footnote, mathjax

Wednesday, September 23, 2020

This is a response to a several things, but, first of all, to a project of Andrea Sangiacomo’s, as reported in this article.[1]The article, which is clearly based on an interview with Sangiacomo, is available in both English and Dutch, but there is no indication I can see as to the relationship between the two versions nor as to the language(s) used in the original interview. I will mostly rely on the English version, for obvious reasons. As usual, I am ill-informed: I don’t know anything about Dr. Sangiacomo’s work besides what the article contains, and, given that such sources are not always very reliable, I can scarcely be said to know even that. I don’t know, for example, whether Sangiacomo has been accurately quoted, or whether his words, even when they are his, may have been taken out of context. So my remarks here should not be understood as criticism of Sangiacomo’s project, still less of him personally.

If not that, though, then what are they? Sangiacomo is reported to have said something to the point:

Many colleagues still consider philosophy to be purely a job for humans. And do not get me wrong, I generally agree with that. The most relevant philosophical insights are still being generated by people who reflect deeply.

The difference between human and non-human is not, as such, of much concern to me in this regard: if horses, oxen, or electronic computers can aid in the generation of relevant philosophical insights, then I am more than happy to share the glory of such with them. What does concern me is the difference between reflective and unreflective. To paraphrase Locke: I grant, AI’s will not come to deep reflection till they be more like humans; and I add, nor then, neither. Natural intelligence is fully capable of executing an algorithm.

I introduce this word, “algorithm” (which does occur, but not very prominently, in the article under consideration) with some hesitation, and not only because, as Popper somewhere notes, the word should really be “algorism.” What I want is a word for the thing non-human AI’s can definitely do, according to Sangiacomo, because to do that thing — to implement or execute what I am calling an algorithm — does not require deep reflection. “Algorithm” may or may not be the best choice for this, and I certainly don’t intend to import any of its technical content from theoretical or applied computer science. In what follows, it should taken to mean: a process that can be carried on without reflection. This I think an informative definition because, despite having read and been thoroughly confused by a whole canonfull of philosophers who all use the term “reflection” in various ways, I still find the definiens clearer than the definiendum.

With only slight exaggeration, then, I would say: the present post expresses a reflection prompted by the output of a certain algorithm. One stage of the algorithm in question, namely the part that made Martin Lenz’s link to the article come up on my Facebook feed, was implemented by electronic computers. But, before that, I would guess it was a human who decided that the University of Gronigen needed a story about this under “About us \(>\) Latest news \(>\) News \(>\) News articles,” and I know it was a human, by the name of Jorn Lelong, who actually wrote the story.[2]I find with some Googling that Jorn Lelong is a freelance journalist based in Ghent, Belgium, who has authored various pieces (in English) about developments in technology research for various different outlets, and who also has a Twitter account (in Dutch). There is even a picture of him, so, barring deepfakery, he is definitely an actual human being — a kind of cool, cheerful looking young human being in a bright red hoody and dark sunglasses. There is no credit for a translator. Perhaps this means that Lelong wrote both the Dutch and the English versions. But I mean no disrespect in saying that I doubt the humans in question required any deep reflection to perform these tasks, and in fact I even imagine that the best current AI might be trained up to do them tolerably well. Moreover, the main factor in the decision to publish the story is presumably what gets mentioned at the very end, namely that Sangiacomo has received an ERC grant of 1.5 million for a follow-up research project. How much deep reflection is involved in awarding such grants? I hope Martin will not be offended if I add that, whatever reflection he may have engaged in upon reading the article, it was likely not an essential part of his decision to post the link. And, as for my own decision to follow that link: what chance has reflection in the face of clickbait? In conclusion, then, Sangiacomo (as reported) and I (as self-reported) agree thus far: we both hold that philosophy requires reflection, and yet we both concede that the output of an unreflective, algorithmic process can present a useful occasion for such reflection.

The question, however, is whether some such algorithms are superior to others. Sangiacomo, in conjunction with the Data Science team at Gronigen’s Centre for Information Technology, has come up with a new algorithm to do this, and plans to spend 1.5 million following up on this work, because he is unsatisfied with the results of a different algorithm, namely, the one whose output he characterizes as “a few great works by a few great authors.”[3]At least: these words are in quotation marks in the article, which I think is supposed to imply that they are due to Sangiacomo, rather than to Lelong. This algorithm, too, has various stages which have been implemented on different hardware. The first and longest part was carried out by what you might call a neural network — anyway, it was a network of some kind, as Sangiacomo reportedly asserts. In a section of the article titled “The importance of a network,” Lelong writes:

Sangiacomo also found that social connections also have a significant role to play in the popularity of certain philosophical theories. As such, in addition to the shifts in meanings, he also sought to trace the mutual relationships and networks of early modern philosophers.

Working over the course of some centuries, this network has brought a few great authors to prominence. The second and shorter part is carried out by the current institutions of philosophy, as part of the process of professionalization: students, professors, journal editors, etc., are trained up to recognize these few great authors and to produce and respond selectively to the work of philosophical scholars which is confined, in a certain way, to a few great works by them.

