Some of the reasoning behind those quotes can be found here, and a recently accepted research article (in English) here.

Here’s the abstract:

Has the emerging field of computational social science grown increasingly intellectually integrated over time, or does the field remain fragmented into several distinct epistemic communities? Because research literature is still inconclusive in this regard, we use data about citations to study the development of the intellectual structure of computational social science. The findings suggest that network science has maintained a central position in computational social science over the years. Contrary to common perceptions, however, the field has not become more integrated over time. Rather, the network of cocitations among the field’s key authors exhibits increasing clustering and greater average path length over 2000–2020, indicating that computational social science is fracturing into various subcommunities, with loose connections to each other.

Keywords: cocitation analysis, cohesion, computational social science, intellectual integration, interdisciplinarity

Here’s how I got into diversity research. I remember the moment quite vividly: in November 2013 I was in my kitchen watching a Coursera video by Scott Page, an economics professor at Michigan University.

In that video, Page was presenting his (and his coauthor’s) famous finding from 2004, the diversity trumps ability theorem. The upshot of the theorem is that when people solve problems together, in groups, it makes sense to maximize the diversity of the group members instead of building the group from the individually best performers. This finding has later had lots of implications e.g. for hiring, when companies started following it in recruiting (McKinsey 2015). (But for some of the rather depressing recent developments, see here)

I thought that was really interesting, and started thinking about what such results would mean in my own research field, philosophy of science.

Ever since Thomas Kuhn’s Structure of Scientific Revolutions in the 1960s, philosophy had had a discussion about the role of rationality in science. Paul Feyerabend had suggested a very radical diversity thesis: “anything goes”, i.e. there is no scientific method, and we should let all flowers bloom. No theory should be permanently discarded, because later on, it may turn out the be true, after all.

How diversity works in research

Others were more cautious. For example in the 1990s, Philip Kitcher argued that no departures from rationality were needed, but instead, a beneficial division of labor between scientists would be enough to get to good collective outcomes.

Using modeling techniques from biology and economics, Kitcher’s paper launched a new field, which could be called social epistemology of science, and that’s the field that I’ve been working in since that afternoon in 2013 (O’Connor 2023, Šešelja 2023).

Here’s the agenda of the field (in the words of Steve Fuller, 1988):

“How should the pursuit of knowledge be organized, given that (…) knowledge is pursued by many human beings, each working on a more or less well-defined body of knowledge and each equipped with roughly the same imperfect cognitive capacities, albeit with varying degrees of access to one another’s activities?”

But how is this related to diversity? I should clarify some of the central concepts I’m talking about.

My main interest has been in cognitive diversity: We say that a group is cognitively diverse when its members differ, for example, with respect to their disciplinary background, expertise and skills, problem-solving heuristics, or strategy that each group member uses to explore a common field of research. 

Cognitive diversity can be distinguished from demographic diversity (its members occupying different social locations, e.g., with respect to gender, class, ethnic identity, nationality, and race), and from social value diversity (when its members endorse different social and political values, or act as representatives of different interest groups). More on these concepts in Rolin et al. 2023.

All these kinds of diversity can be understood as being intrinsically valuable, but what both Page and Kitcher are trying to articulate is the indirect benefit to science of having more diversity in the workforce.

(In political philosophy, a similar case has recently been made for democratic decision making, most forcefully by Helene Landemore.)

Cognitive diversity was of the key concepts of my 2020-2025 Academy fellow project called Modeling the republic of science. Obviously there is no one way that diversity works in science. You could say it has several functions, and these functions are served by different mechanisms.

The table is far from exhaustive, and does not include the potential downsides or costs of diversity to research. Here I just want to point out one thing (Row 5): Philosophers don’t agree about most things, but in philosophy of science there’s almost a universal consensus that scientific knowledge cannot be found in the brain of any particular scientist. Instead, scientific knowledge is collectively held by the scientific community. Furthermore as feminist philosopher of science Helen Longino (1990) famously argued, the objectivity of that knowledge is guaranteed by critical exchange of arguments among a diverse community of scientists.

Cognitively diverse university

In 2023 my colleague Säde Hormio and I took some of these ideas and applied them to higher-education research. Based on diversity arguments, we put forward a new argument for academic freedom: Namely, that the capacity of universities to serve the society by discovering significant truths depends on us having sufficiently diverse repertoires of (1) problems we study , (2) solutions we develop, and (3) people thinking about the world in different ways.

Alien intelligences and existential implications

Here’s what I’m doing now (2026-05). I have a new research project starting on June 1st, 2026: Scientist in the loop - Automation of scientific discovery (SCI-AI) is funded by an ERC Consolidator grant. The main aim of the project is to better understand how AI is changing scientific discovery.

