I have been trying to make sense of the changes that are happening in science for many years at this point. Sometimes, I try to write some of it down. This time, I want to focus on some of the effects surrounding computer science. Specifically, I want to look at the success of computer science and at recent activities by tech-companies.

The ubiquity of computers, and more recently generative AI, is well known and a lot has been, is, and will be said about it. Here, I want to focus specifically on the impact on the organization of science and the notion of leading sciences. I claim that tech-companies are, knowingly or not, constructing the institutions that would install a new leading science. This would be a significant change to the organization of science and, as I will illustrate with historical examples, has implications far beyond science into society.

What is a leading science?

The term leading science is my bad translation of the german term ‚Leitwissenschaft‘. For the rest of this text, I will use the term Leitwissenschaft, since the entire notion seems to have received more attention in the german language region which has a designated term for it. This also means that I have to spend some time explaining the term, since you, dear reader, may be quite unfamiliar with it.

One of the difficulties of defining the term Leitwissenschaft is that every Leitwissenschaft has achieved and used this status differently, which means that any specific feature that one could put in a definition would have an exception. I will still give a rough outline, but then I will focus on illustrating the concept through examples that should make this abstract definition tangible. At its core, Leitwissenschaft is a status that a scientific discipline can have within a given social organization of science. It marks this particular discipline as the one of highest authority and importance. This can show itself in more down-to-earth things like salaries or the order of faculty members during ceremony, but it also has implications on scientific disputes, hierarchies of knowledge and theories, or what is considered as truth. At its core, I would say that a Leitwissenschaft is the final arbiter of truth (within its organization of science). It has — in a sense that is defined by its social context — access to deeper or more ‚real‘ truth than other disciplines. It is important to note that the position of the Leitwissenschaft is constructed and maintained through institutions and/or society. Its claim to deeper truth requires that, whenever someone argues otherwise, it is not taken seriously or the person is silenced. We will go over several examples below and any reader will likely identify at least one example, where they find the claim to deeper truth to be ridiculous. Which example that is depends on your personal idea of science-organization.

With this, we can go to some examples. These examples serve two purposes: They should make the definition above more concrete and they should illustrate what Leitwissenschaften can do, how they get established, and how they can lose their status. We will be focusing on Europe, since the notion of Leitwissenschaften, the way I am using it, is coming from Europe. For the next two subsections, I am working heavily with Chapter 12 of ‚Geschichte der Universität in Europa, Volume 2‘ (ISBN 3-406-36953-7)

16. century - Theology

The leading science of the 16th century gives us a good idea of some core features of a Leitwissenschaft. The structure of science in 16th century Europe were still heavily influenced by the structure from the high medieval period, which grouped higher education (and with it scholarly activity) into four groups with a clear hierarchy: Theology, law, medicine, and the „liberal arts“ (for lack of a better word; it covered the Trivium and Quadrivium). Theology was at the top of this hierarchy, which showed itself in many ways including social status, salary, and the claim to truth. The claim to truth is one of the most crucial elements of a Leitwissenschaft: If in doubt, it is the final arbiter over True and False1. Theology was in this position because of the dominance of the catholic church in (western) Europe in the time leading up to the 16th century, the idea of truth that came with this dominance, and the notion of abstract truth (perhaps best understood in a metaphysical sense) that lends itself much more to theology than the study of law or medicine.

Beyond the claim to truth, the Leitwissenschaft can also define the other disciplines. For example, the notion of practical science in the 16th century — which I will not explain in detail — allowed to give purpose to and thus legitimize the field of law as an additional discipline of science. However, this time-period also shows that not every discipline needs to be defined in terms of language built for and around the Leitwissenschaft: Medicine was fairly isolated in the scientific constellation of 16th century Europe, which may have allowed it to go a more separate path that would play an important role further down the road. Despite the significance of Theology in science and society of the time, the scientific theory that made Theology the Leitwissenschaft did not penetrate every discipline.

