the
possibility of expliciting also negative preferences (e.g. ¬Kids).

41. Upshots
I have showed how the cognitively-inspired TCL
commonsense reasoning framework has been applied in the
context of two different applications
This commonsense reasoning framework can be used for more
applications since it models an ubiquitous phenomenon
We plan to extends the applications to other domains
(museum recommendations, music etc.).

42. References
• Lieto, A., & Pozzato, G. L. (2020). A description logic framework for commonsense conceptual combination
integrating typicality, probabilities and cognitive heuristics. Journal of Experimental & Theoretical Artificial
Intelligence, 32(5), 769-804. https://doi.org/10.1080/0952813X.2019.1672799
• Chiodino, E., Di Luccio, D., Lieto, A., Messina, A., Pozzato, G. L., & Rubinetti, D. (2020). A Knowledge-
based System for the Dynamic Generation and Classification of Novel Contents in Multimedia Broadcasting.
In Proc. of ECAI (Vol. 2020), 680 - 687. http://ebooks.iospress.nl/volumearticle/54949
• Lieto, A., Perrone, F., Pozzato, G. L., & Chiodino, E. (2019). Beyond subgoaling: A dynamic knowledge
generation framework for creative problem solving in cognitive architectures. Cognitive Systems Research,
58, 305-316. https://doi.org/10.1016/j.cogsys.2019.08.005
• Chiodino, E., Lieto, A., Perrone, F., and Pozzato G.L. "A goal-oriented framework for knowledge invention
and creative problem solving in cognitive architectures", in ECAI 2020, 24th European Conference on
Artificial Intelligence, 2020, 2893 - 2894, http://ebooks.iospress.nl/volumearticle/55235
• Lieto, A. and Pozzato G.L. "Applying a description logic of typicality as a generative tool for concept
combination in computational creativity", Intelligenza Artificiale, vol. 13, no. 1, pp. 93-106, 2019. https://
content.iospress.com/articles/intelligenza-artificiale/ia180016
42