Pedro A. Ortega

AGI and Cybernetics Researcher

About

I was VP of research at Kosen Labs and prior to that, lead of the Safety Analysis Team at DeepMind. My research focuses on artificial general intelligence and the formal principles of intelligence, covering aspects such as learning, planning and decision making in both machines and biological organisms. My approach lies at the intersection between machine learning, computational neuroscience, theoretical economics, and physics.

Research Interests

Most of my work centers on information-theoretic and statistical mechanical approaches to learning and control, leading to contributions in bounded rationality models and recasting adaptive control as a causal inference problem. I have also worked on causal induction, and on game- and decision-theoretic models in computational neuroscience.

Free energy

To get a sense of my work, please refer to:

my PhD thesis on bounded rationality and Bayesian control,
my paper on Thompson sampling as Bayesian causal inference,
my paper on bounded rationality through thermodynamic principles,
or my publications.

I was part of the former Tuebingen group on Sensorimotor Learning and Decision Making, where some of the now widespread ideas on Thompson sampling, causality, and thermodynamic methods for reinforcement learning have originated.

News

Feb 4, 2024	Panel speaker at the Warwick AI Safety Summit 2024, 17 February 2024.
Nov 22, 2023	New website!
Nov 14, 2023	Talk on information-theoretic bounded rationality at the Oxford AI Society.

Selected publications

A minimum relative entropy principle for learning and acting

Pedro A Ortega, and Daniel A Braun

Journal of Artificial Intelligence Research, 2010

Bib PDF

@article{ortega2010minimum,
  title = {A minimum relative entropy principle for learning and acting},
  author = {Ortega, Pedro A and Braun, Daniel A},
  journal = {Journal of Artificial Intelligence Research},
  volume = {38},
  pages = {475--511},
  year = {2010},
}

Thermodynamics as a theory of decision-making with information-processing costs

Pedro A Ortega, and Daniel A Braun

Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2013

Bib PDF

@article{ortega2013thermodynamics,
  title = {Thermodynamics as a theory of decision-making with information-processing costs},
  author = {Ortega, Pedro A and Braun, Daniel A},
  journal = {Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences},
  volume = {469},
  number = {2153},
  pages = {20120683},
  year = {2013},
  publisher = {The Royal Society Publishing},
}

Meta-learning of sequential strategies

Pedro A Ortega, Jane X Wang, Mark Rowland, Tim Genewein, Zeb Kurth-Nelson, Razvan Pascanu, Nicolas Heess, Joel Veness, Alex Pritzel, Pablo Sprechmann, and others

arXiv preprint arXiv:1905.03030, 2019

Bib PDF

@article{ortega2019meta,
  title = {Meta-learning of sequential strategies},
  author = {Ortega, Pedro A and Wang, Jane X and Rowland, Mark and Genewein, Tim and Kurth-Nelson, Zeb and Pascanu, Razvan and Heess, Nicolas and Veness, Joel and Pritzel, Alex and Sprechmann, Pablo and others},
  journal = {arXiv preprint arXiv:1905.03030},
  year = {2019},
}

Shaking the foundations: delusions in sequence models for interaction and control

Pedro A Ortega, Markus Kunesch, Grégoire Delétang, Tim Genewein, Jordi Grau-Moya, Joel Veness, Jonas Buchli, Jonas Degrave, Bilal Piot, Julien Perolat, and others

arXiv preprint arXiv:2110.10819, 2021

Bib PDF

@article{ortega2021shaking,
  title = {Shaking the foundations: delusions in sequence models for interaction and control},
  author = {Ortega, Pedro A and Kunesch, Markus and Del{\'e}tang, Gr{\'e}goire and Genewein, Tim and Grau-Moya, Jordi and Veness, Joel and Buchli, Jonas and Degrave, Jonas and Piot, Bilal and Perolat, Julien and others},
  journal = {arXiv preprint arXiv:2110.10819},
  year = {2021},
}

Neural networks and the chomsky hierarchy

Grégoire Delétang, Anian Ruoss, Jordi Grau-Moya, Tim Genewein, Li Kevin Wenliang, Elliot Catt, Chris Cundy, Marcus Hutter, Shane Legg, Joel Veness, and P.A. Ortega

In International Conference on Learning Representations (ICLR), 2023

Bib PDF

@inproceedings{deletang2022neural,
  title = {Neural networks and the chomsky hierarchy},
  author = {Del{\'e}tang, Gr{\'e}goire and Ruoss, Anian and Grau-Moya, Jordi and Genewein, Tim and Wenliang, Li Kevin and Catt, Elliot and Cundy, Chris and Hutter, Marcus and Legg, Shane and Veness, Joel and Ortega, P.A.},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year = {2023},
}