The example of Theology also highlights how a Leitwissenschaft is not just an academic matter but can have substantial impact on society. To cite the historian Wilhelm Schmidt-Biggemann (translated from german): „The function of the Leitwissenschaft was science-political; it depended on the success of a science. Without the scientific leadership function of Theology at the beginning of the 16th century, the Reformation in the Holy Roman Empire, France and England are unthinkable. Theological concepts defined the models that lead to the religious and political upheavals of the 16th and 17th century.“ This example highlights that the very setup of science — how is science organized; what is the final authority on True and False; what institutions maintain this structure — has direct consequences on societal developments of large magnitude. We will see that the choice of Leitwissenschaft in different times aligns with respective important societal developments. And we will see that the scientific and societal dimensions tend to feed into each other, rather than one outright causing the other.

Theology also gives us an important illustration how the Leitwissenschaft can fail. Throughout the 16th century, the Reformation caused confessional fractures across (formerly) catholic Europe. The inability of Theology to settle the intellectual disagreements of this fracture destroyed its credibility as a Leitwissenschaft. By the turn of the century, it would be displaced by Jurisprudence, the science of law.

17th century - Jurisprudence and the sovereignty of the state

Weakness of a given Leitwissenschaft generally does not suffice to displace it. As long as the social machinery around it is running, it is easy to imagine that it can formally hold its position for quite a while. This changes when there is an alternative. One that is better at dealing with pressing questions. A great example for this is the case of Jurisprudence, which displaced Theology: The significance of Jurisprudence in the 17th century starts with the need to answer questions about legal principle. The arguments over the Reformation destroyed the credibility of Theology regarding any question over which there were confessional divisions. What was left was a ‚theology of nature‘, the Theology that comes before humans ‚muddle the water with their confessional disputes‘. This created a need for legal principles coming from ‚nature‘2. At the same time, the discoveries and conquests in the Americas meant that Europeans were encountering people that did not fit into the framework of canonical (and roman) law. There was a need for principles of law that naturally applied to all people and thus could not be based on christian revelation.

On top of that, the 17th century is an important moment in the formation of states in Europe. The Jurisprudence of this time had to answer questions about the legal relationship between a ruler and its subordinates: For example, what is the legal situation of a commune whose confession differs from the confession of its ruler (due to conversion by either, or a change in ruler)? Similarly, what is the legal legitimization of a ruler when the weakening of the church has also weakened the common religious justification?

Jurisprudence was able to give more attractive, more practically executable, and confessionally independent answers to these questions than Theology. In addition, the answers by Jurisprudence gave worldly rulers a justification to outrank the church. For one reason or the other, people of the time, especially decision makers, preferred to pick up the answers by Jurisprudence over the answers by Theology. The ability to answer big questions in attractive ways3 helped Jurisprudence to start outranking Theology. Just like Theology, Jurisprudence didn’t last forever though. The scientific changes that started in the 17th century and continued throughout the enlightenment and the French revolution would bring fundamental changes to the intellectual organization of science and with it, there would be a new Leitwissenschaft. To prevent these examples from taking up even more space, I will not discuss the very interesting 19th century and skip to the 20th century.

The short 20th century - multiple claims, but no successful establishment

The first half of the 20th century saw a substantial set of changes in the fundamental sciences, in technology, and in society. By the beginning of the postwar period, we are in a very different world from the early 1900s. Based on the research I did for this text, there is no clearly established Leitwissenschaft during the second half of the 20th century. This is not for a lack of attempts. Several disciplines tried to stake their claim as Leitwissenschaft, for example Sociology, Neuroscience, and Mathematics, but I could not find an obvious consensus.4

This raises an important question. Why did no discipline succeed at establishing itself as Leitwissenschaft when there were several attempts? While there are many factors that feed into that, one important factor is that no discipline had the necessary institutional backing that would give their claims the necessary authority. This is not the sole reason, but the absence of institutions that would declare a Leitwissenschaft certainly helped in preventing the establishment of one. While some institutions would certainly declare a discipline as a core discipline, for example, a university might center its entire educational program around one discipline, no institution had the necessary reach, nor would a critical mass of institutions find consensus on putting a particular discipline above others.

