Alexander Ku
I'm a research scientist at Google DeepMind and a PhD candidate at Princeton, working with Tom Griffiths, Jon Cohen, and Mike Mozer. I study how intelligent systems solve problems under computational constraints, drawing on methods from both cognitive science and artificial intelligence. My current work focuses on three questions: (1) how intelligent systems combine familiar parts to solve unfamiliar problems, (2) how those parts are represented and what it costs to process them, and (3) how systems adapt their representations and computations to reduce those costs when solving recurring problems.
Prior to this work, my research focused on multimodal learning, embodied agents, and applications of machine learning to genomics.
Email / Google Scholar / CV / GitHub / Twitter
Selected papers
Recent or representative papers:
- Ku, A., Campbell, D., Bai, X., Geng, J., Liu, R., Marjieh, R., ... & Griffiths, T. (2026). Levels of analysis for large language models. Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 384(2320). (pdf)
- Mieczkowski, E., Ku, A., Eisape, T., Arumugam, D., Matters, J., Collins, K. M., ... & Griffiths, T. L. (2026). Improving the Efficiency of Language Agent Teams with Adaptive Task Graphs. preprint arXiv:2605.06320. (pdf)
- Ku, A., Griffiths, T. L., & Chan, S. C. (2026). An evolutionary perspective on modes of learning in Transformers. The Fourteenth International Conference on Learning Representations (ICLR). (pdf)