Edit from the future (2025): This post’s points are now outdated according to my current thoughts, given technological developments since.
Artificial intelligence, or AI, is a loaded term. When it was first used, it would refer to the abilities of computers to play perfect-information games like chess. But now that is “old tech,” and we don’t consider that as much a part of AI. In fact, I contend that AI has always been the term used to refer to the bleeding edge of computer science at any given moment. It’s why we might never “solve” or “figure out” AI, since once we figure out AI today, the term AI will just be redefined tomorrow to be the new horizon to reach for.
Nevertheless, AI has taken on a distinct character recently that might prompt us to consider a closer definition of AI in terms of what it means today. This is the question we will investigate now.
In general, AI has been the marker of comparing the abilities of computers to those of humans. One by one, many games like chess fell to the power of computers which could beat the top human players in the world. But I wouldn’t consider chess to be AI today.
One distinction we can make between chess and the tasks that are thought of as AI today, is that chess is formally specified: the rules can be encoded in a logically precise manner. On the other hand, modern subjects like natural language processing and computer vision are fundamentally informally specified.
Nevertheless, informal specification has always been a part of technological development, even when we haven’t referred to it as AI. In fact, we can argue that informality has been a major part of many products across tech. To see this, note that if product development were completely formal, then once the original feature set was implemented correctly, the code would never need to change and the product would never need to evolve. But this is often not the case in practice: codebases and product features change constantly, based on human innovation and shifting consumer expectations, both of which are informal.
(Of course, some computational tasks are still formal – for example, the problem of sorting a list. But the point still stands that many tech products outside of what we consider AI have fundamentally informal specifications.)
Thus, this distinction alone can’t define AI. It therefore seems unclear to me how to define the subject today, beyond a listing of the fields that are now considered to be under the umbrella of AI – natural language processing, computer vision, and so on.
