iiinova — where your digital self keeps evolving

Reposted by Szymon Miłkoś

Cailin O’Connor @cailinmeister.bsky.social · 06/07/2026

Philosophers! Craig Callender and I are circulating an open letter calling for better COI disclosure in our journals. If tech companies fund and employ philosophers working on AI, we need to know and to be thinking about the impacts 1 docs.google.com/forms/d/e/1F...

docs.google.com

Open Letter about Industry Funding and Disclosure in Philosophy

*For full list of signatories, updated every few hours, please click here.* In recent years, philosophical writing on areas such as climate change, technology, misinformation, and artificial intellige...

14733

Szymon Miłkoś @milkos.bsky.social · 02/07/2026

For sure, we should not place our trust based on the domain of science, but on the rigour of how results are obtained. If science is not trustworthy, what is? ofc, the best case would be a society with research literacy that allow them to "reasonable doubt" the science they approach.

020

Szymon Miłkoś @milkos.bsky.social · 02/07/2026

But this is real problem in big or interdisciplinary research - there is no single author who gets it’s whole, nor single reviewer who can asses.

120

Szymon Miłkoś @milkos.bsky.social · 26/06/2026

Yet, there are better and worse ways for conference discussion. Moreover, there is still a room for some kind of engineering - like conference apps where people can add and vote for questions. Nevertheless, there should be not strict rules - at the frontier of inquire, the way forward is not settled

200

Szymon Miłkoś @milkos.bsky.social · 25/05/2026

The problem is that this debunking is not recognized as „proper” scientific contribution. In most fields, there is expectation for novel or own answer, not merely critique

010

Szymon Miłkoś @milkos.bsky.social · 30/10/2025

It’s evidence for the claim that language is primarily a tool, and as such follows metacognitive optimization, efficiency in this case. For me, my native polish don’t depends on word ordering, therefore one can change the meaning of the sentence on the fly, which is helpful in oral discussions.

010

Reposted by Szymon Miłkoś

Julia M. Rohrer @dingdingpeng.the100.ci · 05/05/2025

This Thursday I will give a talk on (causal) mediation analysis -- happy to have finally worked out what I want to tell people. And happy to pilot a new mode of slide sharing: just putting them on my website (juliarohrer.com/resources/)

Title slide: Trust the process? (Causal) Mediation analysis

Decision flowchart "So you want to conduct a mediation analysis"

1314633

Szymon Miłkoś @milkos.bsky.social · 11/04/2025

CI is great to defend and criticize methodological ideas. Similarly to metascientific findings, it often shows that actual science is bad and good science is hard. Purely statistical work, and offloading causal conclusion to reader is easier. Actual science condition is part of adoption problem.

110

Reposted by Szymon Miłkoś

Julia M. Rohrer @dingdingpeng.the100.ci · 20/03/2025

New blog post! In which I explain the issue with mediation analysis and sketch out one way to deal with the underlying causal inference problem -- in just a bit over 1,000 words! If you have never found the time to read up on this, now is your chance. www.the100.ci/2025/03/20/r...

the100.ci

Reviewer notes: That’s a very nice mediation analysis you have there. It would be a shame if something happened to it.

Mediation analysis has gotten a lot of flak, including classic titles such as “Yes, but what’s the mechanism? (Don’t expect an easy answer)” (Bullock et al., 2010), “What mediation analysis can (not) ...

26452138

Szymon Miłkoś @milkos.bsky.social · 14/03/2025

You mean knowing part of mechanisms constituting object of inquire is not sufficient. Agree. But it is necessary, although I have wider concept of mechanistic knowledge, probably closer to yours concept of some „theory” (not the theory!). Like knowing the symptoms, not failing selection bias, etc.

000

Szymon Miłkoś @milkos.bsky.social · 14/03/2025

Any example of causality without mechanistic knowledge?

100

Szymon Miłkoś @milkos.bsky.social · 14/03/2025

But how can you have better evidence without mechanisms? I mean, how do you know that in fact some empirical evidence is better w/o even minimal mechanistic knowledge?

210

Szymon Miłkoś @milkos.bsky.social · 14/03/2025

This is practically but not theoretically correct. If one would have fully specified mechanisms in form of structural causal model, even in the case of fat handed intervention one would predict correctly. The problem is that we have such knowledge only for constructed things (see causal chambers)

010

Szymon Miłkoś @milkos.bsky.social · 14/03/2025

@hayoungsong.bsky.social gz! What about negative „aha” like spotting the contradictions? Moreover, if we would combine both negative and positive aha in study, I expect the stronger the contradiction, the stronger the positive insights that would resolve it. Would be lovely to se it!

000

Szymon Miłkoś @milkos.bsky.social · 14/03/2025

This is another partial evidence to more general claim: insights are signals about significant metaproblem change about problem (here re representing elements of narratives)

100

Reposted by Szymon Miłkoś

Carl T. Bergstrom @carlbergstrom.com · 06/03/2025

Let's talk about this Nature piece in more detail. I've rarely read something so anti-scientific anywhere short of the National Review. www.nature.com/articles/d41...

nature.com

Three AI-powered steps to faster, smarter peer review

Tired of spending countless hours on peer reviews? An AI-assisted workflow could help.