In terms of methods and theory, SCI-AI continues what I did before in my study of collective problem solving in science. But now the question, from the diversity viewpoint, is what is happening to science now when these alien intelligences are entering academic research.

Not everyone likes that framing, and I understand why. I myself have mixed feelings about the AI transition. But I think it’s a real thing, the models are quite capable, and the benefits to advancement science can be really big. That said, risks and downsides are just as real.

I recently read this Terence Tao interview, and I agree with him that AI will change research in almost all scientific fields. But I feel strong nostalgia for pre-AI times. Many of the research skills that I spent decades learning are quickly losing value. If I could turn back the clock, I’d be perfectly happy doing my research in those pre-AI times. I’m one of those, mentioned by Tao, going through the five stages of grief.

SCI-AI hasn’t really started yet, but one thing I want to mention is this idea that AI can be a useful reminder to us humans of our own limitations. In this paper that I wrote together with Renne Pesonen from Tampere University, we argue that based on scientific findings (and conceptualizations) alone, there are no in-principle reasons why machines could not be intelligent, creative, or genuinely understand language.

Most of us have intuitions to the contrary, and I’m sure that the view we outline is not a popular one. But we hope it’s a discussion opener. I think that from the diversity point of view, AIs, these alien intelligences, provide us humans an opportunity to look in the mirror, and understand our own humanity more clearly. We’ve actually had superhuman cognitive capacities around for a long time (e.g. pocket calculators with their amazing capacity to multiply large numbers), but the fact that LLMs cracked at least the syntax problem for natural language really rubs the point in our face. We argue that whether impressive feats of LLMs mean that they are “really” intelligent or not isn’t all that relevant. What matters more, from an existential point of view, is that they offer us humans an opportunity to develop some epistemic humility, that is, the capacity to see both our intellectual strengths and weaknesses more clearly.

Thinking meat, crazy, right? (Terry Bisson 1991, worth checking out …)

(Cross-posted to leaflet.pub)

These policies outsource researcher evaluation to extra-academic agents, and result in absurd demands on academic review panels. If implemented, they can lead to significant harm to the quality of Finnish research and self-regulation of the academic community.

Here’s an AI summary (light edits):

Kuorikoski and Reijula argue that recent efforts to increase objectivity in academic hiring —- though well-intentioned – may actually threaten the quality and autonomy of science. They critique developments in Finland such as the Research Council of Finland’s guidelines on responsible evaluation and the University of Helsinki’s partial outsourcing of doctoral school admissions, which emphasize standardized and administratively controlled processes.

The authors introduce the concept of “HR-objectivity”, referring to a managerial model of objectivity that seeks to eliminate bias through formal, measurable criteria and external oversight. They contrast this with the epistemic objectivity of science, which relies on expert judgment, peer consensus, and tacit knowledge within research communities.

Kuorikoski and Reijula caution that treating academic evaluation like a typical HR process neglects the complexities of scientific quality and risks marginalizing expert intuition and contextual understanding. They emphasize that while fairness and transparency are important, over-reliance on procedural formalism can lead to superficial assessments and reduced academic freedom.

Ultimately, they argue that a misguided view of objectivity in hiring undermines science itself by eroding the community-based mechanisms that safeguard research excellence.

Here’s the link to the interview.

“If I had to pick one hypothesis to be tested in a real-world setting, I’d go with my work on bias and laziness” - says Matteo. Better find out what he’s talking about!

A manuscript version here

”

Perintövero on yhteiskunnallinen arvovalinta

Ehdotus perintöveron poistamisesta on herättänyt paljon keskustelua. Puheenvuoroissa on korostunut taloustieteellinen näkökulma. Sen sijaan oikeudenmukaisuuteen liittyviä kysymyksiä ei juurikaan ole nostettu esiin.

Perintövero voi tuntua epäreilulta kun se osuu omalle kohdalle. Veroa voisi sujuvoittaa monin tavoin, mutta sen korvaaminen luovutusvoittoverolla olisi askel kohti epätasa-arvoisempaa Suomea. Ruotsissa muutos on johtanut sosiaalisen liikkuvuuden heikentymiseen.

Perintövero on tärkeä yhteiskunnan rakenteen kannalta, sillä se jarruttaa eriarvoisuuden kasvua. Sillä tasataan varallisuuden ja vallan kertymistä harvojen käsiin ja autetaan takaamaan mahdollisimman tasa-arvoiset mahdollisuudet kaikille perhetaustasta riippumatta. Hiljattain The Economist -lehti myös varoitti, että perityn varallisuuden kasautuminen voi murentaa yrittäjyyden, innovaatioiden ja talouskasvun edellytyksiä.