That the first, long part of this process was algorithmic in the sense in which I am using that term, which it to say, not properly reflective, is Sangiacomo’s point when he (reportedly) goes on to say:

A priori, there are no obvious reasons as to why some philosophical theories become much more popular than others. Throughout history you sometimes see that really crazy ideas become mainstream, while good ideas are side-lined. To explain that, you have to look at social connections, because people are the real driving force of history. In that respect, this project provides a lot of information.[4]Again, these words are in quotation marks in the article.

Now, the discovery that some ideas are “really crazy” (Dutch: erg gekke) even though they have “become mainstream” (Dutch: mainstream worden), is, in other words, enlightenment. As such, it requires, one would think, significant philosophical insight, and must therefore depend on deep reflection. This calls up the image of an alternate version of history: a history in which the owl of Minerva gets up early in the morning and flaps around all day, pecking out the really crazy ideas and making sure that the good ones become mainstream. Sangiacomo, however, regardless of what he may or may not imagine, nevertheless does not propose either to apply reflection now, in order retrospectively to put such a history together, nor to try and institute the rule of reflection from now on. Instead, he offers a new algorithm.

But before considering Sangiacomo’s new algorithm: I have yet to describe the second, shorter stage of the old one. This stage, as it happens, is well describe by Martin Lenz himself in his own recent blog post:

People become enthusiastic if they recognise something. . . .  I think much the same is true of our talk of “great thinkers.” We applaud recognised patterns. But only applauding the right kinds of patterns and thinkers secures our belonging to the ingroup. . . .

This is why calling someone a “great thinker” is to a large extent self-congratulatory. It signals and reinforces canonical status. What’s important is that this works in three directions: it affirms that status of the figure, it affirms it for me, and it signals this affirmation to others. Thus, it signals where I (want to) belong and demonstrates which nuances of style and content are of the right sort.

The algorithm at this stage has a familiar purpose, one of the main things both artificial and natural intelligences are trained to do. It is an algorithm for recognizing patterns, and, more particularly, for recognizing places in a pattern: figures that stand out against a ground, enclosing what is inner and excluding what is outer. Some of the places, or topics, it is trained to recognize are great thinkers of the past, whereas others (metaphysics, philosophy of mind, etc.) are contemporary, but, one way or another, knowing how to recognize such topics; how to market oneself as belonging to one of them (or to the “intersection” of two of them); how, on the basis of this topic assignment, to enter into a mutually beneficial group of researchers who own that topic, in which everyone is bound to cite each other’s works; how (if one is a journal editor) to determine the topic of a submission and ensure that it is refereed by members of the group who own that topic — all this, and more of the like, is what professionalization in academic philosophy currently entails.

Or — and this will make a difference to our understanding of what Sangiacomo proposes to do — is that description already somewhat outdated? In using the term “topic” I allude, on the one hand, to Kant and, via Kant, to Aristotle. More on that below. But I also, on the other hand, allude to Brian Weatherson’s fascinating History of Philosophy Journals: Volume 1: Evidence from Topic Modeling, 1876–2013. Unlike Kant and Aristotle, Weatherson cannot be held responsible for the word: “topic modeling” is just the usual terminology for the type of algorithmic technique he uses. But still, it is interesting to inquire into what “topic” ends up meaning for him, i.e. what “topics” the algorithm comes up with when applied to the contents of philosophy journals during the indicated period. It is particularly interesting in the present connection because, to quote the Wikipedia article just linked to:

HLTA [a method of topic modeling][5]HTLA stands for hierarchical latent tree analysis. Weatherson uses a different method, LDA (latent Dirichlet allocation). For some technical details and, should you be interested, links to more, see the Wikipedia article in question and Weatherson’s chapter on methodology. was applied to a collection of recent research papers published at major AI and Machine Learning venues. The resulting model is called The AI Tree. The resulting topics are used to index the papers at to help researchers track research trends and identify papers to read, and help conference organizers and journal editors identify reviewers for submissions.

So, in other words: the AI researchers themselves feel that their algorithm, when implemented by artificial intelligence, can fulfill at least part of the function I assigned to professionalized human philosophers above — or rather, it can fulfill at least part of the corresponding function in that field. So an interesting question would be: will Weatherson’s related technique succeed in finding the corresponding kind of “topic” in philosophy, such that it, too, could be used for those purposes? The interesting answer, as I understand Weatherson’s results, is that his method can succeed in this respect, but only during a certain, relatively short period of the history of philosophy (and of the history of philosophy journals).

