A clearer mental model for the statistical / symbolic divide
An exploration of how statistical and symbolic systems differ in the way they process and reason and what that distinction actually means for how AI systems behave in practice.
.png?width=2000&height=500&name=Thinking%20Fast%20and%20Slow%20in%20AI%20(7).png "Thinking Fast and Slow in AI (7)")
AI has always been inspired by how humans think.
Now, researchers designing AI reasoning systems are turning to Kahneman's fast/slow framework as a blueprint. But is that really how humans think and is it the right way to design AI?
Callum Hackett explores what human psychology genuinely has to tell us.
🕐 Thursday 21st May
📍Online - no cost to attend
🎫 Hosted by Callum Hackett, Research Science Lead and Louis Mahon, Senior Applied Scientist
What you’ll learn from the webinar
A more critical lens on vendor claims
A grounding in what terms like "hybrid AI" and "neurosymbolic" actually refer to so you can engage with vendor conversations more precisely, and ask better questions.
A more considered view of where LLMs may be the wrong tool
Particularly in regulated workflows where a confident-sounding but incorrect output carries real compliance cost and where the architecture itself may be the issue.
A way of thinking about whether AI reasoning is genuinely structured
Some tools are fluent without being reliable. This session explores what it actually means for reasoning to be structured and why that distinction matters for how you evaluate AI systems.
Why attend?
Think more rigorously about AI in 2026 by interrogating the most popular framing
If you want to understand what hybrid AI actually means
Callum will explore what terms like "neurosymbolic" and "hybrid AI" actually refer to under the hood and why the distinction matters for decisions you're already making.
If LLM reliability is a concern in your organisation
The session explores which types of tasks LLMs handle well, where they tend to produce confident-sounding errors, and what a more structured alternative might look like.
If you're curious whether AI has always been inspired by human thinking
From neurons to neural networks. The System 1/System 2 framework is the latest chapter in that relationship. This webinar asks whether it's the right one.
If you work in or around regulated industries
The session focuses on what a thoughtful division of responsibility between AI techniques looks like when compliance and auditability are non-negotiable requirements.
If you want a richer internal conversation about AI
Engaging with this framing critically, rather than accepting it at face value, tends to lead to better questions internally and better conversations with vendors.
If you'd rather think than be sold to
This is a research scientist exploring a concept, not a product demonstration. The audience is enterprise leaders who want rigour and that shapes the conversation.
Thinking about AI reasoning through the lens of human psychology and asking how well it holds up
Psychologist Daniel Kahneman's framework describes two modes of human thinking: System 1, which is fast, automatic, and pattern-driven; and System 2, which is slow, deliberate, and logical. Applied to AI, the parallel is intuitive — LLMs as System 1 intuition, symbolic reasoning as System 2 logic. It resonates because it captures something real about how these technologies differ.
But the analogy has limits. Human cognition isn't as cleanly divided as Kahneman's model suggests, and AI systems are even less so. In this webinar, Callum Hackett explores where the framework genuinely illuminates how AI works, where it starts to obscure more than it reveals, and what a more precise way of thinking about the division of labour in modern AI systems might look like.
An hour of thinking, not selling
Enterprise leaders who want to engage seriously with how AI actually works, not be pitched at, are who this session is designed for.
About the speakers
Callum Hackett
Research Science Lead, UnlikelyAI
Callum Hackett leads research science at UnlikelyAI, where he works on the architecture of hybrid AI systems that combine large language models with symbolic reasoning. His work focuses on the practical question of where each technique belongs in high-stakes, structured workflows — particularly in regulated industries where fluent-sounding errors carry real consequences.
An Oxford alumni linguistics and AI researcher, Callum's work integrates statistical and symbolic reasoning, publishes evaluation approaches, and helps enterprises judge fitness for purpose against their own workflows. This session builds directly on the Trust Gap event, extending the conversation toward what a principled division of labour inside modern AI systems should actually look like.
.png?width=1414&height=2000&name=Untitled%20design%20(6).png "Callum Hackett")
Louis Mahon
Senior Applied Scientist, UnlikelyAI
An Oxford-trained ML researcher with a PhD and 9 years of experience across academia, industry, and nonprofit work, specialising in large-scale model training, multimodal systems, and signal processing, with 20+ peer-reviewed papers in top venues including ICLR, ACL, and AAAI.
Technically fluent across the full ML stack and distributed infrastructure, with hands-on applied work ranging from dictionary-linking tools at Oxford University Press to high-precision AI for regulated domains. Combines rigorous research with a track record of real-world impact, including leading pro bono data science teams for educational charities in Tamil Nadu.
Who this session is for
This session is for enterprise leaders who are genuinely curious about technology and who want to think more rigorously about AI, not be sold a version of it.
- You're making or influencing AI decisions in your organisation
- You want to understand AI more deeply, not just follow the consensus
- You're sceptical of vendor framing and want a more grounded perspective
- You work in or around regulated industries where AI reliability matters
- No technical background required — curiosity is the only prerequisite
Register before places fill
Seats are limited. Senior leaders are already confirming attendance.
Agenda
What will happen on the day
Welcome & framing
Introduction to Kahneman's System 1/System 2 framework and why it has become a useful lens in AI research and practice.
Part 1: Why the AI community has reached for Kahneman's framework
What the fast/slow distinction is trying to explain about modern AI systems and why it has become a common way of talking about how statistical and symbolic reasoning differ.
Part 2: Why the analogy breaks down
The Kahneman framework maps onto the startistical/symbolic divide — but that divide is only one part of how AI reasoning systems actually work. Callum explores what the fast/slow framing obscures, and why treating it as a complete picture leads teams to the wrong architectural decisions.
Part 3: Designing the right division of labour
If the Kahneman framing isn't quite right, what is? Callum explores how to think more precisely about which parts of a task should be handled by which kind of AI, particularly in regulated environments where the cost of getting it wrong is high.
Live Q&A
Ten minutes of open questions from the audience directly to Callum Hackett.
About UnlikelyAI
UnlikelyAI is a UK pioneer in AI Innovation and an award-winning UK startup. The unique neurosymbolic AI technology that we have developed is exceptionally precise and explainable, trusted by Lloyds Banking Group and other global partners.
To advance research into dependable enterprise AI, we established the UnlikelyAI Lab. This is a dedicated research group led by Oxford academic Callum Hackett, a linguistics and AI researcher known for his work on explainable technology. Our research integrates LLMs + symbolic reasoning and publishes evaluation approaches and result summaries so enterprises can judge fitness for purpose against their own workflows.
Our mission is bridging the trust gap between human expertise and artificial intelligence, solving one of the most critical challenges in AI.
.jpg?width=2000&height=1500&name=IMG_1020%20(7).jpg "IMG_1020 (7)")