I also want to highlight that the historical examples we looked at so far illustrate that there is nothing inevitable or natural about a particular discipline being the foundation of our idea of knowledge and science. It used to be something else and it will change again. In particular, it is absolutely possible, and even likely, that a new Leitwissenschaft will be established. This brings us to the 21st century, where the constellation of institutions and technology has changed.

The present, the future, and computer science

The most important institutional change in the past few decades lies in the activities of Tech-companies and their power to make changes that affect all of society. The way that digital technology has penetrated societies across the globe, together with the availability of enormous amount of capital has allowed Tech-companies to not only build technologies, but to shape societies and to weigh in on matters of science. We are only interested in the impact on (the organization of) science here, so let’s narrow down to this. We see the following trends that have been developing for quite a while and have accelerated in recent years:

  1. The recruitment of scientists other than computer scientists by tech-companies. Neuroscientists have been of interest for questions of artificial intelligence and neural implants for a while. Social media companies, like Facebook, have hired sociologist and psychologists to study the effects of social media on mental health and democracy for a while and — very importantly — have opted not to publish the results when they didn’t like them. The recent developments in generative AI have expanded this interest into various other aspects ranging from biology to philosophy.
  2. Generative AI allows one to receive true-or-false statements from a machine on a far broader range of subjects than before. While computers have of course provided answers to computations for a long time, one can now ask questions about law, life, or relationships and receive outputs that can be taken as a direct response. It is important to highlight here that — in many instances — it is difficult or downright impossible to retrace how the computer has arrived at its output.5 Because of this, these outputs effectively provide a new source of True or False that has been socially adapted very quickly.
  3. For various motivations, ranging from marketing to the imagination of creating a machine that does research, tech-companies are attempting to create devices that are „correct“ (i.e. don’t hallucinate, whose statements hold up to verification) virtually all the time and that are able to answer questions in science. In this context, they have also attempted to make use of automated verifiers like Lean, which is able to check formal proofs in mathematics.6 In the context of science, this is essential to make the generative AI an authority that determines whether something is True or not.
  4. Computers have become essential throughout science in many ways. Not unlike the needs for Jurisprudence during the 17th century, computer science plays a crucial role in the creation of the computation infrastructure that humanity is building. To utilize these computational resources, researchers have put a lot of effort to „put the world into the computer“ (to use the language of David Gugerli). Putting the world into the computer is not a perfect process, but it is useful. Generative AI has added a maximally pervasive layer to this. Putting the world and its truth into the computer is a highly imperfect and non-unique process7, but it is very useful and — very importantly — once it is done, it is fast.

If we put all these points together, we can see how the actions of tech-companies create a set of institutions that establish a new organization of science: Researchers are working for private tech-companies. They help them teaching generative AI to perform sufficiently accurate in all scientific disciplines so that generative AI gets maximally adapted. At the end of this, the computer becomes the source of True and False and the science of computers becomes the core of science, since it is the science about the machines that produce the True and False.8

It is important to highlight that it is not necessary for computer science (or generative AI) to actually be right in what it does. It just needs to be successful enough and have the necessary institutional backing so that people can call it the arbiter of True and False without getting a rebuttal that is taken seriously. As has hopefully become clear from the historical examples, a Leitwissenschaft is not an actual fundamental source of truth. It is just considered as such by a society structured to take it so.

In a similar spirit, it is not necessary for computer science to do everything. It might be tempting to argue that a computer cannot do experiments (outside of simulations) or field studies, but that makes little difference if the experiment is primarily conceived by a computer and, perhaps even more importantly, if it is ‚read‘, reviewed, and contextualized by a computer, which is increasingly the case. The developments and progress in robotics over the last few decades also make it likely that computer science can be placed at the core of experiments and field studies as well.

To be clear, I don’t think many tech-leaders think of themselves as doing this. Rather, the historical combination of technology, state of marketing, control of capital, and incentives makes this happen. The result is an institutionalization of a new Leitwissenschaft. Like the ones we have discussed from the past, its institutionalization has unique features and consequences.

Addendum: A concrete example

I wrote most of this text in August 2026. However, the events around the Navier-Stokes equation in early September are such a great case study for this topic that I want to add it as an example.

