Invited Speech

Why We Can Be Confident of Turing Test Capable AI Within a Quarterly Century

Ray Kurzweil

Children still exceed the machines

• Minsky pointed out that problem solving occupies adults -- we still can't tell a machine to tie shoelaces
• Collecting images used to be a daunting task, but not anymore. (images for the blind, collecting data for pattern recognition)
• Recent development of tremendous database (database on net)

150 years ago: apply software to model human intelligence (analytical engine, speculation on AI)
Key to success in invention is timing. One should study technology trends and mathematical models for technology trends.

• Discovery of DNA
• Measure of information technology
• The power of information technology doubles every year
• Spatial resolution of brain modeling doubling every year
• Next 50 years should be 32 times greater than now, given the pace progress

• Put more resources, more research that accelerates the pace progress
• Technology improves biology -- monogenic data, brain scanning data, etc.
• We haven't got much information from brain science yet, but plug in without understanding how it works - get good results

You can't tell the future vs. examples of how predictable it is

• Unpredictable particles & molecules -> overall predictable result (random interactions at low level, predictable at cumulative level)

It's hardware and software -- improvement

• 1988 CMU's super computer for chess: 2400 points (human 2800 points) -- pattern recognition decision and some hand-tuned rules
• Deep Blue with more computational power, better performance, better pattern recognition

Demo: reading machine for the blind

• in hand usage machine
• camera takes picture of the book
• intelligent image processing

AI deepened our infrastructure
15 years ago, very few examples of how AI is used, but now it is deeply integrated in our economic structure.

Paradigm Shift

• Technology modeling applied to application
• Mass use of inventions
• Countdown to singularity
• Cumulate evolutionary process (technology evolution has accelerated)
• Three primary changes

• Linear is good approximation for exponential only locally
• Exponential growth doesn't grow forever

Examples

• Moore's Law -- until we won't be able to shrink any more
• Supercomputer Power
• Average Transistor Price -- doubling electronic performance
• DNA sequencing cost
• Growth in Genbank

Every form of communication technology is doubling

• Internet backbone
• Internet hosts

Miniaturization

• Nanostructures
• Decrease in size of mechanical devices
• Respirocyte (an artificial red blood cell)
• 3-D molecular computing

Exponential growth in computation power

Reverse Engineering in Brain

• The ultimate source of th templates of intelligence reach human level without any input from brain science?
• There are interesting hints - how human intelligence performs and functions

The (converging) sources of the templates of intelligence

• AI research
• Reverse engineering the brain
• Research into performance of the brain (human thought)
• Language: an ideal laboratory for studying human ability for hierarchical, symbolic, recursive thinking

How complicated the human brain is

• Trillions of interwoven neurons
• Data from brain is doubling every year -- turn data into useful models
• Simulation models of brain in the right direction
• As we get enough data we can model and simulate the brain

The cerebellum

• Gathering data from multiple studies
• Their simulation includes over 10000 simulated neurons and 300000 synapses and includes all of the principal types of cerebellum cells
• The design is a billion times simpler

Modeling systems at the right level

• The genome doesn't have that much information
• Self-organizing system with relatively simple design
• It is not that the brain is simple, but it is a level of complexity we can manage

Models often get simpler at a higher level, not more complex

E-commerce revenues: exponential growth

Contemporary examples of self-organizing systems

• Rapidly improving in pattern recognition
• Tools to reverse engineer the brain -- progress will be accelerated
• Human pattern recognition is limited to certain types of patterns
• machines can apply pattern recognition

The criticism

• From incredulity
• Exponential trends can't go on forever (rabbits in Australia)
• There are limits but they're not very limiting
• Need to verify the viability of a new paradigm
• Molecular computing is already working
• The history of AI is the opposite of human maturation
• Tell the difference between a dog and a cat...

• Genome has only 30-100 million bytes of compressed code
• Brain is very highly complex (a recursive probabilistic fractal): design is million times simpler than manifestation
• No evidence that quantum computing takes place in the tubules
• Human thinking doesn't show quantum computing capabilities
• Even if it were true, it would not be a barrier -- would just show that quantum computing is feasible

Promise versus Peril

• Would drive dangerous technologies underground
• Would deprive responsible scientists of the tools needed for defense

Links:

• http://www.kurzweilai.net/
• Slides available at http://www.kurzweilai.net/pps/AAAI/

Transcribers: Chih-yu Chao & Varun Aggarwal