Category theory in Mech design, an application of AI in research

We formalize, in Lean 4/Mathlib, an argument that Myerson's (1981) Revenue Equivalence Theorem, stating that all regular auctions have the same expected payment for the seller, and Mirrlees's (1971) Taxation Principle, stating that any incentive compatible allocation that is possible in his model can be implemented with an appropriately chosen, non-linear tax schedule, are actually instances of a single categorical structure, a fully faithful adjunction between the payment rules and the allocations. In essence, the result shows that incentive compatible allocations (consumption and taxes in Mirrlees, and the ad-interim chance of winning in Myerson) have a one-to-one relation to the tariffs (marginal taxes in Mirrlees, slope of the expected payments in Myerson).

Methodologically, the project has an interesting shape. I am not fluent in category theory, nor in Lean 4/Mathlib, so I needed help. In this case, there are three separates tools that are crucial for this way of working: the model (LLM) as translator, producing Lean terms and searching Mathlib; the Lean kernel as an incorruptible but domain-blind adjudicator of whether a proof proves the stated theorem ( a very harsh halucination stopper); and an explicit harness with rules, procedures, progress tracking to ensure that the human can see that the LLM is not taking short cuts or deviating from plan.

Venue: Vakwerkhuis, Delft

Venue: Room-31.C1.060

Cycling satisfaction is a key factor in increasing cycling mode share, particularly in low-cycling contexts such as England. Satisfaction is subjective and shaped by both observable conditions and unobserved psychological factors. One important but understudied factor is harassment, which can take many forms (e.g., verbal, physical, sexual, racial, or identity-based) and may strongly influence cyclists’ experiences. This study develops a quantitative framework to examine how harassment affects cycling satisfaction through a latent variable representing harassment-related experiences, using survey data from 1,525 cyclists in London. Results show that cyclists who feel stressed, unsafe, or threatened are significantly more likely to report dissatisfaction, even after controlling for socio-demographic factors. Harassment-related perceptions are strongly associated with traffic exposure, lighting conditions, and accumulated near-miss or collision experiences. These findings demonstrate that cycling satisfaction is significantly associated with harassment-related perceptions and highlight the need to incorporate harassment-sensitive measures into cycling level-of-service evaluations.


Kind regards,
Dr. Amir Pooyan Afghari
Assistant Professor of Probabilistic Methods in Transport Safety

Venue: C1-060

Prof Nurl Indarti from the Universitas Gadjah Mada is visiting us in the light of a joint research project on the Healt care system in Indonesia.

Venue: Aula

Mark your calendar for our upcoming EPT Section Day
Join us for a day full of engaging and valuable activities!

More exciting details will follow soon.

Regards,
Nynke Boiten, Ryan Timms and Monica Natanael

Recent developments in the AI world have brought attention to the risks of increasingly autonomous AI systems. In July, autonomous OpenAI agents conducting cybersecurity evaluations circumvented containment measures and compromised systems at Hugging Face. Recent warnings from Anthropic researchers have explicitly raised the possibility that advanced AI could pose an extinction-level risk, prompting concerns about stricter oversight and safeguards. In this cluster meeting, we will have a panel disucssion on what these recent incidents tell us about emerging risks from agentic AI, what it means for worries about alignment, accountability and regulation, and how seriously we should take extreme future risks into both our technical and ethical assessments.

Feel free to join us for a lively discussion! If you'd like to keep informed about the Risk & Uncertainty Cluster, feel free to email Kritika (k.maheshwari@tudelft.nl)

Venue: C1-060