GaŠper BeguŠ

Gašper Beguš investigates how intelligence—human, animal, and artificial—unfolds through language. He is an Associate Professor of Linguistics at UC Berkeley, where he focuses on interpretable AI and combines linguistics, cognitive science, machine learning, neuroscience, philology, and marine biology.

Beguš leads efforts to develop techniques that help us better understand the inner workings of AI. In the Berkeley Biological and Artificial Language Lab he develops interpretable machine-learning models, including an “artificial baby” that learns speech from raw audio the way infants do.

As Linguistics Lead of Project CETI, he applies similar tools to the language of whales and recently discovered that sperm whales produce sound patterns analogous to human vowels and that orcas have vowels and consonants.

Beguš works with industry through InterpretAI to make neural networks more transparent.

His research has been covered by The Guardian, The Economist, National Geographic, The New York Times, Financial Times, The Atlantic, NPR, BBC, WIRED, Quanta, Harvard Magazine, Noema Magazine, and others.

Beguš regularly appears as an invited speaker in diverse venues such as NYU Stern School of Business, OpenAI, Centre Pompidou, NYU Law, the National Science Foundation, and the Santa Fe Institute.

His models inspired parts of the La Biennale di Venezia exhibition and a science fiction book for young audiences.

Beguš is also an advocate for residential education. He served as College Principal of Bowles Hall, UC Berkeley’s oldest residential college, and as a resident tutor at Harvard during his graduate studies.

Research

Research Projects

Interpretable AI for language – Building “artificial baby” models that learn speech from raw audio and tracing their inner workings against brain data.
AI-driven discovery – Exploring how our AI interpretability techniques can guide new insights for science.

Cross-species linguistics – Decoding sperm whale codas to test which properties of language extend beyond humans.
Origins of language – Modeling evolutionary pathways from vocal imitation to compositional syntax.
Law & society – Translating our interpretability findings into policy debates on the legal status of non-human animals.

Outreach

Our models are powering a part of the Next Earth exhibition at the Biennale di Venezia 2025.

Our models also inspired parts of a science book by Kathryn Hulick for young audiences.

The book imagines how to use our AI interpretability technique in case aliens contacted us.

FinneGAN: Latent Spacecraft

FiwGAN model trained on Joyce's Finnegans Wake helps us interpret one of the graetest novels and latent spaces in AI

A new rule in Vedic meter

Whale vowels on CBS News Bay Area

Talk at NYU Law Future of Rights & Governance

Our research featured on the front page of The Guardian.

Whale vowels

Collaboration with the artist Agnieszka Kurant, Phantomatics (2026), currently exhibited in Palazzo Diedo, Venice (concurrent with the Biennale). For the model, see ConlangCrafter.