Joe Barnby

Very happy to announce next week's SoCR lab guest, Andreas Hula, who will be talking about the uses and misuses of I-POMDPs in model-building human social decision-making 🧵

@catiamoliveira.bsky.social @alexandrapike.bsky.social

We are incredibly grateful to @wellcometrust.bsky.social, @kingsioppn.bsky.social, University of Western Australia, and @perroninstitute.bsky.social for supporting the vision.

If you're into computational modelling, mental health, or UX, watch out for the application links.

The new funding will expand our team, develop our cognitive fingerprinting tech, and fund essential validation.

We’re partnering with Emyria (WA), @orygen.org.au (VIC), and C-FIT (KCL; London) to build a seamless clinical UXUI and a platform ready for treatment development.

Announcement from Hypatia regarding second-round investment from the Wellcome Trust and 2026 hiring for five roles: Project Coordinator, Backend Engineer, UXUI Engineer, Artist/Designer, and Director of Operations. The graphic showcases Hypatia's international presence in the UK and Australia, featuring team photos and logos of partner institutions including King's College London, UWA, C-FIT, Orygen, and Emyria.

Excited to say that Hypatia has secured a second round of funding thanks to the @wellcometrust.bsky.social 💫

This milestone allows us to build the technology and team and conduct validation efforts across the UK and Australia to bridge the ecosystems.

Work put together with the excellent @nitalon.bsky.social @stefansarkadi.bsky.social Lion Schulz Jeff Rosenschein & Peter Dayan

ipomdp firing with random noise

The model offers another multi-agent route to paranoia alongside hypermentalising.

If the anomaly detector is tuned too tightly, it yields false +ves. The agent becomes overly defensive against benign peers.

This connects generalist accounts of delusions to social contexts.

With this, more equitable social outcomes are achieved, and exploitation is reduced when compression & expectation are tuned within normal bounds. It provides a plausible, mechanistic formal framework in which agents can compete without recursion.

It augments agents with and ℵ-IPOMDP mechanism:

1. Anomaly Detection mechanisms: "My predictions are consistently wrong."

2. Out-of-Belief Policy: A severely 'defensive' response.

In effect: 'I don't know _how_ you're tricking me, but I know you are, so stop or we'll both suffer'

In recursive modeling (I think that you think...), agents eventually hit a cognitive limit. It’s a logical trap - humans can't just "think harder" forever.

It might also be the case that humans infrequently want to avoid using recursive strategies when they can -

Two side-by-side images depicting the nested hierarchical IPOMDP and the non-hierarchical x-IPOMDP mechanism.

What happens when we can't use recursive belief to compete? We can use anomaly detection instead!

Here, we (led by soon-to-be-Dr @nitalon.bsky.social ) devise a multi-agent account where compression & reward expectation are used to notice deception

jair.org/index.php/ja...

Talk Poster, Prof. Marco Wittmann, 19/02/2026 @ 9am GMT

It's a pleasure for me to announce our next Social Computation and Representation Lab invited speaker @mkwittmann.bsky.social for a talk on dimensionality reduction and basis functions in social cognition

Here we go! #ToM4AI2026 @aaai.org - thanks @reuth-mirsky.bsky.social for the hats 🎩

www.tom4ai.GitHub.io

@nitalon.bsky.social @stefansarkadi.bsky.social

NEPTUNE project logo with affiliations & funders

New job alert 💫 We’re hiring a 3-year postdoc for the NEPTUNE project to study the causal mechanisms of paranoia and social learning.

Work with us on experimental psychopharmacology (THC), social cognition, and psychosis 🧑‍🔬

Apply here: lnkd.in/gQqnNvjR](my.corehr.com/pls/kclrecru...)

