Monday, November 5, 2012

Artificial versus Natural Intelligence

Too often we dismiss the Semantics of a problem space as a superficial element of the larger whole. However, that is often not the case at all, especially where the Semantics of describing the problem directly interfere with the ability to achieve expectations related to the problem space. Case in point is the field of "Artificial Intelligence" or AI within IT. AI was first describe as a discipline in the late 1950's and has been a part of our popular culture ever since Arthur C. Clarke wrote 2001, A Space Odyssey in 1968. While the question of intelligence in computers or robots had been dealt with previously in popular science fiction; 2001 gave us an entirely new perspective on AI - one that seemed entirely plausible and not too far off from being realized (and hopefully not the part where HAL has a nervous breakdown). HAL or Heuristically programmed Algorithmic computer, was an intelligence that had not been programmed but rather had learned in somewhat the same way humans do; this construct is also closely tied with certain aspects of the computer science discipline of AI and is referred (somewhat inaccurately) to as Machine Learning. Yet, nearly 60 years after AI was envisioned, many feel the field has yet to reach any of its original objectives.

Inside HAL - 2001 A space Odyssey
This is not to say that phenomenal advances in IT have failed to materialize, because we have experienced remarkable gains in a variety of areas. However, the nature of those advances seems somewhat different than what we have been expecting in relation to the question - "what is intelligence." Or perhaps the question more precisely needs to be; what is intelligence in relation to viable Information Technology solutions? Today in IT, we tend to use the term intelligence sparingly; we have one major field associated with analytics called Business Intelligence, but we haven't been willing to go much beyond that in claiming that what we provide exhibits anything we might consider related to the concept of intelligence.

So, let's go back and ask a fundamental question, what is it that we meant by Artificial Intelligence in the first place? Here's the wikipedia definition:

"Artificial intelligence (AI) is the intelligence of machines and the branch of computer science that aims to create it. AI textbooks define the field as "the study and design of intelligent agents"[1] where an intelligent agent is a system that perceives its environment and takes actions that maximize its chances of success.[2] John McCarthy, who coined the term in 1955,[3] defines it as "the science and engineering of making intelligent machines."

Let's contrast this with the definition for the term Intelligence by itself:

"A very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather, it reflects a broader and deeper capability for comprehending our surroundings—"catching on," "making sense" of things, or "figuring out" what to do."

Looking at these definitions, a couple of things become apparent immediately:

  1. That computer science though AI has moved towards fairly narrow definition of Intelligence.
  2. Computer Science has not to date managed to reproduce the vast majority of attributes that we tend to associate with (human or generic) Intelligence.
  3. The two terms being defined, which upon first hearing them may seem fairly similar, are not in fact closely related at all.

That last point is critical because it has everything to do with developing and or achieving expectations. As the expectations within the IT domain for AI have in fact gotten somewhat narrower than those which they began with, the expectations across the general public for what Artificial Intelligence will encompass have remained the same or even expanded. It can be summarized thus; most people when they hear the term AI expect that it will result in some sort of thinking machine and thinking cannot be defined or conceived of by most people in terms different than are applied for characterizing human thought.

Our focus over the past several decades has been directed more towards brute force computing - the ability to jam ever more computations into smaller machines at an ever-expanding rate of speed. While this is no doubt vital to the overall goals of achieving true intelligence it has not led to any serious development to what might be referred to a computational cognition. Cognition is the set of functions related to taking the lower level processing and apply real meaning to that "sensory input" without the aid of outside interpreters (e.g. human beings).

Watson won big on Jeopardy
So this brings us to our premise; are we in fact trying to do two separate things in the context of one field of practice and research when we should be splitting it in two? If we consider that there may be at least two separate types of intelligence that can be achieved in computer science, one Artificial in nature and one Natural, might we not be able to realign efforts and expectations accordingly?

This would need to start with some definitions, so we'll return to the Semantics. What if we retain the narrow view currently associated with AI (which is focused on statistical theory, agent technology etc.) as it exists now and consider that Artificial Intelligence is any technology or resulting intelligence that aids, facilitates or other empowers human cognition. In other words, Artificial Intelligence becomes a super-charged version of Decision Support. Then we could pursue a separate (albeit potentially related) field of endeavor that can be referred to as Natural Intelligence. We call it Natural because in fact what we're after is something that functions or behaves in a more human way - something specifically designed to mimic human cognition. 

This year we witnessed Watson defeat several Jeopardy champions; however it is not likely even with that impressive result that we could characterize how Watson operates as Cognition. However, what IBM is trying to do now with Watson is much more akin to Natural Intelligence than what is considered to be traditional AI - even if some of the same technology is being applied. The difference is one of goals and objectives - the expectations. IBM's expectation was the Watson had to parallel human cognition in a typically human task (closer to real time as opposed to earlier efforts with Chess) and it represents perhaps the first real step in achieving Natural Intelligence and realizing the larger set of AI goals first described in 1950's. The larger challenge now is to stop confusing these fields with one another and direct the necessary efforts towards improving the outcomes associated with each. We will discuss how that might be accomplished in future posts...

Copyright 2012  - Technovation Talks, Semantech Inc.


Post a Comment