Professor Timo Honkela presented an argument that there is a analogy between the developments that took place in Ludwig Wittgenstein's philosophy and in the artificial intelligence when turned away from relying rule-based systems. Honkela also discusses more in general epistemological questions, the underlying questions regarding the objectives and motivations of formalisation. Formalization often relies on assumptions such as the basic or primary role of objects, relations and properties or truth values and propositions. Honkela proposed a pattern and distribution based epistemology as an alternative.
Timo Honkela: From early to later Wittgenstein and Artificial Intelligence
1. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
From early to later Wittgenstein and AI:
On human language and knowledge
Timo Honkela
University of Helsinki
Tue, May 15, 2018
Chaired by
prof. Thomas Wallgren
https://www.aljazeera.com/indepth/features/2017/05/scientist-race-build-peace-machine-170509112307430.html
http://cmi.fi/2017/04/12/dialogue-million-people-finnish-professor-envisions-machines-can-build-peace/
2. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Analogy between how Wittgenstein's philosophy and
artificial intelligence have developed?
● Tractatus Logico-Philosophicus (1921)
→
Philosophical Investigations (1952)
● AI based on symbol manipulation and
logical reasoning ( most intensively → 1980s )
→
AI focusing on statistical machine learning, neural
networks, pattern recognition
( gradually 1990s → )
3. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Intellectual context
● Bertrand Russell
● George Moore
● Gottlob Frege
● and others
Also to be noted: Wittgenstein was educated as an engineer
4. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Some observations on Tractatus
● 1.1. Die Welt ist die Gesamtheit der Tatsachen, nicht
der Dinge.
Ogden: The world is the totality of facts, not of things.
GT: The world is the totality of facts, not things.
● 7. Wovon man nicht sprechen kann, darübe muss
man schweigen.
Ogden: Whereof one cannot speak, thereof one must
be silent.
GT: Whereof one can not speak, thereof one must be
silent.
5. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Some word frequencies
● der 667
●
die 520
●
ist 449
● und 302
● nicht 270
●
dass 214
●
das 211
●
in 208
● von 195
●
es 192
● Satz 189
●
Sätze 119
●
Form 89
● Zeichen 80
● Welt 79
●
Logik 77
●
Bild 59
●
Sinn 58
● Satzes 52
●
Tautologie 36
6. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
From early to later Wittgenstein
● Primacy of language
● Primacy of verifiable
propositions
● Language as a
logical system:
words ↔
sentences ↔
● Method of description
● Language games
– Contextuality
● Family resemblance
– Link to distributions
– Vs. Socrates et al.
● Understanding where
we do not understand
7. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
From GOFAI to modern AI
● Knowledge representation
based on hierarchical
symbol structures and
● Reasoning based on
application of deductive
rules
● Learning based on
inductive; problem solving
on abductive reasoning
● Stastical machine learning applied
for various purposes
● Fuzzy sets can be used to model
similarity
● Pattern recognition builds a bridge
between world of senses and
symbols
● Neural networks can, for instance,
build novel conceptual systems
● Meaning represemted in continuous
multidimensional spaces
8. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
LehtolaAarnoandHonkelaTimo(1987):
AWARE-DAGTransformationsforSemanticAnalysis.
NordicConferenceonComputationalLinguistics,Copenhagen.
KIEL
9. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
A carefully
crafted hierarchical
system of
rule systems
each designed
for its purpose
separately
LehtolaAarnoandHonkelaTimo(1987):
AWARE-DAGTransformationsforSemanticAnalysis.
NordicConferenceonComputationalLinguistics,Copenhagen.
10. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
LehtolaAarnoandHonkelaTimo(1987):
AWARE-DAGTransformationsforSemanticAnalysis.
NordicConferenceonComputationalLinguistics,Copenhagen.
KIEL
SQL query
Logical
representation
of meaning Universal
query
language
Dependency
parsing
11. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Questions / challences
● Is there world beyond (what is or can be
described using) language?
● What is the ontological, epistemological and
methodological status and role of emergence?
● What is the ontological role of formalization?
Are there psychological motivations for formalization?
● Is the interpretation of linguistic meanings
subjective to some degree (as Honkela
suggests)? Cf. private language argument.
12. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
On mathematics of meaning
● Spatio-temporal: space and numbers
● Euclidian versus Riemannian and Lobochevskian
● Objectivity of mathematics?
● Mathematics in/for
– Physical reality
– Biological reality
– Cultural reality
13. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Weakness of objects
as a basis of epistemology
● We see and formalize the world as a collection
of objects, their relation and properties as well
as events in which objects are involved
– Note: object / objectivity
● This does not mean that the world would
simply/solely consists of objects, reletionship
and events
● We can (re)consider the basis of our ontological
assumptions and epistemological formalizations
14. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
On numbers and how we
conceptualize the world
3 + 2 = 5
Matters in physical, biological and
cultural domains: matters dependent
and not dependent on our decisions,
agreements and behaviours
15. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Truth versus understanding
● If we broaden our view from logical systems
into particular mathematical systems, we can
considered truth as a relative/contextual
notion, and understanding to be a primary
notion (understanding the world and the others)
● The reason for the fear for the relativity of truth
may stem from (1) ”anything goes”, (2) we
would not have solid basis for epistemology
16. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Potential roles of AI in philosophy
● AI as a means to test philosophical ideas
● AI as a means to build philosophical arguments
● AI as a means to build bridges between empirical
sciences and philosophy
● AI as a means to build philosophical thought systems
– Here AI refers to statisticial machine learning, pattern recognition,
language and speech technology, text mining, modeling of socio-
cognitive systems, neural networks, models of conceptual spaces,
applications of catastrophy and chaos theory, evolutionary algorithm,
fuzzy sets, mathematical models taking distance from natural sciences,
algorithms developed for humanities and social science, etc.
17. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Thank you for your attention!
18. Timo Honkela on AI and Wittgenstein in the von Wright and Wittgenstein research seminar, 15th May 2018, Metsätalo, University of Helsinki
Abstract
Natural language processing and knowledge representation have been central tasks in artificial
intelligence (AI) research since its early stages.
These tasks most often aim at reaching some technological solutions.
Attempts to develop AI systems are based on ontological and epistemological assumptions that
guide the development explicity or implicitly.
Philosophical understanding can guide the development of AI systems the tasks of which are
inherently linked within human language and knowledge.
On the other hand, the development of AI systems may potentially provide understanding of
philosophical questions that would be difficult or even impossible without the use of computational
methods and models.
In this presentation, some elements of Wittgenstein's philosophical work and its development is
considered in parallel with the developments that have taken place in AI research over the past
half a century.
Ways on how computational modeling could be used as empirical theoretical philosophy are
considered.
In addition, some current and hypothetical future means for studying philosophical works through
computational means is discussed.