The findings that Dr. Strogatz presents toward the end of his article, however, are where I currently begin to differ in opinion. In this latter portion, he begins to take on what seems a somewhat nihilistic view of the future. Continuing his description of the potential of programs that can teach themselves, he starts to highlight the tremendous advancements that they may make for humankind. I certainly cannot disagree. But where matters seem to diverge are where he begins to describe humans as seemingly-mere shoe-shiners in comparison to the grand titans these machines are. For instance, he points out that we would for a future learning algorithm “sit at its feet and listen intently,” and essentially have our role in science be reduced “to that of spectators, gaping in wonder and confusion.” 

The problem that I see in such a view of humans as having their “role” reduced and thereby importance essentially negated is in failing to fully recognize the mutual growth between us and the machine. In other words, Dr. Strogatz essentially writes that as of now, such algorithms “can’t articulate what they’re thinking” to us, and therefore that their “understanding” is essentially reserved from us. He provides a seeming further example of this with “the four-color map theorem,” a proof that was solved by a computer and which involved so many steps as to not be human verifiable and thus also not particularly instructive for us. The idea he presents, therefore, is this apparent one-way insight: we code these algorithms, and then they operate independently, coming to findings without ever revealing their process of growth to us. The result, if following this route, is that we are essentially victim to this single pathway of intellectual comprehension from humanity to its gadgetry. But it is here, in this key line of thinking, that I think more can be said about what actually seems to be a bidirectional conduit.

In other words, with the growth of these algorithms, their “insight,” as Dr. Strogatz describes, seems to be more translatable than what he may merit. A key example in chess, for instance, is with the play of AlphaZero. Dr. Strogatz presents humans as seemingly-inactive figures that can only watch such growth from afar; yet grandmasters have already begun to use AlphaZero as sources of growth for their own play. Thus, though the article’s discussion of our biological computing power being far reduced from that of an algorithm is true, it does not mean that we too cannot learn from these machines. They are not isolated from us. Instead, rather than being taken out of the game, such algorithms instead reinvigorate our role as the players and makers of our own future by providing us new knowledge with which we may experiment. In this way, insight is indeed thereby gained by ourselves implementing what are only new lines of thought, from which we may then find reason. In particular, with the example of top-level chess players implementing certain tactics of AlphaZero’s play, we may “try out” the techniques of what such a computer has found, and then ourselves see why these tactics and tricks work. Multiple notes have been made, for instance, about AlphaZero’s dominance of the space of a chess board, and how it renavigates pieces in at first seemingly non-intuitive ways so that the timing of a turn finds it favorably. (For example, amongst other notable figures in the chess community, agadmator, a chess YouTuber mentioned elsewhere on this website, includes videos demonstrating such prowess from AlphaZero here and here, and has related its influence on top-ranked players.) These are findings that become apparent through reverse-learning from the solution itself. 

These patterns of what learning algorithms can achieve are therefore not secluded from us, but rather the results themselves of repeated experimentation, trial and error. Though a computer can of course run these trials far more quickly than can the human brain, it is thus but a matter of time before humans “catch up” for any given solution and gain the understanding of why its finding has arisen. As a result, machine learning is not a one-way street: instead, it gives us the results of rapid evolution that we can then verify and implement through similar tests and learning of our own, and which in that way, allow us to ourselves attain greater insight far more quickly than if left to our own devices. So sure, an algorithm like AlphaZero may not tell us why it is has arrived at the conclusion that it has, or even why it may be thinking what it is at any given point in time; but the results it gives are ones we can then learn from rather than merely “spectate” on; and in doing so, develop our own knowledge independently as well. Machine learning, then, visible with AlphaZero and otherwise, is therefore not some mystical “oracle” with powers beyond our grasp, but rather a teacher of our own construction and for our own advancement. And like any teacher, the knowledge it brings is not simply disseminated from a higher existence, but rather built upon by both the student and instructor to enhance the experience of education. In this way, the students are not secluded from intellectual growth, but rather provided a way to come to their own findings, and then in their own right, enhance that of the teacher’s, in a process that thereby becomes one of mutual understanding and knowledge gained.