To start, I want to give a very simplified explanation of the Navier-Stokes smoothness and existence problem. (I will call it the Navier-Stokes problem.) Navier-Stokes is a partial differential equation (PDE) that is important for fluid mechanics. The Navier-Stokes problem asks four questions: A) Under a certain extra assumption on the equation, is there always a smooth solution? C) Without this extra assumption on the equation, does there exist initial conditions so that there is no smooth solution? Parts B) and D) are the same questions as A) and C) but for the 3-dimensional torus, instead of R^3.9

So what happened in early September 2026: OpenAI announced a proof of statements C) and D) of the Navier-Stokes problem. Right before the announcement, Tristan Buckmaster (NYU) and Levent Alpöge (Anthropic) reported their work on the Euler equation (an equation where one might ask similar questions and try similar strategies, but a little easier). The strategy used by OpenAIs model follows the same strategy used by Buckmaster and Alpöge.10 Buckmaster and Alpöge were working with Codex, a model by OpenAI, in their research and Buckmaster suggested that their work in Codex may have become training data of the model that OpenAI used to get their solution. OpenAI has stated that „While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models.“ As I am writing this, this matter is not yet resolved.

While there are many things to talk about both in this dispute and the results themselves, I want to focus specifically on how it relates to the topic of Leitwissenschaften. Since Buckmaster-Alpöge and OpenAI used the same strategy, it is quite possible that they may have managed to push their results from Euler to Navier-Stokes. This would have required more time, but because OpenAI was pressing the issue, this time was not available. Even their work on Eulers equations is not in the form that they would like, according to Buckmasters initial post on the matter. As far as I could find out, OpenAI started putting resources into Navier-Stokes when they heard rumours that it might have been solved.11 This is important, because Buckmaster-Alpöge and OpenAI have used two very different modes of working with generative AI (genAI) and these two modes are relevant to our topic here. We don’t fully understand how Buckmaster and Alpöge were using genAI, but there are some things we can infer. For example, they clearly did not have access to the same amount of funding to pay for genAI-use as OpenAI itself. It is also clear that Buckmaster is familiar with the origins of the strategy they were using, who came up with it, and why it can be used to work on this problem. Furthermore, the way Buckmaster writes about the first published version of their work suggests that they were operating under the idea that the output of the genAI needed to be reworked by them. Whether this simply means to improve the writing, fixing mistakes and gaps, or understanding the content of what the genAI outputs, I cannot say with my current knowledge. But it is clear that their mode of working with genAI involved substantial work by humans both before and after the use of the computer. They are displaying a mode where genAI at the very least heavily interacts with human knowledge and expertise. There might be even more human steering, but I cannot tell so far. We know less about how the research team at OpenAI actually got their final output, but they are definitely presenting it as an output by a genAI where human interference has been removed as much as possible. The genAI is presented as the main provider of the result, of the True.

These two modes portray fundamentally different hierarchies between humans and genAI in research and two fundamentally different organizations of science and knowledge. Even though both approaches use genAI, they represent completely different ideas of the role of computers and genAI in our attempts to understand something about the world. They conflict in the position of computers and genAI in the organization of science, specifically how much leading power computers (and their experts) hold. As Buckmaster said, they made their work public earlier than they liked. OpenAI created this time-pressure. OpenAI could have waited for Buckmaster and Alpöge to finish their work, perhaps even generalize it to Navier-Stokes. Since they used Codex for their work, it would have been a showcase of what genAI (specifically a genAI of OpenAI) can do in science. But that was not satisfactory to OpenAI. It had to be done by them and it had to be done in their mode of interaction between humans and computers. By doing this, they have12 directly attacked the idea of science-organization that Buckmaster and Alpöge displayed through their way of working. It is a great example how the actions of a tech-company actively pushed towards organizing science in a particular way, that sees computer science as a Leitwissenschaft in a particular way, with particular structures and particular hierarchies.

Is this a problem?

