History of AI

History of AI

Classic AI

Concentrated on:

Human-level intelligence. Abstract reasoning.
Linguistic intelligence.
Symbol using and knowledge representation.
Rule-following and logic.

Archetypal problem - Chess

AI and games - Metric exists.
Great success: AI beat best human at chess.
But now people say: Chess is the kind of thing that is easy for computers. Tennis is hard.
Chess impresses us because we can't do it. Walking doesn't impress because we can. Doesn't mean walking is easy. Rather it means all humans have solved hard problem.

The history of AI: A repeated cycle of excitement and disillusion

Slower than expected progress led to biologically-inspired movement in AI, since mid-1980s.
This itself has also now made slower than expected progress!
Conclusion: AI is hard!

Will probably need both approaches but no one knows how to combine them.

Robotics
- Chess is easy. Walking/running over rough ground is hard (which adult humans, children and animals can all do).

Biologically-inspired AI

Since mid-1980s - Biologically-inspired movement in AI.

AI probably now split 50-50 between "Bio-" and "Non-bio-"
(to use crude stereotypes that do not necessarily fit either)

Themes of Biologically-inspired AI

Action Taker and Evaluator (AUTONOMY)
Learning, Evolution, Self-organisation v. Design, Search
Robustness, Duplication, Multi-minds v. Brittleness (AUTONOMY)
Non- (or Sub-) Symbolic v. Symbolic
Sensorimotor v. Cognitive
Animal Behavior, Cognitive Science v. Engineering (AUTONOMY)
(Whole) Animals v. (Parts of) Humans
Situated (in stream of IO) v. Isolated
If human-level AI is too hard, look for "Animat path to AI" (e.g. how nature actually did it)
A-life v. A-animals, bacteria, insects, dogs, chimps, H.erectus and infants
Symbol-grounding, origin of representations

Analysis v. Synthesis

Whole point of AI is division between:

Analysis (biology, neuroscience, psychology) - reverse engineering with no manual, very difficult.
and:
Synthesis (AI as part of cognitive science) - try out models of mind by building them! We know what the component parts are (we can label them!). Much easier to study the whole system.

Ironically, the more you allow machines learn and self-adapt (especially evolution of neural networks or program code) the more you run into analysis problems of the final result - just like nature again!

Reading

The Classic-AI v. Bio-AI debate

AI is possible .. but AI won't happen: The future of Artificial Intelligence, Mark Humphrys, 1997. - Online. - My simple, short, popular-science summary of the history of AI, emphasising the recent biologically-inspired trends.

Intelligence without Reason, Rodney Brooks, 1991. - Online. - A lengthy history of AI from the biologically-inspired angle.

Today the earwig, tomorrow man?, David Kirsh, 1991. - Online. - A reply to Brooks.

Bio-AI

Why not the whole iguana?, Dennett, 1978. - See Full reference. - A classic call to build whole creatures.

The animat path to AI, Stewart Wilson, 1990. - Online. - Dennett's call taken up by a whole new movement.

Out of Control: The New Biology of Machines, Kevin Kelly, 1994. - Online. - A mind-bending description of the type of machines we're trying to build.

The failure of bio-AI

AI is still not solved. Why is AI so hard?
Some more of my discussion of the history of AI, and where AI is going:
- AI is too big a problem
- "Not enough stuff" in AI
- Maybe "Not enough time" can also be added to the above. We get 20 years to grow our intelligence. Maybe a machine needs a 20 year childhood too (and switched on and interacting with the world 15 hours a day - whereas most robots in labs are actually switched on and interacting for less than 1 percent of their life).