Supervisors

Suitable for

Abstract

Prerequisites: basic AI/ML courses 

 Background 

Motivation: When working to build machines that have a form of “intelligence”, it is natural to be inspired by humans. Of course, humans are very different from machines, in their embodiment and myriad other ways. Humans exploit their bodies to experience the world, create an internal model of it, and use this model to reason, learn, and make contextual and informed decisions. Machines lack the same embodiment but often have access to both more memory and more computing power. Despite these crucial disanalogies, it is still useful to leverage our knowledge of how the human mind reasons and makes decisions to design and build machines that demonstrate behaviours similar to those of a human.

In this project, we aim to investigate a novel AI architecture, Slow and Fast AI (SOFAI), which is inspired by the Thinking, Fast and Slow cognitive theory of human decision-making. SOFAI is a multi-agent architecture that employs both “fast” and “slow” solvers underneath a metacognitive agent that is able to choose among solvers as well as reflect on, and learn from, past experience.

Related Work: Cognitively inspired architectures are a widely studied area of research and have produced many applications and different architectural paradigms (Kotseruba and Tsotsos 2020). In this project, we examine a specific novel architecture introduced in (Booch, et al. 2021, Fabiano, et al. 2025), along with its underlying paradigms (Kahneman 2011, Graziano 2013).The architecture has been shown to be valuable in several use cases, such as automated planning and constrain-drivennavigation (Fabiano, et al. 2025).

Focus

This project centres on the Thinking Fast and Slow paradigm and related dual-process theories, rather than on the SOFAI architecture itself. We aim to investigate how System 1/System 2 interactions can support learning, modelling, and decision-making in new settings. Some research directions include:

Use the Thinking Fast and Slow paradigm to explore novel problem settings, e.g., controller synthesis.

Investigate methods to derive deductive rules from experience and use those rules to guide and constrain reasoning.

Enrich the architecture with online learning so both systems can readily adapt from experience.

Develop solver-agnostic ways to transfer information from System 1 to System 2.

Explore non-crisp notions of correctness: “how can we evaluate solutions when a formal notion of correctness is absent?”

Shift the role of System 1 and System 2 from the purely “solving” phase to the modelling phase.

Method

The project will begin with a literature review. Building on this, we will formalize ways to enrich the interaction between System 1 and System 2---for example, by developing principles that guide S2 reasoning using S1 outputs, or by evaluating the architecture’s behaviour through relaxed notions of correctness. The later stages will focus on implementing and testing these ideas in controlled settings, such as synthesis or planning tasks, to assess how the proposed S1/S2 mechanisms improve modelling or decision quality. Expected outcomes include a proof-of-concept implementation, a clear conceptual and theoretical framework, and, if results are promising, a publication-ready paper detailing the findings.

Bibliography

Booch, Grady, Francesco Fabiano, Lior Horesh, Kiran Kate, Jonathan Lenchner, Nick Linck, Loreggia, et al. 2021. “Thinking Fast and Slow in AI.” AAAI Conference on Artificial Intelligence. 15042-15046. https://ojs.aaai.org/index.php/AAAI/article/view/17765.

Fabiano, Francesco, Marianna B. Ganapini, Loreggia, rea, Nicholas Mattei, Keerthiram Murugesan, Vishal Pallagani, Francesca Rossi, Biplav Srivastava, and K. Brent Venable. 2025. “Thinking Fast and Slow in Human and Machine Intelligence.” Commun. ACM (68) 72–79. doi:10.1145/3715709.

Graziano, Michael SA. 2013. Consciousness and the social brain. Oxford University Press.

Kahneman, Daniel. 2011. Thinking, Fast and Slow. Macmillan.

Kotseruba, Iuliia, and John K. Tsotsos. 2020. “40 years of cognitive architectures: core cognitive abilities and practical applications.” Artificial Intelligence Review. https://doi.org/10.1007/s10462-018-9646-y.