Historical Perspectives - AI and Cognition

Mind computer analogy

1. mental processes can be understood in terms of similarity to a computer

2. a mental process can be understood as information processing through a system in a stepwise fashion

Attkinson-shiffrin model of memory

external input (from environmental stimuli) → sensory memory (iconic and echoic memory; short-lived) → STM (remembering a phone number and then forgetting) → LTM

Chase & Simon Experiment

• viewed board for 5 seconds then had to rebuild it from memory

• measured number of trials to get the board direct

• LTM helping you do a STM task

• if board was randomized, beginners did better!

• showed that we have models in our heads in LTM that are used to process things

Mental Models

• Johnson Laird = theory that mental representations have 3 forms that can be stored in many ways

• propositional → language like (experts)

• mental model → resembles real world and on a continuum

• mental imagery → visual (novices)

Rumelhart + McClelland = Parallel-distributed processing

• ran binary code processors (neurons) in software that could be stacked in layers (cortical tissue) on top of one another

• the layers connect and energy (action potential) is transmitted to stimulate information

• same process in our brains when we are learning!

Cognitive Modelling in 1970s

• Atkinson-shiffrin box model remained persistent

• evolved to computer simulations and math models that can solve problems!

What is cognitive science?

• the interactive study of the mind (blend of fields to make an integrative idea of the human mind)

• includes cog psych, neuro, com sci, AI, robotics, linguistics, etc.

AI Mind Modelling

• AI strives to model a mind separate from humans that produces goal-directed behavior in the most efficient way possible

Technological evolution

• computers started with vacuum tube (binary process) → transistors (1970s) → microchips (embedding multiple transistors) (1990 -what we use today)

What is the promise of AI?

• promise is that AI will outperform humans and that AI will be more optimized than human thinking

Chess Thinking

• algorithm method = analyzing all possible outcomes

• heuristic method = using shortcuts to analyze best case scenarios

• usually heuristics win but with DEEP BLUE IBM vs Gary Kasparov he lost against AI

IBM Watson & Jeopardy

• 2 top champions vs computer

• Watson reaches confidence level then answers; humans able to get some right

• Computer has trouble with language

• In the end, humans unable to defeat computer

Go game

• players try to capture territory

• Einstein used to play

• opens with blank board; has higher branching factor and board configurations

Alpha-go

• alphago software defeated Lee Sedol in go game in 2016