A perfect philosophical work, at whatever time and in whatever way written and/or published, ought to have a topic, a place, in a formal (logical) sense. That is: it ought to have some leading concept.[6]You might think the condition should instead be that the work have a leading judgment (a thesis). For Kant, however, such a unity of thesis depends on and presupposes a conceptual unity: “A judgment is the representation of the unity of the consciousness of different representations or the representation of their relation insofar as they make up a concept” (Jäsche Logic, §17, Ak. 9:101,5–7). For, as Kant says in the Amphiboly of the Pure Concepts of Reflection:

One can call any concept, any title under which many cognitions belong, a logical place [Ort]. Upon this is founded [gründet sich] the logical topics of Aristotle. (KrV A268/B324)[7]Kant, reasonably enough, treats τόποςtopos as the equivalent of Ort, and reserves Topik for a theory or system of such places (parallel to other terms such as Logik and Physik). But I will stick, somewhat reluctantly, to the more common practice of calling each individual logical place a topic. (Hume, following a different interpretation of the Topics, uses “topic” in yet another sense.)

The (logical) perfection of such a work would consist in the consequence of all its parts from, and their joint adequacy to, that single leading concept:

In every cognition of an object there is namely unity of the concept, which one can call qualitative unity, insofar as only the unity of the comprehension of what is manifold in the cognitions is thought thereunder, as, for example, the unity of theme in a play, an oration, a fable. Secondly, truth with regard to consequences. . . .  This one can call the qualitative plurality of the notes that belong to a concept as their common ground. . . . Finally, perfection, which consists in this, that this plurality, [taken] together, reduces, conversely, to the unity of the concept, and completely agrees to this [concept] and to no other, which one can call qualitative completeness (totality). (B114)

I hasten to add (in fact, this is the whole point of §12 of the B edition) that these three qualitative moments of quantity are necessary but not sufficient for reference, i.e. for “cognition of an object”; in other words, even a work that is logically perfect in this sense might lack a subject matter (what might be called an objective or material topic, although Kant never uses such expressions, to my knowledge). And then again, on the other hand, a work might be full of objective content and yet lack formal–logical perfection. This latter condition, roughly speaking, is what the methods of “topic modeling” assume: namely, that a given article will contain a mixture (in some determinate proportion) of the consequences (notes, characteristics) of various different logical topics. Words are used as proxies for such consequences, and a statistically recurring cluster of words is taken as the sign of an underlying conceptual unity from which all (and, ideally, only) the members of that cluster flow.

Even assuming, however, that all or most of the articles published in philosophy journals have at least such imperfect logical locality, it doesn’t follow that the topics located by topic modeling will be topics of the aforementioned kind, namely the topics that guide the current algorithm of professionalized philosophy. For one thing: if a topic is to be usable in that way, it must be, not only a logical place, but also, so to speak, a social or political one. It is not enough for it to comprehend, in a conceptual unity, what is manifold in certain cognitions; it must also comprehend, in a political unity, what is manifold in certain people. This is explicit enough in one part of the function of such topics: if the editor is to decide, based on the topic of a paper, who would be appropriate reviewers, or rather, to which reviewers that paper is appropriate, then there must be an identifiable group of people who have a right — in a broad sense, a property right — to the topic in question. This is the external deference to or recognition of the political structure which is owed by the journal editor; internally, the main political tie, is, as I have mentioned above, the implicit covenant to cite one another’s papers.[8]It came out recently, in certain infamous cases, that the editors of some journals had not paid this political deference to the owners of certain topics, e.g. trans studies, and, as a consequence, had accepted articles for publication that failed to cite the literature. No doubt these articles were also flawed in their content. I highly doubt they were based on deep reflection or contained many relevant philosophical insights. But people were quite right, in my view, to complain, not only about problems in content, but about the merely procedural errors in the acceptance process. This lack of deferral to experts was itself, indeed, an unjust act, given the way other topics are currently treated. Following the usual terminology, I will call such combined logico-political topics specialties. For topic modeling to pick up topics of the right kind, at a minimum, the constitution of professional philosophy — the constitution under which the body of writing to be analyzed was produced — must involve division into specialties.

In the absence of such a constitution, the method may well detect some topics, but these topics will not be specialties. It might still pick up the logical topics of individual works, even though, since these topics are not specialties, the information will not be usable by journal editors etc. in the contemplated ways. On the other hand, however, it might detect political topics directly, even though these are not specialties, i.e., not correlated with logical topics. A group of people who selectively read and mention each other, and selectively control one another’s ability to publish and otherwise flourish, will naturally develop some linguistic markers of their own. They would do so, I’m certain, even under an imagined constitution in which political topics were formed arbitrarily, say by lot, which means that topic modeling could easily discover even such purely political topics. Philosophy has never been so constituted, and I hope it never will be.[9]Some systems of academic patronage can contain elements of such a constitution, however. This may be what Sangiacomo has in mind when he (reportedly) talks about “the network of a well-known male author,” and also when he (reportedly) says that “people are the real driving force of history.” It’s not what you know but who you know, as they say. But philosophy has very often, in fact almost always, had another constitution that might produce, so to speak, an even stronger political signal. I refer to the system of philosophical schools, also known as sects: αἱρέσειςhaireseis. In this system, a few great authors — authors in the original sense, authorities — are treated, not as logical topics, conceptual themes that a philosophical work might be about, but rather as heresiarchs, mythical or historical originators and successive rulers of political topics that preserve themselves through time.