Please RT :)

Poster showing the schedule for Theory of Mind in AI workshop at AAAI 2026, featuring 3 keynote speakers, a hackathon, posters, and flash talks. Sponsored by Edith Cowan University and Ketryx

Full details for the 2026 ToM4AI Workshop @aaai.org are now live (ToM4AI.com)

100+ registrations, 40 papers, 3 keynotes, & a hackathon, spanning cog sci, philosophy, psychiatry, computer science, and robotics

@nitalon.bsky.social @stefansarkadi.bsky.social @reuth-mirsky.bsky.social

Would be fun to chat about whether your data also conforms to our model of self-other generalisation... its in the correct format (and uses a fehr-schmidt function as in ours) so it would be quite straght forward to fit!

The utility function captures the latent/type of partner within a phase, but I meant between phases. The way we formulate it is to say that representations of self and other have some degree of porosity, so the latents that define each transfer a bit between phases.

We found a very similar thing! you mention that an RL generalisation model didn’t quite fit vs the utility model, but did you try the utility model with a generalisation component? Would be keen to see if your data fitted our model well

elifesciences.org/articles/104...

Yes I think this complexity is certainly front-and-centre in our appraisal of the results, although we do see some (tentative) evidence of symptom specificity. We're digging into this further.

Yes that would be really satisfying, both if this model could distinguish therapies or track reported improvement in interpersonal satisfaction - I suspect the story will be far more complex and against my best predictions, but one can hope :)

My prediction would be that, like our Elife paper, raw predictive accuracy would not change, but the process to get there would be more integrated (with successful therapy, folks would become less model M4 and more model M1, as defined in the paper) - we're working on testing this 👀

Yes thats true. Concerning Autism and I defer to @leoschilbach.bsky.social for a more detail, but my analysis is that neurotype communication is successful within-group, and only in autistic/non-austistic pairs do you see mentalising discrepancies (see journals.sagepub.com/doi/pdf/10.1...)

Thanks for sharing! Of interest, we note that those with a diag. of BPD were no less accurate at predicting the future actions of others; they just used a different method to get there. We argue that this difference might also raise the risk of instability in interpersonal bonds.

Curious why we sometimes see minds where there are none? 🧠 Join us to uncover the foundations of attributed agency.

Fully funded PhD for next year as part of DRIVE-Health, working with me, Adam Hampshire & @stefansarkadi.bsky.social

Deadline: 12/1/26
Reach out for an informal chat!

Curious why we sometimes see minds where there are none? 🧠 Join us to uncover the foundations of attributed agency.

Fully funded PhD for next year as part of DRIVE-Health, working with me, Adam Hampshire & @stefansarkadi.bsky.social

Deadline: 12/1/26
Reach out for an informal chat!

💭 In the first interview for Inside Discovery, @joebarnby.com discusses how deeply human questions about our social world can be explored through elegant, technical science.

Read the full conversation: buff.ly/CKx4k5c

Excited to receive 57 paper submissions for our 2026 workshop 🙌 Stoked about the interest in this space!

We also have 4 free student regos to award; tough choices ahead with many worthy authors

@reuth-mirsky.bsky.social @nitalon.bsky.social @stefansarkadi.bsky.social

tom4ai.github.io

I am pleased to announce that @davidschultner.bsky.social will be our next guest as part of the SoCR Lab's Invited Talk Series. Dr. Schultner will present recent work that offers evidence for a parsimonious and mechanistic explanation of human social learning strategies via reward learning!

Just one more week to get submissions in for this!

Join us in Singapore to discuss the intersection of cog sci, comp. sci, psychiatry, and philosophy to understand Theory of Mind 🧠

DM me for informal qs 📜

@stefansarkadi.bsky.social @nitalon.bsky.social @reuth-mirsky.bsky.social

1/6 Excited to see our NMH paper featured here for #WorldMentalHealthDay! rdcu.be/dYl7T

Many psychiatric symptoms are inherently social. @xiaosigu.bsky.social, @joebarnby.com, & I call for cognitive models that reflect processes most relevant to our questions, esp. for those about social symptoms🧵