621622688

Reposted by Szymon Miłkoś

Juan Mateos-Garcia @jmateosgarcia.bsky.social · 26/11/2024

AI for science could be more impactful than chatbots. It is already helping win Nobel prizes and accelerating drug development and materials discovery. Today we published an essay about it: why it matters, how it’s happening and its implications. Here is a summary from an econ / social sci lens.

27930

Reposted by Szymon Miłkoś

Iris van Rooij 💭 @irisvanrooij.bsky.social · 16/02/2025

How hard is cognitive science? 🎬📽🍿 Video: m.youtube.com/watch?v=2bdK... 📖 Paper version: psyarxiv.com/k79nv/ Summary in #PaperThread below 🧵 1/n

m.youtube.com

How hard is cognitive science?

YouTube video by Iris van Rooij

612529

Reposted by Szymon Miłkoś

Paul Hünermund @p-hunermund.com · 17/02/2025

More evidence that peer review penalizes academic risk-taking from a new paper Pierre Azoulay and Wesley H. Greenblatt: "Does Peer Review Penalize Scientific Risk Taking? Evidence from NIH Grant Renewals." www.nber.org/papers/w33495

0163

Reposted by Szymon Miłkoś

Berna Devezer @devezer.bsky.social · 11/02/2025

A revised version of "Minimum Viable Experiment to Replicate" is up, where we expound why standard expectations from replications are unrealistic and experiments that may deliver on those replications are rare, if not nonexistent. #metasci #sts #philsci philsci-archive.pitt.edu/24720/7/Mini...

Minimum Viable Experiment to Replicate
Berna Devezer and Erkan O. Buzbas
Department of Business, University of Idaho
Department of Mathematics and Statistical Science, University of Idaho

Abstract: In theory, replication experiments purport to independently
validate claims from previous research or provide some diagnostic evidence about their truth value. In practice, this value
of replication experiments is often taken for granted. Our research shows that in replication experiments, practice often does
not live up to theory. Most replication experiments involve confounding factors and their results are not uniquely determined
by the treatment of interest, hence are uninterpretable. These
results can be driven by the true data generating mechanism,
limitations of the original experimental design, discrepancies
between the original and the replication experiment, distinct
limitations of the replication experiment, or combinations of any
of these factors. Here we introduce the notion of minimum viable experiment to replicate which defines experimental conditions that always yield interpretable replication results and is
replication-ready. We believe that most reported experiments
are not replication-ready and before striving to replicate a given
result, we need theoretical precision in or systematic exploration
of the experimental space to discover empirical regularities.

610832

Szymon Miłkoś @milkos.bsky.social · 06/02/2025

True, the link between causal models and their target systems remains fuzzy in causality. Not sure we can grasp this framework from stochastic thermodynamics without mastering the framework foundations first. Getting into causality is though.

100

Reposted by Szymon Miłkoś

Mark Rubin @markrubin.bsky.social · 06/02/2025

"Reliability and validity are important properties for research methods to have. Yet, we also show that reliability and validity fall short of other epistemic virtues that are crucial to the quality of research methods" (Ventura, 2025). doi.org/10.1007/s112... #Methodology #MetaSci #PhilSci

How should we measure the quality of experimental research? With talk of a looming “replicability crisis”, this question has gained additional significance. Yet, common measures of research quality based on reliability and validity do not always track core epistemic virtues. To remedy this issue, we draw on information theory and propose a measure of research quality based on mutual information. Mutual information measures how much information an experimental method carries about the world. We show that this measure tracks epistemic virtues that reliability and validity do not. We conclude by discussing implications of this information-theoretic measure of research quality and address some limitations of this approach.

2365

Reposted by Szymon Miłkoś

Kristian Kersting @kerstingaiml.bsky.social · 24/01/2025

Thrilled to share our #ICLR2025 work on Meta-Causal States! 🌟 Causal graphs evolve with dynamic systems & agent actions. We show how to cluster causal models by qualitative behavior, revealing hidden dynamics & emergent relationships 🚀 #Causality #ML arxiv.org/abs/2410.13054

0126

Szymon Miłkoś @milkos.bsky.social · 23/01/2025

Do you mean that if Y is caused by $X then difference between changing $ or X is meaningful only in dynamic systems ? Changing parameter is changing causal relationships on all phenomenon level, changing X is changing its state on units level. They are not same, even with same numerical effect.

000

Reposted by Szymon Miłkoś

Soazig Le Bihan @soaziglebihan.bsky.social · 23/01/2025

Should science communication be only about its societal benefits or should it be about the proper understanding of the nature of scientific practices? This article looks into how scientists answer that question: link.springer.com/article/10.1...

link.springer.com

How Scientists Perceive NOS and Its Value for Science Communication - Science & Education

A primary goal of science education and communication is to promote a functional scientific literacy that enables people to efficaciously engage with socioscientific issues (SSI), such as COVID-19 and...

