At the Capgemini event last week, they had a scribe to capture details from discussions. Here’s what we got from my talk - scaling is not the answer, AI could offer huge potential to (say) NHS triaging, and the future is neuro-symbolic :-)

Looking forward to this event on the 23rd of September - I’ll be talking about LLMs and (if I can) neuro-symbolic AI

I'll be heading to Glasgow for a keynote at the "Playtime is Over GenAI" event by Capgemini. Title: "Large Language Models & Reasoning -- Bridging Symbolic and Neural Approaches". Thanks to Dave Ord for the invitation, and to Richard Veit.

At HAR in Paris, I have a short position paper on why logic and probability could jointly play an important role for cognition in virtual and physical robots.

Jessica and Chiara are in Santa Cruz for the neurosymbolic AI conference! Presenting (resp) our work on morality + verification and constraining ML models for counterfactual fairness via tensorised logical representations.

On the trails of a recent position paper on logic and its role for AI, journals.sagepub.com/doi/10.1177/... I now have an update at HAR in Paris on what specifically first-order logic can add to the picture.

Hello Göteborg! Neurosymbolic + linguistics + philosophy = gu-clasp.github.io/LARP/index.h...

LARP - Home gu-clasp.github.io/LARP/index.h... Heading to Sweden today for this talk. One of the key arguments is about the modeling of belief / theory of mind with LLMs, including symbolic executors, as well as the limitations of ascribing beliefs & anthropomorphism with these models.

Ambrose and I have a new paper at HAR in Paris: we look at LLMs to solve the so-called Wason Task, and outline steps a system would need to take to solve the task.

Excited to be in Sweden next week to give a keynote talk on large and small models in the neurosymbolic AI context. Event: Language models And RePresentations (LARP) organised by the Centre for Linguistic Theory and Studies in Probability, Uni of Gothenburg gu-clasp.github.io/LARP/index.h...

Pablo Barcelo and I have a paper in Jelia 2025, in Tblisi, Georgia, that we are not able to attend. Tommie Meyer will present our talk; slides below -- A Uniform Language for Safety, Robustness and Explainability -- doi.org/10.5281/zeno...

Slides from my Keynote talk at PFIA 2025 in Burgundy, France: pfia2025.u-bourgogne.fr/invites/ in July. -- Neuro-symbolic systems for Responsible AI -- doi.org/10.5281/zeno...

Chris and I have a new paper at HAR in Paris: We discuss how beliefs are understood w.r.t. LLMs, and why Anthropomorphism puts us off our footing.

Ruta and I have a new paper accepted at HAR in Paris: we try to quantify and qualify theory-of-mind reasoning abilities in LLMs!

Jessica and I have a new paper accepted at HAR in Paris: Its on integrating cultural preferences into reinforcement learning through the so-called Cross-Cultural Moral Reward Machine framework.

How can users effectively assess AI model predictions? We have a new study that explores combining uncertainty estimates with explanations in human-AI collaboration; considers interaction between model uncertainty, user confidence, & trust. doi.org/10.3389/fcom...

ML courses often overlook explaining a model's decision-making process, critical in high-stakes AI domains. We have a new paper that discussses how to go about a structured XAI course to equip students with vital interpretability techniques! doi.org/10.3389/fedu...

This opportunity is related to the Doing AI Differently white paper, where I'm also honored to have been one of the co-authors.

Interesting. Looking forward to reading it!

Drop in to explore the intersection of art and responsible AI at Edinburgh Art Festival's Tipping Point Exhibition, by our AHRC BRAID program. From Louise Ashcroft's "Real Stupidity" to Rachel Maclean's AI "They've Got Your Eyes!" & Arctic-inspired "Models of Care"

www.eventbrite.co.uk/e/bridging-r... Happening today: tipping point - chat with us about artistic pieces here and AI.

I'll be joining Bev Hood for Tipping Point/Authenticity Unmasked Tour - where we'll talk about technical insights & discussions about the art and implications for the field of responsible AI. If you are here for the Edinburgh Fringe, pop in inspace.ed.ac.uk/bridging-res...

