Approaches to AI from Symbolic Systems to BacktoBasics Evolution

1
Approaches to AI from Symbolic Systems to
Back-to-Basics Evolution

• Name Kim Thompson
• Profession Transit Administration
• Affiliation Washington Metropolitan Area
  Transit Authority
• Relation I find AI fun to think about.
• Question Can greater understanding of
  intelligence help us better relate to
  each other?

2
Local Representations

• Symbolic representations are the lynch-pin of
  classical AI.
• Units of information are shunted around and
  operated on by the model.
• This type of representation is called local
  because information is kept together in a
  locatable package.

3
Distributed Representations

• Neural Networks
• A distributed representation is spread out across
  the whole network.
• Neural networks are built up from atomic units
  artificial neurons but these units are rarely
  used to represent anything in themselves.

4
Complex Activity

• Information is represented by a complex pattern
  of activity over a wide number of neurons.

5
Complementary Approaches

• Connectionism is an approach inspired by the
  neural structure of human and animal brains.
• Symbolic AI, an approach that relies on locatable
  symbols, was the accepted conceptual vocabulary
  of choice for a long time.
• Historically, supporters of symbolic AI and
  connectionism were divided into rival camps.
  Today, most now agree that the two approaches
  complement each other.

6
Can Neural Networks Think?

• Computers manipulate meaningless symbols.
  Computers do not understand the symbols.
• Connectionism can offer some insights to this
  problem
• 1) neural networks are physically different
• 2) neural networks compute at a sub-symbolic
  level

1
2
7
The Chinese Gym

• Searle believes that connectionism cannot bring
  anything new to the AI debate.
• He describes a Chinese Gym filled with
  non-Chinese speakers each representing a neuron
  in a neural network.

8
The Symbol Grounding Problem

• By themselves, symbols are meaningless shapes
  realized by in the case of a conventional
  computer a pattern of electrical activity.
• Any meaning we confer to the symbols is parasitic
  on the meaning in our heads.

9
Breaking the Circle

• Could you ever learn Chinese as a first language,
  with only the aid of a Chinese-Chinese
  dictionary?
• How does one get out of a symbol-symbol
  merry-go-round?

10
Psychologist Stevan Harnard imagines

• A classical symbolic system sitting on top of a
  sub-symbolic connectionist system.
• The connectionist system has inputs that are
  grounded in the outside world through sensors.
• In this way, symbolic representations are no
  longer defined in terms of other symbols, but are
  instead related to iconic representations with
  are directly linked to the sensory surfaces of
  the system.

11
The Demise of AI?

• After a half century of research into AI
  expectations of building a machine that can match
  the cognitive abilities of humans are not being
  met.
• the cognitivistic paradigms neglect of the
  fact that intelligent agents live in a real
  physical world leads to significant shortcomings
  in explaining intelligence.
• --- Rolf Pfeifer and Christian Scheier

12
New AI

• We use to argue whether a machine could think.
  The answer is No.
• What thinks is a total circuit, including perhaps
  a computer, a man, and an environment.
• Similarly, we may ask whether a brain can think,
  and again, the answer is No.
• What thinks is a brain inside a man who is part
  of a system which includes an environment.
• --- Gregory Bateson

13
The Problems of Conventional AI

• Scalability

14
Another Problem ...

• Robustness, or the inability of many systems to
  react well to unforeseen circumstances.

15
Final Problem

• Operating in Real Time
• The sense-model-plan-act cycle leads to massive
  amount of information processing requiring too
  much time.

16
The New Argument from Evolution

• The problem of creating intelligent agents has
  already been solved over the course of 4.5
  billion years.
• Biological evolution builds on existing designs
  by adding the occasional improvement.
• MIT roboticist Rodney Brooks believes that our
  knowledge of evolution can inform AI
• and argues that we should first aim to build
  basic mechanical creatures before we try to build
  mechanical humans.