Maybe it seems that there is little to choose between our constitution of specialties and this older one of sects. Isn’t the latter just an alternate way of dividing up the same territory? But actually everything is different, because no sect ever owns its authorities. If anything, it is the other way around, although sovereignty should not be confused with property. In any case: this relationship, however it should be described, leaves the authorities still available to anyone else. The ancient Peripatetics and Stoics could not prevent, and had no right to prevent, the Neoplatonists from using Aristotle or Epictetus for their own purposes. The German Idealists might have many things against Schopenhauer, or the Marburgers against the Phenomenologists, and vice versa, but no one could coherently complain: Kant is ours. If Natorp gave a talk, and Husserl raised his hand during the Q&A, he would not have to begin his question with an apology: “This is not my field, but …”. The authorities were no one’s field. They remained a commons.[10]This is not to say that there can’t be problems about property rights to authorities. Problems will come up when we want to use an authority, e.g. Confucius, who belongs to an entirely different tradition, a tradition that has its own institutions of teaching and writing and need not necessarily welcome being appropriated by the Western academy (descended from Plato’s Academy). In general, a commons belongs to the village, not to the world.

And, sure enough, Weatherson does pick up at least two sects as “topics”: post-Kantian idealism (dominant in the earlier years of his journals) and ordinary language philosophy (peaking sometime in the 1950’s). Weatherson is well aware that these are not “topics” in the same sense as most of his others: he does call idealism a “school,” and he says of ordinary language philosophy that “what the model is finding is a style as much as a content,” although I think “school” (or “sect”) would be more appropriate in that case, as well. He explains here that, while the particular model that forms the basis of his book did not detect another sect, pragmatism, this sect did appear as a topic in many other runs. Since the reign of Analytic philosophy began with the capture of most of these journals by a single sect,[11]As discussed by Weatherson, ibid., referring in turn to an article by Joel Katzav and Krist Vaesen. most of the well-defined sects in the postwar period do not appear in his model, or at best have a very small signal.[12]Clear examples of this would be Straussians and Cavellians, both of whom tend to publish elsewhere. Continental philosophy, while it could be regarded from the outside as one big sect, also continued to have a stronger sect-based constitution internally during much of this period, but, again, their publication mostly did not appear in these journals. The nature and evolution of Continental philosophy’s internal constitution are beyond my scope here, and in part beyond my knowledge (though I do have some ideas).

But — and this needs to be emphasized because, even though the main events happened within living memory, they have already become hard to imagine, let alone recall — this does not mean that his method detects mostly specialties from, say, the late 1940’s on. On the contrary: no specialties yet existed, or at most very few (ancient philosophy might be an example). What we had instead was, to begin with, a single sect, Analytic philosophy, which then began the process of breaking into subsects. Ordinary language philosophy was an early and ultimately abortive example, but it is not hard to imagine an alternative future for our mid-to-late 20th century past in which Anglophone philosophy would have been divided up among followers of such figures as Popper, David Lewis, Dummett, Putnam, Rorty, etc., with none of those sects coming to own any logical topic. In other words: we might have Popperian logic, Lewisian logic, etc.; and likewise for metaphysics, political philosophy, philosophy of science, philosophy of language, philosophy of mind — each of these figures felt free to write about any of those logical topics, and so might all of their followers have felt free. In fact, Lewis, writing around 1988–9,[13]“Academic appointments: Why ignore the advantage of being right?,” first published in Ormond Papers (Ormond College, University of Melbourne, 1989), based on a lecture given at Ormond College in July 1988; now available in Papers in Ethics and Social Philosophy (Cambridge University Press, 2000, ISBN 978-0521587860), pp. 187–200. I have discussed this article, which abounds in relevant philosophical insights, at greater length elsewhere. seems to confidently assume that such a future is in store: the implicit bargaining between the members of his lucky philosophy department concerns which school[14]Lewis in this article conceives schools doxastically, as groups of people who happen to agree about certain things, rather than politically, as I do here. I take this as a piece of characteristically Lewisian irony. will get a new hire, not, as generally happens now, which AOS will get it. Allowing for Lewis’s perceptions of the profession to be a little out of date,[15]If that is really the explanation. He had perhaps already acknowledged Leibniz as a historical specialty earlier: “Anything I might say about Leibniz would be amateurish, undeserving of others’ attention, and better left unsaid” (On the Plurality of Worlds (Blackwell, 1986, ISBN 978-0631139942), p. viii). Alas, what I wouldn’t give to know Lewis’s amateurish thoughts on Leibniz! this agrees well with my own sense that the rise of specialties occurred sometime in the 1980’s. It also agrees with Weatherson’s characterization, here, of his Era 3 (1966–1981): “where the classic works of contemporary analytic philosophy were written by writers like Kripke, Lewis, Putnam, Rawls, Thomson, Singer and Frankfurt.” “Classic works of contemporary philosophy” (as I have had occasion to point out recently on Facebook) is, like “Classic New Coke,” a contradiction in terms. But we can all understand Weatherson’s meaning: since this was the era of the last authorities, the most influential works of this era — the ones that would have become central texts of newly emerging sects — are, so to speak, frozen in, as both “classic” and, at the same time, permanently “contemporary.” And this also explains why he describes his Era 4 (1982–1998) as “the one I have the hardest time conceptualising.” This era, being the era in which empire of specialties arose and grew to its maximum extent, is the era Weatherson (at least subconsciously[16]I don’t know Weatherson personally, nor have I read his extensive other professional publication, but his writing in this book is thoughtful (perhaps even contains some deep reflection and relevant philosophical insights), so I don’t want to suggest that he is unaware of the matters I bring up here. I should emphasize, as well, that he does not aim (at least, in what he has published so far) to study the change I am talking about; I am putting his work to a use somewhat different than what he intends. In particular: the author of an article is not among the data that his algorithm works with. Still, I think it not too unfair to say that his work is animated by a background feeling that division into specialties a normal state for philosophy.) imagines extending over the whole history of philosophy. How could he differentiate it in the way conceptualization requires?

