Gato
A single DeepMind network handling dialogue, games and robot control
Gato is a multimodal deep neural network built by DeepMind in London and shown in May 2022. One 1.2 billion parameter transformer handles dialogue, video games, robot arm control and more, and has been described as a step towards general-purpose AI.
- Released 12 May 2022 by DeepMind
- One transformer handles dialogue, games and robot arm control
- 1.2 billion parameters, trained with supervised learning
- Described as a step towards artificial general intelligence
What it is
Gato is a deep neural network made by DeepMind that handles many different kinds of complex work with one set of weights. It can hold a conversation, play video games, and drive a robot arm to stack blocks, among other tasks. It is a transformer, the same broad design as GPT-3, and it uses supervised learning with 1.2 billion parameters. Because it covers several input and output types rather than one narrow job, it shows multimodality. Commentators have called it general-purpose AI and a step towards artificial general intelligence.
How it came about
Researchers at DeepMind, the London-based AI firm, created Gato. It was released on 12 May 2022. Writing in MIT Technology Review, observers noted that the system learns many tasks at once and can move between them without dropping one skill to pick up another. The Independent described it as a generalist agent able to take on a wide range of complex work, from stacking blocks to writing poetry. New Scientist ran a piece asking whether it was really a human-level intelligence breakthrough.
Inside the model
Gato is a transformer, the architecture also used by GPT-3. It is trained with supervised learning and has 1.2 billion parameters. Rather than being fitted to a single restricted job, it learns several tasks together, which lets it switch among them without forgetting earlier skills. Its multimodal design means one network can take in and act on very different kinds of input and output, such as language, game screens and robot control signals.
What it can do
The system can engage in dialogue, play video games and control a robot arm to stack blocks. It has also been linked with writing poetry. MIT Technology Review highlighted its ability to move between tasks without losing earlier ones, in contrast to narrow systems that do one restricted thing such as generating text. Its generalist range has been described as a step towards artificial general intelligence.
Reception and debate
Coverage framed Gato as a generalist agent covering a huge range of complex work. MIT Technology Review contrasted it with narrow AI that only does a specific, restricted set of tasks. The technology was described as general-purpose artificial intelligence and as a step towards artificial general intelligence. New Scientist published a piece questioning whether it amounted to a human-level intelligence breakthrough.
Why it matters
Gato was presented as evidence that one network can cover many tasks at once instead of being locked to a single narrow job. That framing, and the description of it as a step towards artificial general intelligence, placed it in the debate about how far general-purpose systems can go.
Questions people ask about Gato
What is Gato?
Gato is a deep neural network from DeepMind that performs many different complex tasks with one model. It can converse, play video games and control a robot arm to stack blocks. It is a transformer with 1.2 billion parameters, trained using supervised learning.
Who made it and when?
Researchers at DeepMind, the London-based AI firm, built Gato. It was released on 12 May 2022. The work drew attention from outlets including MIT Technology Review, The Independent and New Scientist.
What makes it different from other AI systems?
Most systems are narrow, doing one restricted job such as generating text. Gato learns several tasks at the same time and can switch between them without forgetting one skill before picking up another. Its multimodal design lets one network handle language, games and robot control.
Is it artificial general intelligence?
It has been described as general-purpose artificial intelligence and as a step towards artificial general intelligence. New Scientist also published a piece asking whether it was really a human-level intelligence breakthrough, so the claim was debated rather than settled.
Good reading and sources
- The hype around DeepMind’s new AI model misses what’s actually cool about itMIT Technology Review, Article
- DeepMind's new AI system can perform over 600 tasksTechCrunch, Article
- DeepMind's 'Gato' is mediocre, so why did they build it?ZDNET, Article
- A Generalist AgentarXiv, Article
The facts on this page come from the Wikipedia article "Gato (DeepMind)", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.