184

Szymon Miłkoś @milkos.bsky.social · 27/12/2024

It is acceptable, but functional opacity has to be countered by external evaluation of the outcomes. For example competent judges should grade random 10% of the outcomes.

010

Szymon Miłkoś @milkos.bsky.social · 06/12/2024

Agree. I was not saying that it’s wrongly selling itself as something more.

020

Szymon Miłkoś @milkos.bsky.social · 05/12/2024

This is because causal framework is selling itself as something more then statistic, while basic argument is that it could provide experimental results without them. Nevertheless, there is a lot of work dealing with both experimental and observational data

110

Szymon Miłkoś @milkos.bsky.social · 03/12/2024

Now I get it. You mean HB paradigm as way of doing behavior economy. I was thinking about HB paradigm as program in cognitive science. You probably know the Anderson Rationality analysis regarding HB?

110

Szymon Miłkoś @milkos.bsky.social · 03/12/2024

But maybe I don’t understand what you mean by HB program, and why impossibility of finite set is so devastating to such program

100

Szymon Miłkoś @milkos.bsky.social · 03/12/2024

You say that HB will fall under its own weights because HB is not finite. However, notice that your argument could also support the claim: the HB changes on the go. So it’s not finite because it’s dynamical, and also the space of possible HB is changing.

100

Szymon Miłkoś @milkos.bsky.social · 03/12/2024

Moreover the test and train set seemed very similar, so the questions of test distribution and overfit may make the competition obsolete

020

Szymon Miłkoś @milkos.bsky.social · 03/12/2024

The logic of this comp is weird. They have the same method applied to all 25k test sets. It’s not a typical application, where we usually know smtg about the object of inquire and could choose algorithm that fit best to our problem. They measure best universal naive method, not real problem solving.

320

Szymon Miłkoś @milkos.bsky.social · 02/12/2024

Dude. But what about dynamic set of heuristics? We crate them and change them on the go.

100

Szymon Miłkoś @milkos.bsky.social · 02/12/2024

Meta (cognition) is intertwined self-monitoring and self-control. I think first-order science have meta-scientific elements: we are „just” specializing. I think that positive feedback loop can be very powerful for science advancement - yet hard to see, as within first order practice itself.

100

Szymon Miłkoś @milkos.bsky.social · 28/11/2024

Doing things that impact empirical research!

020

Szymon Miłkoś @milkos.bsky.social · 28/11/2024

Speaking as philosopher and social scientist and causal methodologists: clarifying what proper A and B is; criticizing the way to measure/intervene on them; picking/idealizing right phenomena; reassuring that it’s harder job then physics (I studied that too!); nudging them to have better theory.

010

Szymon Miłkoś @milkos.bsky.social · 25/11/2024

I think the @jakobrunge.bsky.social causal inference for time series is great. The very aim is to bridge methodological and domain specialist language. Also it build on Pearl Causal Hierarchy paper which is great intro to fundamental ideas. www.nature.com/articles/s43...

nature.com

Causal inference for time series - Nature Reviews Earth & Environment

Earth sciences often investigate the causal relationships between processes and events, but there is confusion about the correct use of methods to learn these relationships from data. This Techni...

031

Reposted by Szymon Miłkoś

Carl T. Bergstrom @carlbergstrom.com · 13/08/2024

Taylorism is a management philosophy based on using scientific optimization to maximize labor productivity and economic efficiency. Here's the result of making the false Taylorist assumption that the output of scientific research is scientific papers—the more, faster, and cheaper, the better.

The AI Scientist: Towards Fully Automated
Open-Ended Scientific Discovery
Chris Lu1,2,*, Cong Lu3,4,*, Robert Tjarko Lange1,*, Jakob Foerster2,, Jeff Clune3,4,5, and David Ha1,
*Equal Contribution, 1Sakana AI, 2FLAIR, University of Oxford, 3University of British Columbia, 4Vector Institute, 5Canada CIFAR
AI Chair, Equal Advising
One of the grand challenges of artificial general intelligence is developing agents capable of conducting
scientific research and discovering new knowledge. While frontier models have already been used as aids
to human scientists, e.g. for brainstorming ideas, writing code, or prediction tasks, they still conduct
only a small part of the scientific process. This paper presents the first comprehensive framework for
fully automatic scientific discovery, enabling frontier large language models (LLMs) to perform research
independently and communicate their findings. We introduce The AI Scientist, which generates
novel research ideas, writes code....

21410164

Reposted by Szymon Miłkoś

Marcin Miłkowski @mmilkowski.bsky.social · 24/10/2024

We’re excited to announce the addition of the pure text version to the MDPI Open Peer Review Corpus 2!

151

Reposted by Szymon Miłkoś

Kristoffer Magnusson @rpsychologist.com · 19/11/2024

Me and @larsklintwall.bsky.social have been working on an app that will allow participants to interactively build networks of perceived causal relations. I’ve put together a demo website, that I’m soft-launching while we continue to pilot the app pecan-tool.rpsychologist.com 🧵👇

pecan-tool.rpsychologist.com

Interactive Perceived Causal Problem Networks (PECAN) Demo

73511