I have one last matter to consider from Weatherson, before getting back to Sangiacomo and, by way of him, back to Kant (and Hume). This is the nature of his Era 5 (1999–2013), which he describes somewhat anticlimactically as “dominated by a number of distinctive topics, such as reasons, vagueness, contextualism and Williamsonian epistemology and metaphysics.” I suppose Era 5 is all of that, but it is also the period which saw the appearance of the “bad topic” he discusses here and again here. This bad topic (“bad” in the sense that it is not the kind of result Weatherson wants) does not appear in the list he eventually settled on; it resulted from an attempt apply too many iterated “refinements” to the same underlying model — I refer you to the above linked text for technical details. Here is Weatherson’s description of what happened:

One signature problem with the kind of text mining I’m doing is that it can’t tell the difference between a change of vocabulary that is the result of a change in subject matter, and a change of vocabulary that is the result of a change in verbal fashions. . . .

So after 100 iterations, we ended up with a model that wasn’t a philosophical topic at all, but was characterized by the buzzwords of recent philosophy.

To be specific, the words the over-refined model gave most weight in identifying this topic were: accounts, role, commitment, commitments, account, proposal, constitutive, practices, challenge, typically, claims, worry, approach, relevant, project, focus, features, issue, appeal, provide. I blush to say that I have relied on some of these myself in the past and that, even after having seen the list, I sometimes find it difficult to avoid them. Weatherson hesitates between two possible explanations of why a bad topic like this only appeared towards the end of his period:

  1. There has been a linguistic revolution over the last generation, and philosophers now write in a very different style to how they wrote a generation ago.
  2. This is an artifact of model building, and if you stopped the model at any time, and ran the same study I did, you’d get results like this. That is, doing what I did will get you weird results whenever there is linguistic drift, and there is always linguistic drift.

But, although I have not a shred of that meticulous empirical evidence which Weatherson requires and could supply, I don’t doubt for a moment that the answer is (1). This certainty of mine you should probably distrust: the old are always affronted by the language of the young, after all. I can only say how it seems to me. The rise of these buzzwords, namely, seems to me a sign that it is not only Weatherson’s algorithm that can’t tell the difference between subject matter and verbal fashion, but that, rather, the algorithms of professionalization themselves are beginning to have the same problem. We are beginning to train up producers and detectors of buzzwords.

Back to Sangiacomo, then. In place of the algorithmic — that is, as I am using the term: unreflective — process which has brought up a few works by a few great thinkers for our consideration, he offers a different algorithmic process which will bring up different pieces of text. Granted, there may be little reason to suppose, with Hegel, that the old algorithm, the algorithm of history, was suffused with the cunning of the spirit, such that it could be trusted always to output exactly what deep reflection requires. Whatever Hegel means by “the end of history,” it probably is not the kind of end we face now. Still, it would be reasonable to ask: what was wrong with the old algorithm, and how will the new one improve on it? Sangiacomo (reportedly) explains:

Our entire understanding of early modern philosophy is based on the works of five authors.[17]The five (or six) authors he counts, in the order in which Lelong lists them, are: Spinoza, Descartes, Hume, Locke, Newton “and, at least according to some, Kant.” A few other names, e.g. Leibniz, could probably have been added to this. So naturally, that is an incredibly distorted view. But what is the alternative?…

I realized that I would never be able to gain a complete picture of that period. Because if it turns out that there were actually as many as a thousand philosophers active during that time, what would be the advantage of adding ten to fifteen names to the existing canon? We need new tools.

