5/5 Collaboration between @aalto.fi and @oistedu.bsky.social, supported by Aalto Science Institute (ASci) and the Helsinki Institute for Information Technology (HIIT).

4/5 This work contributes to a young research agenda seeking to understand creativity beyond the realm of humans or highly developed animals. We argue for the SO model as a fascinating candidate to study the effect of learning on creativity from the bottom up - in life as it is and as it could be.

3/5 More specifically, we demonstrate that modifying the SO model learning parameters gives rise to four different regimes that can account for both creative products and inconclusive outcomes, thus providing a framework for studying and understanding the creative potential of learning systems.

2/5 Developed to model complex adaptive systems in ALife and advocated as a candidate for minimal agency, the Self-Optimization (SO) model can be considered as the 3rd operational mode of the classical Hopfield Network, leveraging the power of associative memory to enhance optimization performance.

4/4 Locally, we're part of a rich AI research ecosystem, including the Finnish Center of AI (FCAI), Helsinki Institute for Information Technology (HIIT) @icthiit.bsky.social and the Finnish @ellis.eu institute, a world-class research hub in AI and machine learning.

3/4 You can work w/travel to our fantastic collaborators Natasha Jaques @natashajaques.bsky.social (Washington U./DeepMind), Sebastian Deterding @codingconduct.cc (Imperial College), Christoph Salge @christophsalge.bsky.social & Daniel Polani (UH), & Julian Togelius @togelius.bsky.social (NYU)!

2/4 The AIR group (www.autotelic.science) embraces diversity in research, thought and identity and has a track record of research on (computational) intrinsic motivation, bridging between AI, Psychology and Cognitive Science. Aalto University is 15 min away from Helsinki and surrounded by nature.

5/5 🤝 A collaboration between Aalto University @aalto.fi and Imperial College @imperialcollegeldn.bsky.social, generously supported by the Research Council of Finland and Helsinki Institute for Information Technology @icthiit.bsky.social

4/5 📆 Catch us in person: We will be presenting this work first at RLDM @rldmdublin2025.bsky.social in June (in person) and then at CogSci @cogscisociety.bsky.social in July (likely remote). Please join us for exciting discussions of this research and the directions in which we take it next!

3/5 🤔 Why it matters: By matching psychological theories with AI models, we foster transparency and testability in motivation research. More generally, our work can support a cycle of theory development by inspiring new AI models and experimental designs, which can then be used to refine the theory.

2/5 🔎 Overview: We match models of intrinsic motivation from RL with 4 facets of competence identified in previous work: effectance, skill use, task performance & capacity growth. This uncovers previously hidden preconditions in SDT and supports refining our understanding of intrinsic motivation.

3/3 Why it matters: Understanding how we judge machine-made art can guide HCI design, support fair system evaluations, and inspire co-creativity applications. More exploratory results in paper! Collaboration between @aalto.fi and @helsinki.fi, supported by HIIT and the Research Council of Finland.

2/3 We compared two drawing robots, a mechanical plotter and an anthropomorphized arm, creating the same still-life drawings. Surprisingly, differences in robot embodiment did not affect creativity ratings. But seeing the art being made, and what makes it, boosted how creative people perceived them.

(3/3) We are now extending our proof-of-concept with more experiments, e.g. on transfer learning. A key challenge is quantifying the diversity of learned skills, and we look forward to distilling relevant insights for tackling this from interacting with the wider community!

(2/3) DP is a method for shaping goal selection in reinforcement learning based on an agent’s beliefs about how much pursuing the goals will diversify its skills. DP is applicable to a particular class of intrinsic rewards used to learn a repertoire of diverse skills without task-specific rewards.