10 "ahas" in Artificial Intelligence

The discussion on Artificial Intelligence seems everywhere.  I thought I would spend the summer reading up on it.  I was delighted to find several books, articles, movies, TV shows and videos that were accessible to the lay person.

I have written about my own concerns regarding A.I. in the past that have generated good discussions.  

This time, rather than express my views on the topic, I thought I would identify 10 "ahas" that jumped out at me as I did my summer reading.  (I am aware, if this were a class paper, an A.I. professor might find this not technically proficient enough .... !)

1. Who coined the term?

The label Artificial Intelligence was coined in 1956 at a summer workshop at Dartmouth college organized by computer and cognitive scientist Jack McCarthy and was attended by eleven mathematicians and scientists.

The idea:  Sometimes a label sticks and is an important moment in the development of a field

2. General Intelligence vs Specialized Intelligence

Specialized intelligence is designing computers for specific tasks, such as reading an X-Ray.  General intelligence is to make sense of the operating environment, and make overarching decisions.  A.I. has been at a gallop in developing specialized intelligence applications, and in the process disrupting jobs being done in those areas by humans.  And as these specialized applications are expanding in scope, they are threatening both blue collar and white collar jobs.

But general intelligence has been much more hard to crack.  Here we enter the realm of computers being able to feel emotions.

I once had the chance to ask Ray Kurzweil in person if we would ever achieve general intelligence.  His answer was rather disturbing.  He said, "it will be close enough."

The idea:  At what point will specialized intelligence become "close enough" to general intelligence?!

3. A.I. Winter

As an analogy to nuclear winter, the term A.I. winter is used to refer to periods when enthusiasm for A.I. wanes due to lack of progress and funding dries up.  As with any technology, A.I. is also subject to the hype cycle. 

The idea:  Yet, it seems to pop back up after each A.I. winter.

4. Intelligence explosion

As computers become better and better at replicating human cognitive tasks, and as processing power continues accelerating apace, and more and more of them get connected, A.I. thinkers are forecasting an intelligence explosion.  Similar to the Cambrian Explosion that caused life to take off on earth.

The idea:  Intelligence explosion would put control of what happens next out of human reach

5. Ants

A chilling example of how A.I. decision making can go wrong is given by Max Tegmark through the example of ants.  Let us say we care about ants.  And we are walking on the street and spot ants.  We would be potentially willing to step aside and make an adjustment to our path to avoid killing some of those ants.

Now let us assume, we are building a dam and are about to flood a plain.  And that plain has thousands upon thousands of ant hills with hundreds of millons of ants.

Would the same person, walking to work, avoiding killing the ants in their way, stop making the decision to build the dam, to avoid killing those millions of ants?

The idea:  Regardless of how much controls we put in on A.I. decision making, we could find ourselves in the position of ants in the above scenario.

  

6. Singleton

Singleton in A.I. terms, is a world order where there is a single decision making agency at the highest level.  Singleton is both a mathematical and a programming concept.  However, A.I. thinkers, such as Nick Bostrom, have begun using that term to talk about algorithms powered by A.I. that can supersede any decision making that we would be familiar with.  

An example would be Amazon through Alexa beginning to suggest and direct all our purchases.  We can imagine extending that idea to other aspects of life.

The idea:  Humans will eventually give up decision making to an or a set of algorithms.

7. Energy efficiency of the brain

Our brain uses 20 watts of energy for a given task, compared to 200,000 watts by a computer for a task of comparable complexity, according to researchers at Rice University. That is, the brain is 10,000 times more efficient than today's already astonishing computers.

The reason is because the brain is willing to accept approximate answers and navigate the environment, whereas computers generate the most accurate answer.  So the giant jump in energy is due to the fact that the exact answer is needed for computers to operate.  If we can imagine a heavy robot on an assembly line that is assembling cars - it better be absolutely precise!

However, humans can accept 2+2 = 5 to continue operating in the physical world.

The idea:  Scientists are working on how to make computers more energy efficient by being able to handle approximations.

8. The cortical column

While programs are being written to emulate intelligence, some researchers are trying to emulate the brain itself.  One of the physical models of the brain is that it is composed of cortical columns.  These columns repeat and are a fundamental unit of intelligence operation.  Thus, one could replicate that through engineering.  

The idea:  Don't just replicate the functioning of the brain, replicate the brain

9. Memory and consciousness

I never actually thought about the difference between memory and consciousness and if one is the precondition for the other.  But A.I. experts seem to be spending a fair bit of effort on this topic.  Do you have to remember to be conscious?  Do you need to be conscious to remember?  Can you be conscious if you mis-remember?

The idea:  You can have tons of memory storage, but for intelligence, one needs to replicate the function of cognition.

10. Prediction Machines

Economists have reduced the functional utility of A.I. to one useful construct.  The ability to predict.  What they mean by this is the ability to predict allows a machine to operate efficiently in its environment like a human would.  Driverless cars are good examples of this.  A driverless car can predict how the traffic is going to behave and can thus drive safely and efficiently like or better than a human.

The implication of this is that the cost of prediction will dramatically reduce and change our social structures.  An example of this is the light bulb.  If each light bulb is prohibitively expensive, we would use it in one way.  But if it is down to nearly zero and is ubiquitous, we build whole cities with skyscrapers and live a different way.

The idea:  When the cost of prediction nears zero (like light bulbs), our societies will be unimaginably transformed including every business sector as we know it.


Summary

Of the nearly fifty ideas that popped out during my summer reading, these ten generated the most "ahas," at least for me.

All in all, what surprised me was that uniformly, writers and thinkers in the A.I. field tend to be more concerned about the coming wave of A.I. and our ability to properly handle it.  And those that do not appear to be worried, seem to be those working on specific applications and not thinking about the general problem.

I found this a worthwhile summer topic this year!
  

My earlier posts on A.I. :

Alexa ..... you firing me?

Artificial Love

Artificial Incoming

Inverse Turing Test

Timeless concepts in Westworld TV show

Second half of the chessboard

Virtual Reality at Stanford University



References:

Books

How to create a mind:  Ray Kurzweil - always provocative

Super Intelligence:  Nick Bostrom - uniformly gloomy

Life 3.0:  Max Tegmark - imaginative

Machines that think - Everything you need to know about A.I. - my book of the summer

Prediction Machines - Ajay Agrawal - Distilled essence


Movies

Ex-Machina - best of the lot

The Imitation Game - heart breaker about Alan Turing

Interstellar - big budget Hollywood

Her - intriguing

Short Circuit - funny

I, Robot - weak

RoboCop - kitschy

Blade Runner - did not work for me


TV Shows

West World - stunning

Humans - also ran