How he arrived at the figure of (as many as) a thousand “active” philosophers, I can only guess. But I can say a few things about this. First of all, if the knowledge we need about our history, in order to make reflection possible, consists in knowing the views of as many as a thousand different people, then there is no hope for us, because there is no way we can possibly know that. Secondly, and worse: as philosophy is now constituted, we, the thousands upon thousands of philosophers who are said to be active now, are very far from knowing the views even of those five authors. Each of these authors constitutes a specialty (or rather, as things now stand, a relative specialty, which is a generality to the various specialties lying under it). The owners of a specialty are alone authorized to say, to one another, in the journals, what each one of these authors said, and always on the condition that they cite one another. The rest of us are authorized, more or less, to supply caricatures of them for the purposes of jokes at department meetings, lectures to undergraduates, and chronically unfunny comic strips. Adding ten or fifteen new history AOS’s would only make this situation worse. To be added to the canon, under our present constitution, is to be culturally appropriated: removed from the commons and brought under the corporate control of a specialty. After canonization, the figure in question is no longer known by us: from our point of view, they have been canceled. Canonization is appropriation is cancellation. Only those who are (still) too obscure to be claimed by any specialty can still be known by us at all.

What is the true fear, the true uneasiness, that finds expression as the fear that “our” “view” is “incredibly distorted”? There is an inconsistency between the way we understand contemporary philosophy (the classics of Weatherson’s Era 3, plus the vast output or Eras 4 and 5) and the way we — that is, the specialists among us — understand the history of philosophy. To know philosophy now, to be a professional, means (1) to know to which specialty one belongs, who else belongs to it, when to cite them, and so forth; and (2), increasingly, to be up on the latest verbal fashions, to know what buzzwords to emit. But no one knows anything parallel about philosophy before Era 3. It would be impossible to know (1), since specialties did not yet exist, but some historians, so-called contextualists, now supply a simulacrum by pretending that, say, Descartes, spent much of his time eagerly scanning the latest literature by every (active) member of his “field” (which was …?), as if he lay under our obligation to read recent things so as to able to spew a list of name–date pairs into the appropriate footnote. But this leaves (2) unaccounted for, and (2) is what Sangiacomo’s algorithm promises to supply. Along with his collaborator, Christian Marocico, and others at the CIT, he “used ShiCo (Shifting Concepts through Time), an open source tool developed by the Netherlands eScience Center and the University of Utrecht, to analyse shifts in words in a historical context.” Here is how it worked out:

Of course, the tool needed a little tinkering. ShiCo was developed to record conceptual shifts in 20th-century newspaper reports. Data scientist Marocico adapted the model so that the algorithm was able to analyse 70,000 letters exchanged between philosophers and academics during the 17th and 18th centuries. . . .  They investigated which word associations were formed for certain scientific concepts, and how this changed over the years. Sometimes this produced remarkable results. “For example, we found that in the 17th century the English word ‘spirit’ was used in both a chemical and a religious context. In the 18th century, however, we see that the word has lost its chemical meaning. It illustrates how concepts change.”

It may not be obvious that this result is remarkable, that is, worthy of remark.[18]Contrast it with Locke’s story about a related word: “I was once in a Meeting of very learned and ingenious Physicians, where by chance there arose a Question, whether any Liquor passed through the Filaments of the Nerves. The Debate having been manag’d a good while, by variety of Arguments on both sides, I … desired, That, before they went any further on in this Dispute, they would first examine, and establish amongst them, what the Word Liquor signified. . . .  They were pleased to comply with my Motion, and upon Examination found that the signification of that Word, was not so settled or certain, as they had all imagined; but that each of them made it a sign of a different complex Idea. This made them perceive that the Main of their Dispute was about the signification of that Term; … a thing which when considered, [they] thought it not worth the contending about.” (3.9.16) But we can appreciate what might make it remarkable by comparing it with what Weatherson (somewhere in here) says his algorithm has detected in its “bad topic”: “talking about the ‘commitments of an account’ rather than the ‘consequences of a theory’ is a way to mark one’s philosophical writing as being up to date with modern terminology.” This is the kind of thing a professional ought to know about the present, and so it must be important to know it about the past, as well.

Whether deep reflection, of the kind essential to philosophy, can ever be prompted by knowledge of verbal fashions — either contemporary verbal fashions or the verbal fashions of the 17th century — I honestly don’t know. The spirit, as that term is used in a Hegelian context, rather than a chemical or a religious one — the spirit, whose perfections non comprehendere, sed quocunque modo attingere cogitatione possumus, is always more cunning that we expect; it always overflows our concepts, shift them as we might. It is said to have revealed itself in cracked turtle shells, in images on toast, in words written on subway walls and tenement halls, in riddle games in the dark at the roots of the Hithaeglir, in comic strips, in superhero movies, in Facebook memes, in a brutal command to wipe out Amalek — perhaps, who can say, it reveals itself in philosophical buzzwords, as well. But, if we are to act on God’s apparent, presumptive, antecedent will, rather than his secret, decisive, consequent one, I would have to say that this method does not look promising.

