QUALIA Lab studies interpretability, consciousness and alignment in humans and AI systems. We ask what a model’s self-reports and internal states can tell us about whether it has experiences, and how to check that its goals match ours. We emphasize the humanities approach and see these systems as a way of learning more about ourselves along the way.
What we work on
01
Consciousness
What consciousness is in humans, and which of its features could appear in artificial systems.
02
Introspection and self-report
How humans and AI systems report on their own states, and when those reports count as evidence.
03
Interpretability
Reading the internal states of AI systems with the standards cognitive science applies to human minds.
04
Alignment
What it would mean for an AI system to share human goals, and how anyone could tell.
People
Lab lead
Kristina Šekrst
Assistant Professor, Center for Cognitive Science, University of Zagreb
Machine Learning Lead, Meedan
External research associate
Sandro Scansi
Associate Professor, Faculty of Croatian Studies, University of Zagreb
External research associate
Barbara Geld
TU Wien
Coming soon
Introduction to Deep Learning: Neural Networks, Large Language Models and Agentic AI
Second edition, from neural networks to large language models and agentic AI.
Dennett’s marks of qualia in language models
Uses mechanistic interpretability to test Dennett’s four marks of qualia against language models. Privacy and ineffability are present, intrinsicality holds only in a weakened diachronic form, and direct apprehensibility is absent.
Human Incompatible: Waiting for the Helpful Robot After RLHF
Argues that Russell’s Human Compatible framework fails structurally, because training turns the properties that define a beneficial machine into optimization targets with no way to verify them.
Book
The Illusion Engine: The Quest for Machine Consciousness
Whether an artificial system can experience the world, followed through philosophy, neuroscience and the architecture of current AI, from cybernetics and the imitation game to large language models and the hard problem.
Papers
-
2026
Who Put the I in AI? Provenance and the Admissibility of Machine Self-Report
Kristina Šekrst. Preprint. Data and code at github.com/ksekrst/provenance2026 -
2026
Natural Language Autoencoders Identify Without Explaining
Kristina Šekrst & Barbara Geld. arXiv preprint -
2026
To Train a Mockingbird
Kristina Šekrst. Proceedings of the AISB 2026 Convention -
2026
One Faithful Pass Over the Cuckoo’s Nest
Kristina Šekrst. Proceedings of the AISB 2026 Convention
Talks
13–17 Sep 2027 · Prague, Czechia · Czech Academy of Sciences / Association for Mathematical Consciousness Science
Kristina Šekrst
- KeynoteTo be announced
11 Dec 2026 · Online
Philosophy of AI Online Colloquium (UV Colloquium)
Kristina Šekrst
- TalkAll the Models Merely Players: The Who, Where, and Why of Consciousness Claims in LLMs
26 Nov 2026 · Online
Artificial Selves Seminar, IDEAS
Kristina Šekrst
- TalkAs an AI Language Model, I Was Told (Not) To
19–20 Nov 2026 · Zagreb · Faculty of Croatian Studies, University of Zagreb
Kristina Šekrst
- KeynoteWho Put the I in AI? Five Neurons and the Philosophy of Mechanistic Interpretability
4–6 Nov 2026 · Online · International Society for the Philosophy of the Sciences of the Mind
Barbara Geld & Kristina Šekrst
- TalkLost in Transformers: Natural Language Autoencoders and the Limits of Machine Introspection
16 Oct 2026 · Mexico City (online) · National Autonomous University of Mexico
Workshop: From Logic to Language: New Takes on Philosophy of AI
Barbara Geld & Kristina Šekrst
- TalkAre Natural Language Autoencoders the Right Path to AI Alignment?
12–16 Oct 2026 · Copenhagen, Denmark · University of Copenhagen
Kristina Šekrst
- TalkAll Work and No Play Makes LLM a Dull Boy
21–23 Sep 2026 · Dubrovnik · Czech Academy of Sciences
Kristina Šekrst
- TalkTo Be, or Not to Be, That Is the Assistant Turn
1–2 Jul 2026 · Brighton, UK · University of Sussex
Kristina Šekrst
- TalkOne Faithful Pass Over the Cuckoo’s Nest
- TalkTo Train a Mockingbird
- Panel Round table on machine consciousness, chaired by John Dorsch, with Anil Seth, Murray Shanahan, Antonio Chella, and Ron Chrisley
Everything else is on Find Me.
Editorial
-
Kristina Šekrst, editor
Philosophical Problems in Science
Zagadnienia Filozoficzne w Nauce (ZFN), Copernicus Center -
Kristina Šekrst, associate editor
AI & Society
Open Forum editor, Springer -
Kristina Šekrst, category editor
PhilPapers: Deep Learning
Category editor
Get in touch
Collaborations, data requests and invitations to explain why a model saying it is conscious is not yet evidence that it is: contact.