There is this problem with AI: people who think deeply about it start losing their minds. In my opinion, people fall into two categories: those who use AI but don’t particularly think about what AI is in essence (or who don’t have to explain AI to students), and those who do think about it and freak out. I suspect that people in the second category are at risk of losing their sense of the boundaries of reality, as happened during the era of great discoveries in physics at the turn of the last century and the century before last.
At one time, Lenin’s (I’m not saying “Marxist” because I don’t want to drag Karl into this) answer to this question was heavily hyped in the Russian-speaking world (among people in the second category, that is, the “ones who think about it”): philosophy for proletarians, the book Materialism and Empirio-Criticism.
What do we have as of today?
I’m not a philosopher. But from the (little) I’ve read by Russian-speaking philosophers, I’ve concluded that Lenin’s philosophy is today being critically re-examined through the lens of “decommunization,” while, on the other hand, this philosophy is so firmly embedded in the heads of some philosophers who come from the Soviet era that present-day reasoning inevitably gets packaged into Lenin’s conceptual framework.
Again, I’m not a philosopher and I’m an idiot in general, so maybe I’ve misunderstood everything. So this is not financial advice, but I suspect that the need has arisen once again to defend realism against attacks by the “thinkers” (those in the second category of people when it comes to AI). Lenin is both helpless here (he died 100 years ago, although his body is still lying on public display in downtown Moscow, and therefore he cannot comment on events taking place today) and toxic, because the very subject of Soviet “communism” is toxic.
So we need something of our own, preferably something Western (to expand influence). Among Western thinkers, so far I’ve only come across Roy Bhaskar (also deceased by now) and the little-known English philosopher and sociologist John Michael Roberts (still alive). But they are outside the AI context.
Recently, I took a closer look at the works of the contemporary American philosopher Moti Mizrahi, who has indeed written some notable things about AI. Besides AI, he also has a book about relative realism. All in all, this is, rather, precisely a counterexample of thinking from the second category.
What is to be done, then? Evidence synthesis is no longer what we can concern ourselves with here. Looking at things globally, I don’t think I’d be wrong to say that evidence synthesis has lost its strategic distinctiveness as a narrow discipline and can now, with every justification, be reconceptualized as a branch of applied AI. A branch that is undoubtedly still very poorly developed, but strategically, that is how I see it. I remember a documentary from a few years ago about Industrial Light & Magic, and there was this memorable shot where they cleared out the entire puppet studio and filled the space with computers. I foresee exactly the same thing happening in evidence synthesis.
But how do you remain a realist if you think about AI? I think all of this deserves a more detailed examination, using concrete examples of what it means for the field: how to trust searches performed by AI, how to trust AI-assisted classification according to inclusion and exclusion criteria and risk of bias, how to trust tables containing data extracted from PDFs using AI. How do these processes, when carried out with AI, differ from their familiar human counterparts?
Today, the literature offers numerous answers to these questions. The number of publications on the automation of evidence synthesis has increased severalfold, if not by an order of magnitude. Unfortunately, the vast majority of these answers do not satisfy me, because at best they are fragmentary and fail to provide a coherent picture of the world.
I think that at any level of empirical knowledge, one should strive to bring that knowledge together into a theory – perhaps an incorrect one. But I suspect that the very absence of the aspiration to synthesize at a high level – that is, without following any particular methodology, but essentially in the form of a traditional narrative review – is caused by certain limitations of the postpositivist orientation of clinical epidemiology and, consequently, of evidence synthesis.
The aforementioned Roy Bhaskar (together with his coauthors) proposed solving this problem through the general theory of interdisciplinarity that they themselves developed. This theory is built upon the principles of Bhaskar’s dialectical critical realism, particularly the idea of “seriousness,” as he calls it. According to Bhaskar, it is unserious to deny the obvious – for example, that leaving a house through the door and leaving through a second-floor window are, after all, different things, and no philosophy is needed to understand this. Here I return to the idea of “proletarian philosophy,” that is, the idea that a proletarian (or a precarian – eh?) needs philosophy (if they need it at all – ha!) in order to strengthen (rather than nullify) “naïve” everyday realism.
In the context of evidence synthesis, the idea of everyday realism is oh so relevant. So let’s return to the thought about traditional narrative reviews. To be fair, they haven’t been completely excluded from the taxonomy of evidence syntheses, but, on the other hand, they “flipped” it on the fly, turning traditional narrative reviews from a phenomenon in their own right into a variety of evidence syntheses, that is, the people who devised the taxonomy quietly incorporated them into their own conceptual framework. This is roughly like how I above quietly incorporated the entirety of evidence synthesis into a “branch of applied AI”! This is a perfectly routine carve-up of intellectual spheres of influence – not that it is entirely without objective grounds, but it undoubtedly has political significance as well. (I’m reasoning like a Marxist.)
So here’s what happens to these narrative reviews as a result of this intellectual resubordination: what was a method of universal applicability suddenly turns out to be a narrow method in the arsenal, assigned only a particular place (in the gutter).
Thus, it turns out that solving any question – if viewed through the taxonomy of evidence synthesis – should begin, for example, with a visit to the Right Review website. However, such an approach can hardly be considered necessarily universal – even from the standpoint of wide-angle interdisciplinarity (for example, as Bhaskar and his coauthors depict it).
What I’m trying to say in this convoluted and long-winded morning text is that wherever I poke around, wherever I go in evidence synthesis (whether it’s its authoritative taxonomy or preprints on automation using AI), everywhere, in the most astonishing way, I run up against the very same wall that Lenin described among the Russian “Machists” and that Bhaskar and his coauthors identify in their “general theory of interdisciplinarity”: a wall made of fear of transcendent ontology, a wall made of the refusal to declare unconditionally that the world as a whole really is as we see it – that is, big, cold, insanely diverse, and by no means anthropocentric.
With that, I conclude my thought, because my phone is running out of battery. Thank you for your attention.
Written by Pavel Zhelnov on October 16, 2026. Originally published as two posts in t.me/evidence_synthesis (in Russian): one, two. Translated using ChatGPT 6 High on the same day, revised manually by the author. Last revised October 16, 2026.