AI is not a machine

AI is not a “machine.” It certainly runs on machines, but if we define “machine” in the way the Industrial Revolution framed it, can we say that what AI does is the same as what a “machine” does? The most obvious difference is that machines, in the traditional sense, solve deterministic problems, where the outputs are always exactly the same given the same inputs. This is the image most people have of “machines.” AI is fundamentally different from this. From this perspective, AI does not extend us. It is beyond “medium” as defined by Marshall McLuhan. A boss can think of his employees as extensions of himself, tools or mediums, but because each employee has some degree of agency, they resist that characterization. AI is similar in this respect; it does not simply extend us. It does something on its own that happens to align, more or less, with what we want. The difference becomes clearer when we think about the kinds of people we work with. If you hire an undergraduate, you might be able to offload tasks you would otherwise do yourself. If you hire a PhD student, you are no longer simply offloading, because she may be capable of things you are not. If you collaborate with someone at your own level, the idea of offloading no longer applies at all. Likewise, we don’t simply offload our work to AI. It does something we cannot do. At that point, it is no longer a “tool” or a “medium,” but something closer to a collaborator, with the potential to exceed us. Dyske Suematsu