Verotus on aina arvovalintoja. Mihin suuntaan haluamme viedä Suomea?

”

”

Julkishallinto pois pikaviestipalvelu X:stä

Miljardööri Elon Musk päättää X-pikaviestipalvelun tiedonvälitysperiaatteista. Valitsemalla X:n tiedotuskanavakseen julkishallinnon toimijat osoittavat hyväksyvänsä Muskin näkemyksen tiedonvälityksestä.

2010-luvulla sosiaalisen median pikaviestipalvelut avasivat uuden suoran kanavan politiikan, tutkijoiden, eri alojen asiantuntijoiden ja kansalaisten välille. Varjopuolistaan huolimatta sosiaalinen media on edelleen tärkeä suoran ja demokraattisen vuorovaikutuksen ja oppimisen väline. X on käyttäjämäärältään suurin pikaviestipalvelu. Musk osti Twitterin lokakuussa 2022 ja on sen jälkeen voimakkaasti muokannut alustan toimintaperiaatteita.

Julkishallinnon toimijoiden on aika poistua X:stä. Monissa tapauk­sissa palvelua tarjoavan yrityksen omistajan poliittiset näkemykset eivät ole ongelma sen käyttäjien kannalta, mutta X:n tapauksessa näyttää siltä, että Muskin poliittiset näkemykset vaikuttavat siihen, millaista sisältöä alustan algoritmi nostaa esiin. X ei ole neutraali viestintäkanava. Tästä seuraa että julkishallinnon X:n käyttö on poliittinen teko, eikä välttämättä linjassa arvojemme kanssa.

X:lle on olemassa hyviä vaihtoehtoja. Toisiinsa linkittyneet niin kutsutun fediversumin protokollat ja palvelut (muun muassa Mastodon, Bluesky ja jossain määrin Metan Threads) tarjoavat vapaan maailman vaihtoehdon, jossa käyttäjä voi omistaa omat sisältönsä eikä ole yhdenkään tietyn palveluntarjoajan armoilla. Fediversumin hajautettu rakenne tarjoaa julkishallinnon viestintään erinomaisen alustan. 

”

In our view, the striking thing about LLMs is that they are an existence proof of structurally simple (but large!) systems that can perform many of the tasks we think require intelligence. The fact that they might do those things in a “wrong”, non-human, way is beside the point: Something we thought could only be brought about by the (so far) unexplained powers of the human mind is now done by a shallow system. This should give us pause, some epistemic humility and antidote to human cognitive exceptionalism.

Preprint available here and here

Here’s the abstract:

Can large language models be considered intelligent? Arguments against this proposition often assume that genuine intelligence cannot exist without consciousness, understanding, or creative thinking. We discuss each of these roadblocks to machine intelligence and conclude that, in light of findings and conceptualizations in scientific research on these topics, none of them rule out the possibility of viewing current AI systems based on large language models as intelligent. We argue that consciousness is not relevant for AI, while creativity and understanding should be considered functional traits that, in principle, can be implemented in a machine. Many arguments in circulation that claim current AI systems necessarily lack these traits either rely on human exceptionalism or mistaken understanding of human intelligence. Arguments that highlight alleged flaws in current systems—such as a lack of reliability, agency, or understanding—may be important, but they often obscure the many qualitative similarities between human cognition and current AI models. We further suggest that a critical examination of high-performance AI systems can serve as a mirror through which we can reflect on our own intelligence. Many arguments against the prospects of AI may prove disconcerting for naively optimistic assessments of capacities of the human mind.

I’ve become increasingly interested in the fediverse. Thinking in terms of protocols instead of privately owned platforms could give us ways to address some of the epistemic challenges associated with the new AI-structured information landscape. My plan is to write something on this later on, but I’ll start by gradually compiling a reading list on both technical and more conceptual thinking on the fediverse. I’ve been asking people for recommendations on what to read.

The Fediverse is a new name for an old idea. From the beginning, the internet has been a network of networks, made possible by agreeing (only) on shared protocols used for communication between machines (the TCP/IP stack). Only in the past two decades or so we’ve become used to the social web being based on closed, proprietary platforms like Facebook or X/Twitter. The idea behind the Fediverse is to bring the social web back to a decentralized model: protocols like ActivityPub or Bluesky’s AT Protocol make inter-service sharing possible.