What about the old algorithm, then — let’s say, the old algorithm as it operated (or: in one of the many ways it operated) before the rise of our current specialties? What reason is there, Hegel’s reasons aside, to expect anything useful from that? “Why,” to quote the title of Lenz’s blog post cited above, “would we want to call people ‘great thinkers’?” To tell the truth, although I have often used that phrase in the past, I am growing tired of it. To present day ears, at least, it sounds like an award we are giving to some people, as if we were doing something for them, rather than: that they did something for us, namely, made us possible. I am pleased to find that Justin E.H. Smith makes some remarks along the same lines in a newsletter I received as I was writing this, which I will quote, since he writes so much better than I do. Speaking of demands that we diversity our canon, he says:

An implicit premise … is that it is a good thing for a thinker to end up on a course syllabus, a sort of posthumous prize handed out to the dead by the living. Thus it is presumed that to study the old canon, as we inherited it, the canon of dead white men, is to honour these dead white men.

But scholarship is not a fan club, and to read Descartes or Kant is an undertaking that is entirely neutral with respect to whether they are praiseworthy as thinkers (a fortiori whether they are praiseworthy as people). When we study the history of philosophy we want, first, to know what happened, and second, to know how what happened shaped the world we inherited. This is not a celebration, but a solemn duty.[19]I received this essay as an edition of Smith’s newsletter, to which I urge you to subscribe, but for now you can see it online without a subscription here.

I disagree with some of this,[20]I agree that at least some of the complaints about the old canon arise from a thought like this: that assigning someone’s book is like building a monument to them (which, however, does rather undermine the meme about, “Won’t they be surprised to hear that we can learn history from books?”). But I doubt that all or most of the impetus for a new, more diverse canon can be traced to same source. What students want to see on the syllabus is not someone to whom an award is due, but someone who “looks like them,” who reminds them of themselves. And, in some sense, as I will go on to say, that is exactly what they should want. I can’t think that what is really desirable in this respect is to find someone in the reading list the same color or gender as yourself, but I can see that feeling alienated by the color and gender you exclusively do find might be an obstacle. We all face such obstacles, but some more than others. We should try to remove them if we can. and with some of the other things Smith says in his essay, but I agree with the main point: canonization is not a reward. So, although I continue to think that most of the canonized, if not all, were indeed praiseworthy, and in particular that they were praiseworthy as (finitely!) wise and good — wiser and better than me, at any rate — I would rather avoid calling them great anything. Let me call them my authorities.

How can reading an authority prompt deep reflection? What is “reflection”? It is reflective, reflex. I recognize a pattern, all right, something familiar, but the pattern, the παράδειγμαparadeigma, is familiar, not because I have previously picked it up with my buzzword detector, but because I myself am its image, ut esset tanquam nota artificis operi suo impressa; nec etiam opus est ut nota illa fit aliqua res ab opere ipso diversa. This kind of pattern recognition is called recollection, ἀνάμνησιςanamnesis, and it cannot be performed by algorithm: not because machines, in the end, really only know how to count; not because πᾶν τὸ πρὸς ἑαυτὸ ἐπιστρεπτικὸνpan to pros heauto epistreptikon ἀσώματόν ἐστινasomaton estin; not (if I may allude to a certain harasser) because recollection depends on the actual physical–chemical properties of actual human brains — but simply because recollection is reflection, and, as you will recall, I have defined an algorithm as a process that does not require reflection. To see an electronic computer perform ἀνάμνησιςanamnesis would surprise me no more than to see a human being do so, which is to say: it would surprise me a great deal, and always again surprise me even if I were to see it always and everywhere I looked (although, sadly, there is little prospect of that). But I would also know, of the computer or the human, that they could have performed this recollection only thanks to a reminder. The sign, the trace that can remind me in this way leads always beyond my shifting concepts, לעילא לעילא מכל ברכתא ושירתאle-ʿeila le-ʿeila mi-kol birchata ve-shirata, and is hence everywhere beyond needing any praise from me. But I need everything from it.

I close by coming back to Kant, and from Kant to Hume. What did Kant think was wrong in Hume? Not, alas, that he was a racist: Hume’s racism, if anything, thanks to its tentative, empirical nature, would scarcely have seemed racist enough for Kant. But he did think Hume had committed an error or two. The worst one, as it happens, was his failure to reflect, or rather, his failure to reflect in the right way. Reflection (I will assert here without evidence) is the (logical) topic Kant takes up in the Analytic of Principles, in its third Hauptstück, Of the Ground of the Distinction of all Objects in general into Phenomena and Noumena.[21]The distinction itself between phenomena and noumena is also a local distinction: it distinguishes what is within the bounds of experience (phenomena) from what is without (noumena). These two places constitute what Kant calls transcendental topic — or, in the terminology I have adopted here: they are the two transcendental topics (only, of course, that the second one is empty). Accordingly, he begins there by raising a question about everything that has preceded. Granted that everything you say is true, Kant, why tell us? (For, it must always be recalled: there is a negative duty not to lie, not, absurdly, a positive duty to say everything that is true.)