If you have read this far, you can probably think of a few problems with the organization of science that I have just outlined. For starters, science is a project of all humanity. Institutionalizing it via companies that are under the control of a very small number of people eradicates every control mechanism that a project of all humanity needs (remember the works in psychology and sociology done by Facebook that weren’t published): After all, all of humanity has a stake in such a project, is supposed to benefit from it, and, yes, is supposed to be entangled with it in a way that matters.13 Furthermore, we have discussed above that putting the world in a computer is an imperfect process. It creates blindspots that cannot be uncovered if the computer is at the center of science. This issue is perhaps easier to understand by looking to the past: It is easy to see today that science cannot be reduced to the divine revelation of a deity. The same issue applies to every Leitwissenschaft; it is just easier to see when you are not living in the society that relies on a Leitwissenschaft to fulfill a purpose. If we think back to the second half of the 20st century, we see that it is absolutely possible to organize science in a way that has no Leitwissenschaft and this is likely a much better way to deal with blindspots and to avoid unnecessary hierarchies that are based on status. Indeed, it is very unfortunate that the second half of the 20st century was not used to institutionalize the absence of Leitwissenschaften. We are likely to pay the price for this in the coming decades. Finally, I want to point out a particular characteristic of this currently arising institutionalization. It penetrates every aspect of science. Due to the ubiquity of both computers and the aggressive attempts to make generative AI equally ubiquitous, this institutionalization of computer science as Leitwissenschaft will penetrate everything. As discussed before, even Theology during the highly religious 16th century left some scientific disciplines largely to their own devices.

How does this help?

Thinking of recent (and less recent) developments in terms of the institutionalization of a new Leitwissenschaft can help analyze what broader effects certain decisions can have. It also gives a much better sense of the scope: Every proposal about AI-regulation that I have encountered would likely have little effect on the social reorganization of science. No amount of watermarking AI-generated content changes whether we view it as the arbiter for True. And I would much prefer if we did not walk into this blindly while thinking that we did all the social and political moves that were possible. This analysis also helps in identifying some directions: The most optimistic outcome would be if science would stay without a Leitwissenschaft, indeed if this was institutionalized. More modestly, we can perhaps make sure that the Leitwissenschaft (and a lot of science with it) is not directly in the hands of private capital. And with a similar priority, we should aim to prevent one Leitwissenschaft to penetrate science so completely. We would be well advised to avoid computer science projecting True and False onto science and society in such totality.

  1. or Correct and Incorrect if the Leitwissenschaft is math; or Just(ified) and Unjust(ified) if it’s law… you get the idea. ↩

  2. Even just within the 17th century, the word ‚nature‘ can mean all sorts of things. As interesting as this is, I will not go into it and you will have to accept the term with imprecision. ↩

  3. To bring this into territory that might be more familiar: Imagine if Jurisprudence was able to solve an open problem in mathematics that was considered to be very difficult. ↩

  4. It’s important to distinguish between the discipline and its practitioners here. The practitioners would have wildly differing opinions on such claims. ↩

  5. A fact made worse by the lack of transparency by the companies developing generative AI. ↩

  6. Without going into too much detail: There is a precisely formulated notion of formal proof in mathematics that effectively goes back to David Hilbert. This formal notion of proof can be checked for correctness, which is a crucial aspect how mathematicians reach consensus on what is correct or incorrect. Checking correctness can be automated, if the proof is fully formalized. However, this only checks whether a formal proof is internally correct. This cannot check whether a formal proof proves what it is supposed to prove. The translation from the formal language into the language that mathematics usually uses is separate from the verification of a formal proof and there are some inherent obstacles to automating that part. ↩

  7. Not unlike putting the world into the word of a deity, or reducing it to the language of the law, etc. ↩

  8. It is also quite possible that the machines will need some type of software maintenance from computer scientists. ↩

  9. If this is strange and unfamiliar, don’t worry about it. ↩

  10. This strategy was built earlier by other mathematicians. As is very common for difficult problems, it takes more than one group to build and execute a successful strategy. ↩

  11. This point confuses me greatly. Buckmaster and Alpöge were using a product of OpenAIs. Why did OpenAI have to be first in using their product to make this proof? Someone else successfully using their product independently seems like much better marketing to me. ↩

  12. consciously or not ↩

  13. Yes, generative AI entangles nearly every human being into science via the training data. However, this process is entirely extractive and is neither a proper nor an appropriate entanglement of a person with science. ↩