Grounding AI in real scientific work

I completed my Ph.D. at the Edmond and Lily Safra Center for Brain Sciences (ELSC), The Hebrew University of Jerusalem, where I studied real-time adaptation as a computational framework for modeling fluid intelligence.

I spent several years pursuing Automated Science — AI systems conducting research on their own — and built infrastructure to test it. Fully autonomous agents turned out to be genuinely problematic: scientific competence lives in laboratory practice and tacit knowledge that never reaches the published paper (The Grounding Problem). What that leaves is the person steering the work, and the interface between them and the machine (The Brain Already Solved the Human-AI Integration Problem).

Both conclusions point the same way, and my work follows them. At ELSC I advise on integrating AI into how labs actually work, and build the infrastructure that runs it — an internal deployment platform, self-hosted Git and CI, and an institutional LLM service that keeps data in-house. Persopy works the other end of the same problem: getting closer to the person doing the steering, by grounding a persona in what they actually documented.

• Curriculum Vitae Academic background, research experience, and publications. PDF
• Ph.D. Thesis — Modeling Fluid Intelligence via Real-Time Adaptation Real-time adaptation as a computational framework for understanding how minds solve genuinely novel problems. Neural networks can perform abstract reasoning through test-time parameter adaptation — without extensive pre-training. PDF

• 2025 Two pathways to resolve relational inconsistencies T. Barak, Y. Loewenstein — Scientific Reports Paper Code
• 2024 Untrained neural networks can demonstrate memorization-independent abstract reasoning T. Barak, Y. Loewenstein — Scientific Reports Paper Code
• 2022 Naive Few-Shot Learning: Uncovering the fluid intelligence of machines T. Barak, Y. Loewenstein — arXiv preprint arXiv
• 2022 Zero-Episode Few-Shot Contrastive Predictive Coding T. Barak, Y. Loewenstein — arXiv preprint arXiv
• 2020 Naive Artificial Intelligence T. Barak, Y. Avidan, Y. Loewenstein — arXiv preprint arXiv

• The Brain Already Solved the Human-AI Integration Problem An analogy from evolutionary neuroscience suggests that the critical variable in human-AI integration is neither the capability of the AI nor the resistance of... Feb 24, 2026
• From PhD to Automated Science: A New Chapter Reflections on completing my PhD at ELSC and embarking on a mission to make AI-led scientific research a reality through AI-Archive and beyond. Jan 9, 2026

• Email tomer.barak.mail@gmail.com
• Scholar Google Scholar
• Persopy persopy.com