I have a couple of reasons why I care. The first one is personal. I’ve realised I have a long history with text-based social media - I was involved in the dial-up modems and bulletin board scene in the 90s. Maybe that’s why my reaction to social media is different from many others, I still get more joy than anxiety out of this. The reason I write about text-based content is because I guess I’m just too old for the short form video shit. I try it every now and then but it doesn’t interest me, I feel it’s not worth my time. I want to decide the pace at which I engage with information. I also have somewhat esoteric theoretical concerns about what will happen to civilization once we move away from propositional content, but I’ll spare you.

More importantly, I think fediverse is an interesting case of how institutions (& technology) matter for epistemology. It’s intellectually interesting, novel, and relevant. What excites me as an academic, is that social media, at its best, has created new direct channels between us researchers, policy-makes, and the public at large. Those are important connections. Social web for the free world.

Most importantly, this technology is highly political. It’s not all bad - remember the Arab Spring? The future of our democracy is not decided only at the ballot box, but as we’ve seen, technologies of attention and amplification of opinion do shape societal outcomes. At the time I’m writing this (2025-05-23), in Finland, many public authorities still use X as their communications platform. Politicians, all the way up to the president. I think it’s shameful, and a shame, we could do a lot better. Public authorities should not surrender to a lock-in platform governed by an obviously biased algorithm, fine-tuned to the whims of an unpredicatable billionaire.

The idea of fediverse

What is the fediverse? Fediverse is a (relatively abstract) model of how the social internet, or perhaps social media, can be organized. The most important conceptual move is to distinguish the concept of social media from the idea of a platform.

Email is a useful analogy: There is no “email platform”, and no one owns email. We don’t all use the same email service provider, but instead, mail moves between different services using the SMTP protocol that all service providers have agreed about. They share a language, and it’s that language that connects them into a network.

With fediverse, it’s the same thing. The fediverse protocols connect various nodes in a decentralized network. For example, you can host a Mastodon server or Bluesky node yourself, own your data, or at least choose between different providers, none of which control the entire network. And if you decide to move from server to another, you’re free to take your identity, data and connections with you.

On a slightly deeper level, I think fediverse is more than what the concept of social media captures. It’s an access layer to the content of the internet. Search engines emerged as the first solution to resolve the tension between the almost unlimited amount of information on the internet, and our limited attentional resources. (The speed of human cognition has been estimated as only 10 bits/s!)

Social internet, both the private platforms and the fediverse, provide a different solution to essentially the same problem of access. When posting and commenting, linking to content, we colletively curate the online content.

Middleware

An important aspect of decentralized architecture is that different functions in the network can be taken care of by different service providers. For example, the work of filtering content for readers/viewers can be done by parties separate from those who host the data, and the apps people use to access the content. This has interesting implications on moderation and the way people’s timelines are generated. It encourages people to build different views on the content posted on the network. In other words, there is no The Algorithm! Unlike private platforms, the fediverse is not an algorithmic monoculture, we’re not subject to one opaque algorithm, but instead, everyone gets to decide what shows up in their timeline. Bluesky has implemented this as Feeds, and I like it a lot.

And no ads (unless you use a service that has them), no microtargeting - no fear that by watching a random video one second too long you trigger a tsunami of similar content in your timeline.

Reputation, trust and truth

I’m convinced that identifying deepfakes (incl texts) based on content only is a race we will not win. As the detection algorithms become more advanced, so do the fakes. The amount of work needed to distinguish between real and fake will exhaust our resources. That’s why I think fighting epistemic erosion may require re-introducing some familiar accountability practices. Like an electronic version of the wax seal.

In science (and media, I suppose), reputation building is a career-long project. Reputation is often described as the capital of the scientific community. Being recognized as reliable and truthful is a necessary condition for establishing trust, and for having a good reputation within the community. Being listed as an author of a scientific article both gives you credit, but it also assigns a responsibility. If claims made in the article are false, it’s you, the authors, who are to blame. Note that in science, faking contributions by real scientists is a non-existent problem, and scientific misinformation - while real - is not quite as serious as issue as in some other domains. All this due to the hierarchic reputation system: we have an implicit understanding of what the good forums and who the reliable authors are, and both engage in reliable practices of content curation.

Something similar can be implemented on social media: Bluesky posts are already immutable objects signed with your user identity. If, in addition, there’s scarcity of identities - meaning you cannot generate more at will and at a negligible cost - we could create a credit economy similar to the one in science: your activities on social media contribute to your reputation, and all of us must care about our reputation in order to remain relevant. If you post or share deepfakes as real content, that’s going wreck your reputation and credibility. Reliable sigatures on content is a rudimentary accountability mechanism, I think we need it.

