This time it is different. At some point every pattern breaks, or at least changes.
The black box aspect is a direct result of AI being a learning technology. A higher order technology. It learns to do things we have not explicitly taught it, or might not even know how to do ourselves.
We will eventually find better ways to analyze and interpret how models work. Why a model produced a specific answer to a given prompt.
But the models will keep getting more powerful too. Today they help coders over the speed bumps of unfamiliar language syntax, or esoteric library conventions. In a few years they will be actively helping researchers with basic problems.
I.e. the hurdles for getting to the front of AI as a field as a technology contributor, and the resources needed, are going to get steeper. Of course there will be many people who do, but it won't be in the same way that most technically savvy people have had many years to learn programming languages, apply them, and even contribute to them, without language's progressing out from under them. (Except for C++ of course! /h)
EDIT: It is worth distinguishing between AI like GPT, as in very flexible and powerful models that will be used across disciplines by all kinds of people regardless of technical chops, vs. the overlapping AI algorithms (partly subset, partly different) applied to smaller learning tasks, which has been a commonplace tool for many years and will continue to have its place.
The black box aspect is a direct result of AI being a learning technology. A higher order technology. It learns to do things we have not explicitly taught it, or might not even know how to do ourselves.
We will eventually find better ways to analyze and interpret how models work. Why a model produced a specific answer to a given prompt.
But the models will keep getting more powerful too. Today they help coders over the speed bumps of unfamiliar language syntax, or esoteric library conventions. In a few years they will be actively helping researchers with basic problems.
I.e. the hurdles for getting to the front of AI as a field as a technology contributor, and the resources needed, are going to get steeper. Of course there will be many people who do, but it won't be in the same way that most technically savvy people have had many years to learn programming languages, apply them, and even contribute to them, without language's progressing out from under them. (Except for C++ of course! /h)
EDIT: It is worth distinguishing between AI like GPT, as in very flexible and powerful models that will be used across disciplines by all kinds of people regardless of technical chops, vs. the overlapping AI algorithms (partly subset, partly different) applied to smaller learning tasks, which has been a commonplace tool for many years and will continue to have its place.