Thinking Machines

There is an interesting article on the Atlantic Monthly site about Artificial Intelligence (AI) investigator Douglass Hofstadter.  The article itself is located here and is well worth a read:

http://tinyurl.com/pcfejwf

This column is relevant here for a couple of reasons.

To me, one major takeaway of the column is that Hofstadter’s research is not only qualitatively distinct from what I will call “industrial AI” conceptually, it is intellectually superior.  By which I particularly mean, it is (or will be) vastly more relevant in helping humankind understand ourselves and our world – even though it is quantitatively less productive than industrial AI – if his research is successful and understood.

However, industrial AI (let’s abbreviate it IAI) is more effective at delivering results that can be monetized than anything Hofstadter has done.  And thus, a brute force computing monster like Google can do amazing things, and make lots of money, without really addressing deep questions about learning and thinking that Hofstadter focuses on.

I think that this column indirectly says something in general about questions that are relevant to energy and energy conservation, though I may be making a stretch.  But let me take a shot.

Hofstadter is ultimately interested in figuring out how human beings think (hint, he thinks we are absolutely amazingly evolved analogy machines) whereas IAI is principally concerned with delivering accurate and reliable results.  Which, I would add, is a non-trivial and intellectually interesting field of research.  Nonetheless, Hofstadter is pursuing something more akin to fundamental natural philosophy that could lead to astounding discoveries about how and why we think.  The results-oriented application of his findings would then be layered on top of that conceptual foundation.

Let’s get a little more specific.

I can input an English sentence into Google Translate and it will convert that sentence into serviceable Spanish or French or Chinese or any of a number of foreign languages.  But Google Translate does not know what it is translating, nor does it even know that it is translating.  It is a dumb system that provides pretty smart results.

Hofstadter is looking for something different.  An AI system that would, somehow, “understand” that it was translating, and that would rely upon associations instead of brute force to come up with a workable solution.

Just so.   But the relation to energy?

As I think I have written elsewhere, there seems to be a growing trend to treat energy conservation – via energy policy – sort of like IAI.  Namely, if you throw enough resources and regulations and credentialed “experts” at it, you will get the results that you desire.

And thus, the policy maker might (and probably would) conclude that if $100,000 on energy conservation saves one million kilowatt-hours, then $200,000 will save you two million.

Alas, this is not automatically true.  But why it is not true requires a Hofstadterian investigation into where energy is actually being used, why it is being used, and whether it can be judiciously reduced.  After that, maybe you can or maybe you can’t deliver two million kilowatt-hours of electrical energy savings.  But most essentially, you will know why you can or why you cannot.  It is a deeper and intellectually more profound understanding of the situation.

This is, of course, partially an argument for more control of the framing of technical issues by competent experts, which I have broached elsewhere.

I think, though, that we can broaden this.  The sidelining of Hofstadter (he is not in the mainstream of current AI research) parallels the sidelining of deep competence in many fields of endeavor in the United States today.  Respect for deep competence has been supplanted with respect for, or deference to, gaudy credentials and smartly packaged deliverables.  And so we have semi-competent consultants and policy makers issuing poorly informed advice that is acted upon to the detriment of the enterprise.

This is not a good trend, and it is a difficult one to reverse.  Hofstadter at least shows that the qualitatively superior path can be taken.  This path will not necessarily deliver great immediate profit, but the potential long-term benefits could be enormous.  A smart enterprise, then, should treat consultants with a wary eye, as one option that does not preclude other approaches or ideas.