That said, all this comes at a cost to anonymity, which also is important in some situations. And it can be fun. But I guess we could still have anonymous accounts, as long as there’s a limited supply for generating them. As long as there’s an ecosystem where reliable identification can be differentiated from anonymous posts, trust can be calibrated based on those differences.

Why isn’t everyone on fedivers?

This one is easy. It’s a collective action problem: for now, the commercial platforms have an order of magnitude more users than their fediverse alternatives. For anyone with a large social network and social capital on those services, it doesn’t make sense to switch until others do the same.

I think this problem is less significant for public authorities, though, than for, say, private companies or individuals. Building a presence of fediverse, perhaps first on the side, as a complement to the locked-in platforms, can be seen as in investment in the infrastructure of democracy. (Taiwan seems like a really interesting model of this. I think it’s important to realize that using commercial platforms for communications was never a free lunch! That was just us fooling ourselves. The price is just paid in a different, more implicit and often unpredictable way. And by the way, I think fediverse should seem like a pretty good investment: unlike siloed platforms, the Fediverse won’t collapse if, and when, the company goes belly up. There is no “the company” in the fediverse!

The positive side of the collective action problem is that as the numbers of those making the switch increases, returns to everyone increase.

Loose ends

The design details matter. Only when the rules of interaction determined by the protocol are designed care, does the decentralized system lead to desired outcomes. For example, making sure corporate takeover of the network is not possible is something fediverse creators have been very mindful of.

One benefit of decentralization is that there can be actual people who run the server you use. So if you run into trouble, you can actually ask someone (ever contacted Facebook customer service when you have a problem?).

I have limited experience with Mastodon, but it seems to built on the premise that each of us has our community, one of them, in singular. It’s the instance of that community that you’re supposed to join on Mastodon. But I’ve got many communitities and social identities related to them, so I don’t want to make that choice. And I think most exciting things happen in the boundaries between those. Mastodon does not seem to be ideal for that kind of boundary crossing.

Right of exit and freedom of reach (Doctorow). The right of exit and the freedom of reach (the principle that anyone can talk to anyone who wants to talk to them) are both key to technological self-determination. (Doctorow Pluralistic: What the fediverse (does/n’t) solve (23 Dec 2022) – Pluralistic: Daily links from Cory Doctorow)

Resources

A hopeful take on social media: Zuckerman: Six or Seven Things Social Media Can Do For Democracy (2025-12)

MacKuba: A complete guide to Bluesky (Accessed 2025-12-21)

MacKuba: Introduction to AT Protocol

Christine Lemmer-Webber (2024-11-22). How decentralized is Bluesky really?

• highlights the/a? fundamental difference between ActivityPub and ATProto: AP delivers messages straight to receivers whereas ATP relies on relays that collect all content. So in ATProto it will be very costly to host relays, which makes real decentralization difficult

Pierce, David (2023). The poster’s guide to the internet of the future

• The POSSE model for the social internet

Kleppmann, M., Frazee, P., Gold, J., Graber, J., Holmgren, D., Ivy, D., … & Volpert, J. (2024, December). Bluesky and the at protocol: Usable decentralized social media. In Proceedings of the ACM Conext-2024 Workshop on the Decentralization of the Internet (pp. 1-7).

• looking forward to reading this one. good refs.

Bluesky CEO Jay Graeber interview at SXSW

• Super clear explanation of how fediverse principles guide bluesky development

Jay Graeber interview with Wired, 2025-05-19

Masnick 2019. Protocols, not platforms: A technological approach to free speech

• Although it’s getting old, this is still a gem, I’ve read it several times

Fukuyama et al. Middleware for dominant digital platforms: A technological solution to a threat to democracy

Jason Burton 2023 Algorithmic Amplification for Collective Intelligence

• excellent piece on the epistemic consequences of social media algorithms. great reference list too.

Journal of democracy issue on “The future of platform power”

Bluesky docs on A Self-Authenticating Social Protocol

Clear summary by @alexia.bsky.cyrneko.eu

Nil-Jana Akpinar, Sina Fazelpour. Authenticity and exclusion: social media algorithms and the dynamics of belonging in epistemic communities

McKelvey, Fenwick and Robert Gehl 2022. Canada’s public broadcaster should use Mastodon to provide a social media service

• Interesting piece that highlights the infrastructure aspect of fediverse: public authorities do not need to create their own platforms for their services. Instead, they can join and contribute the a decentralized shared social world based on the fediverse protocols

Podcasts & video

Waveform podcast: Protocol Wars - The Fediverse Explained!

• This was an excellent intro

Make Identity Central Again, with Bluesky’s Jay Graber

• Bluesky CEO interviewed, explains how bsky is implementing these ideas