How to Build a Mind

Unfortunately psychology had an accident aong the way. Hemholtz, Vagner?, … did good experiments but at same time Psychoanalysis came along

Rene Descartes Thinking Substance vs Located Substance For Decartes - Mind must be independent.

Turing machines give an intuition of linearity, how will there be concurrency, ….

0:20:45 - It became lambda because of typesetting problem :rofl:

cut argument and paste it in place of bound variable inside expression. So, lambda calulus is more of a - cut and paste thingy.

• Turing machines can compute Lambda calculus
• Lambda calculus can compute Turing machines

— equally powerful!

We have found other methods of computation:

• Mathematical proofs
• Partial Recursive function
• POS machine - a variation of Turing Machine

Church-Turing Thesis: All definitions of computation have the same power

Principle of Neural Information Processing are computational -

• Compartment models
• Integrate and Fire model - Simplification
• Artificial Neural Network (Weights, Bias, Transition function) - Crude Simplification

• Minds are computer programs (realized in neural hardware configuration and individual neural states)
• Mental content is represented in these programs - Program as Data
• Perception encodes information given at the system boundary of the mind as mental content
• Thinking, feeling, dreaming, … are aspects of operations over mental content

All the universe can do to us is:

• present information
• produce regularities in this information (i.e. compute)

No conceptual difference between reality and a computer program

Neither mind is primary nor is physical world, computation is primary.

• Only information is primary
• Mind and matter are constructions (encodings) over information
• All regularity in perceived information can be captured in computational models

Roger Penrose:

• The brain uses quantum processes to produce consciousness

— minds are more than computers

“The quality of understanding and feeling possessed by human beings is not something that can be simulated computationally.”

0:27:41 But quantum process are computational too!

Test question: can you solve undecidable problems?

If you choose one of the following answers randomly, what is the probability that your answer is correct?

• 50%
• 25%
• 100%
• all of the above 25%

How Difficult is AI?

~100 Bn neurons

Cells in the cortex are organized into ca. 2 million complex, regular circuits (columns)

• Brain has a distinct architecture
• Upper bound of complexity of constructing a brain is given by human

0:32:03

0:32:05 Computation complexity of complete human being is less than Microsoft Windows (If Physics and Chemistry are given) 0:32:24 Its the Upper bound because neurons are just a part of it 0:32:29 Its not easy to reverse engineer - its just not hopeless

0:36:17 Marvin Minsky coined AI at Darthmouth Conference - 1956

Marvin Minksy wrote against neural networks - they won - so for most of the time there was no progress in that part for about a decade

• 0:37:31 Expert systems gave diminishing returns

• Food
• Water
• Integrity

they structure our social interaction

• Affiliation
• Internal Legitimacy

• state of arousal
• direction of attention (inwards for outwards)
• resolution level

these dimensions regulate/modulate our congnition. so we experience, understand something differently

Emotion in Computation Models

• Affect can be seen as a configuration of a cognitive system
• Modulators of cognition:
  • arousal, selection threshold, securing threshold, resolution level
  • estimate of competence and certainty
  • pleasure/distress signals - valence
• Affective state is emergent property of modulation
• 0:47:37 Directed affects (higher-level emotions) (jealosy, pride,…) emerge by association of demand with appetive or aversive objects/situations

• abstract and perceptual,
• how to work with ambiguious
• superpositional
• inheritance and polymorphism,
• any-time construction (using constraints)

• including temporal relations, loops, etc

  0:51:28 Constraints for Partial Represenatation ??

• bottom-up - cue driven - driven by environment - what we see
• top-down: theory/concept driven - driven by what we expect to see
• works on sensory input and memory
• any-time characteristics => works on partial representation??

• Saliency (Similarity and Dissimilarity)- driven
• Goal - driven
• Social (from input from others)

We want to have general solutions

organizing mental represenatation (which are probably Fuzzy distributed hypergraphs) into discrete strings of symbols

• theory of mind
• Personhood and Self concept
• personality properties

• how to integrate analytical and associative operations work together
• Attention management
• complete system insetad of isolated modules
• 0:54:12 Developmental trajectory - start with kids and grow to be more adult like
• persistence - system should be stable

• Must enforce universal problem solving, representation, perception
• Must support incremental development
• Focus on child performance first
• Allow for comparison between approaches
• Inspire a productive research community
• Attract funding

Backlinks

• Joscha Bach