Happy to be a co-author of the Doing AI Differently white paper at the The Alan Turing Institute. This is a call to action for meaningful change in AI, w. collective insights from arts & humanities. And includes a mention of neuro-symbolic AI :-). www.turing.ac.uk/news/publica...

media.licdn.com/dms/document... Have a look at our brochure on the executive education program on generative AI, fairness and regulation

I’m excited to be joining as the Director of Research and Innovation at the Bayes Centre. I’ll be 40% at the Bayes. I’m particularly looking to see how interest in generative AI and neurosymbolic AI could impact engagement with external stakeholders! bayes-centre.ed.ac.uk

Directions in AI, Logic, LLMs and Learning: our Lab’s research Overview

Our summary report on the Royal Society UK/Canada “frontiers in AI” event; Ottawa last year. www.facetsjournal.com/doi/10.1139/...

Reminder about our ExecEd offering in AI and GenAI covering goodies like bias and regulation too AI and Generative AI for Leaders Programme medium.com/@vaishakbell...

There was an owl hovering around at Dagstuhl. This is the parliament. Neurosymbolic AI & knowledge graphs

If you are heading to #ACL in Vienna, do chat with Ruta Tang — we have a paper on LLMs, context free grammars and evolutionary search.

On the Relevance of Logic for Artificial Intelligence, and the Promise of Neurosymbolic Learning - my position paper on the flavors and depth of logic for general purpose AI doi.org/10.1177/2949...

Jessica Ciupa has a new paper accepted on "exploring verification frameworks for social choice alignment" at the Neurosymbolic Conference, 2025. This is one of the late-breaking papers. vaishakbelle.org/papers

So kicked that our children’s book - the girl and the robot - is fully pledged. It explores a range of topics from friendship to projecting feelings on artificial agents.

Ute Schmid on Donald Michie’s ultra strong machine learning #dagstuhl

Ute Schmid on the origins of the neuro-symbolic argument

Artur Garcez at Dagstuhl on neurosymbolic AI

Nicholas Papernot and I co-chaired the Royal Society UK+Canada event last year. With invited talks by Yoshua Bengio and others, read our summary article in the facets journal

The Dagstuhl castle seminar on neuro-symbolic AI and knowledge graphs bit.ly/4lLgKIG

Pablo Barcelo and I have a new paper at JELIA aiming to unify the definitions of fairness, robustness, and explainability, including abductive explanations as well as semantic loss. “A Uniform Language for Safety, Robustness and explainability.”

Parakeet Books were on the book club podcast and talk about our new children’s book “the girl and the robot”

A new @facetsjournal.bsky.social report, from @vaishakbelle.bsky.social and @nicolaspapernot.bsky.social (RSC College), briefly summarizes the objectives and outcomes from the 2024 Frontiers of Science: Artificial Intelligence Conference, which was held in Ottawa. 🔗 Find out more: bit.ly/3Ij2Q1X

New 📰| UK/Canada Frontiers of Science: Artificial Intelligence Report ✨ buff.ly/Dy3olJS @vaishakbelle.bsky.social & @nicolaspapernot.bsky.socia This report briefly summarizes the objectives and outcomes from the Frontiers of Science: Artificial Intelligence Conference. 📷 from @src-rsc.bsky.social

Next week I’ll be in Germany for the Dagstuhl Seminar on Neurosymbolic AI and Knowledge Graphs See you there if you are coming

Morality & AI: what’s to be said? Our Lab’s perspective—

Interested in GANs and in logic, and how to bridge them? Nijesh has a new paper at ICLP on how to use tensorized logical formulas. "Logic Tensor Network-Enhanced Generative Adversarial Network”:

We have a new result on abduction (guessing explanations for observations) + epistemic modal logic, accepted at ICLP. With Sanderson Molick. "Propositional Abduction via Only-Knowing: A Non-Monotonic Approach."

What we need are not LLM-based reviews, called “perspectives” by the misguided. Rather, we need to be stopping reviewers from using LLMs to do their work. So build agents to spot that.

Modeling the non-quantifiable harms in responsible AI is challenging - informal language is both a barrier, and a bridge.

My slide for where neurosymbolic AI could / should go, involving first order, dynamic and modal logics, for modeling beliefs, norms and causality.

Was lovely to connect & reconnect with Francois, Lydia, Anaelle, Henri and many others at the platforme AI: pfia2025.u-bourgogne.fr Slides coming up.