If, then, through this critical investigation, we learn no more than what, even without so subtle an inquiry [Nachforschung], we would anyway have executed [von selbst wol würden ausgeübt haben] in the merely empirical use of the intellect, then it seems that the advantage which one draws from it is not worth the effort and the preparation. (A237/B296)

Leaving aside the conclusion, where does the premise of this objection come from? Who says that we would have done the same thing, anyway, without this subtle inquiry? Who indeed. Who says, for example, that we would have employed the law of causality, that every event must have a cause, even if we not only could not demonstrate the truth of that judgment, but could not even show title to the concepts it contains? The answer is: Hume. Hume says that. So what does the whole apparatus of the Transcendental Analytic contribute that was missing in Hume? If we must, anyway, use the concept of a cause (when, for example, we speculate, rightly or wrongly, about the causes of certain historical–geographical trends), what good to us is a demonstration of the Second Analogy? Kant answers:

the intellect when occupied with its merely empirical use, which does not reflect [nicht nachsinnt] upon the sources of its own cognition,[22]Nachsinnung is not the official equivalent Kant later gives for reflexio (rather: Uberlegung — see A260/B316). But I feel the translation is justified, given that he is talking here about what he goes on to call “transcendental reflection”: investigating the relation of one’s cognitions to the faculties that are their “sources.” can indeed progress very well, but there is one thing it cannot achieve, namely, to determine of itself the limits of its use, and to know what may lie within or without its whole sphere.

What Kant thinks Hume failed to do — or perhaps did incorrectly, so as to obtain a negative (skeptical) result — was transcendental reflection; and he adds that Hume was therefore led to say (as, indeed, Hume does say quite clearly) that reason cannot set its own limits (i.e., transcendentally locate itself). Which, from Kant’s point of view, is an infinitely bad error, both theoretical and practical: reason’s theoretical self-limitation is the destruction of knowledge that makes room for faith, and reason’s practical self-limitation is self-legislation, autonomy. For Kant, Hume is guilty, not of making some particular moral error, but rather of asserting that morality, as such, is impossible.

So why would Kant want to call Hume a “great thinker” (or words to that effect)? Or, as I would rather say: why would Kant take Hume as an authority?

Because Hume is perhaps the most ingenious [the richest in spirit, geistreichste — not, one assumes, in the chemical sense] among all skeptics, and incontrovertibly the most eminent with respect to the influence which the skeptical procedure can have upon the awakening of a thoroughgoing rational self-examination [einer gründlichen Vernunftprüfung], it may well be worth the trouble to represent, insofar as it is appropriate to my aim, the course of his conclusions and the strayings [Verirrungen], which nevertheless began on the track of the truth,[23]It is difficult to translate this passage correctly. The idea is that, when we follow the course or path (Gang) of Hume’s reasoning, we will find that, at certain points, he takes off after the tracks or traces (Spur) of the truth, but nevertheless goes astray (verirren). of so insightful and estimable a man. (A764/B792)

Hume supplied the place — the authoritative topic, so to speak — of Kant’s awakening.

I freely confess: the recollection [Erinnerung] of David Hume was just that which, many years ago, first interrupted my dogmatic slumber and gave to my investigations in the field of speculative philosophy an entirely different direction. (Prolegomena, Ak. 4:260,6–9)

Hume himself — wrong, infinitely wrong as he may have been — was for Kant the trace of truth, the trail, the spoor, the path that leads right inward. In Kant’s terminology: one might say that Hume, in his life and thought, constituted a symbolic, indirect exhibition of the transcendental idea of freedom. The recollection of Hume was, then, for Kant, the recollection of his own true self, of his transcendental ego. To name a tower after such a one, even a monument as high as the moon; to put him on a syllabus, or even, like some notice about the plagiarism policy, on every syllabus; to hire a fantastic young scholar who specializes in his work, or even to hire a whole army of such young scholars, each more fantastic than the last — all this is nothing, not the smallest part of what we owe him, or, rather, what we would owe him, if he needed anything from the likes of us.[24]Here I am speaking about what we might owe Hume in common, not about what is owed to him by those specialists who have appropriated him and used him as a means, extracting the value of his labor to further their own careers. But these latter will have to make their own reckoning.

But do we need him! What have I just been saying, all this time, but that our dogmatic slumber now appears terminal? “Any prospect of awakening or coming to life to a dead man makes indifferent all times and places. The place where that may occur is always the same, and indescribably pleasant to all our senses” (Walden, 5.9). He is, incontrovertibly, the heresiarch most eminent with regard to what we require. We cannot receive him now, cannot receive anyone: Hume and all the others were already canceled the moment they were transformed from authorities into specialties. To quote someone else who had a morally wrong view here and there: only a god can save us. But, if and when the day should come, it has already been prophesied: et servus meus David princeps in medio eorum.


  1. Andrea Sangiacomo has published first results: When he talks about networks, it's in the sense of "network analysis", a subdiscipline of computational social/intellectual history.

    1. Thanks Stefan on both points. I am ill-informed, as I said. But: leaving aside my perhaps kind of silly pun on the word "network": do you think I'm wrong to say that these networks (all added together) are what processed all the text and eventually output a canon?