By Jim Harper

Your semanticist is dissatisfied today. This time it is the buzz
phrase “artificial intelligence”—or “AI.” While casting large shadow ideas
about computing power, the term “AI” does more to obscure than clarify. People
who write about innovative uses of computing should probably avoid it.

The inspiration for the present jeremiad is a recent Foreign Affairs article entitled “A Force for the Future: A High-Reward, Low-Risk Approach to AI Military Innovation.” The thesis is that the United States should embrace advanced computing in the manifold ways it will change warfighting. On that, it’s a perfectly good article. Too bad warfighting remains such a significant part of what our government does, though it did bring about a lot of computing.

The lines that offended my sensibilities were these: “China,
Germany, Israel, and the United States have all used AI to create real-time
visualizations of active battlefields. Russia has deployed AI to make deepfake
videos and spread disinformation about its invasion of Ukraine.” How is
software intelligent when it produces battlefield visualizations and
videos that convincingly impersonate a person’s appearance, speech, and
mannerisms?

Undoubtedly, there is some clever code that takes in imagery
and other sensing, identifies entities, and tracks them to produce real-time
battlefield information that is better than ever before. But I’ll be surprised
indeed if that selfsame code says something like, “Hey, there’s no moon tonight.
Their infantry doesn’t have night-vision goggles. So it’s a great time to . . .”
whatever you do in that advantageous situation.

Same for deepfakes. It’s clever code. But it’s hard to call
it intelligent.

“Intelligent” means a lot of things, but I think it’s at its apex when an agent (typically human, possibly machine) draws information and ideas from diverse fields to enlighten another field. The inventor of the balloon catheter, a tremendous stride for noninvasive heart surgery, was a scrub technician in a hospital who brought to that world ideas gained through tinkering with go-karts and tying fishing lures.

I briefly investigated AI as a concept in my 2020 report, Privacy and the Four Categories of Information Technology:

There is an open question . . . as to what AI actually is. In their textbook Artificial Intelligence: A Modern Approach, University of California, Berkeley, computer science Professor Stuart Russell and Google Director of Research Peter Norvig divide definitions of AI into four different classes. Some definitions see AI as intelligent thought processes and reasoning, while others focus on intelligent behavior, however produced. Some treat approximation of human performance as the goal, while others treat the ideal of rationality as the hallmark of true AI. In academia, at least, AI is a varied collection of milestones that—at odds with the milestone concept—are not easy to measure.

You don’t have to be an objectivist to see value in Ayn Rand’s neologism “anti-concept.” An anti-concept “is an unnecessary and rationally unusable term designed to replace and obliterate some legitimate concept. The use of anti-concepts gives the listeners a sense of approximate understanding. But in the realm of cognition, nothing is as bad as the approximate.”

In most usages, “artificial intelligence” is very much an
anti-concept. It doesn’t have borders, so it doesn’t wall in or wall out
different computing techniques. It doesn’t move readers toward understanding
the utility or consequences of a given computing application. It may be a cover
when authors don’t know enough about such applications to say what is going on.

Where computing intersects with public policy, it will often
be important to know what is under the hood. It is no defense to a wrongful arrest
or search—or an errant missile—if AI was behind it. What data went in? What did
the algorithm(s) do with it? Were these inputs properly calibrated to achieve
their desired ends and avoid undesired ones? Or were they mis-calibrated? The
latter condition indicates all kinds of consequences and remedies, from civil
rights lawsuits, excluded evidence in criminal trials, and debarment from
government contracting to perhaps even courts-martial.

Fear the blog post. Thirteen years ago I went after the prefix “cyber” in similar fashion. You see today that its use has dropped off steeply, and everyone speaks with precision rather than waving their hands at a great mass of different issues in the computer security area. Here, of course, I have shifted from semantics to facetiousness.

The post Artificial Intelligence Is an “Anti-Concept” appeared first on American Enterprise Institute – AEI.