Gato
In plain wordsA 2022 DeepMind AI that used one set of learned settings for hundreds of jobs, from chatting and describing pictures to playing video games and moving a real robot arm — an early step toward general-purpose AI agents.
A single transformertransformer, n.: The design behind today's chatbots and much of modern AI. It handles text by letting every word look at every other word and decide which of them matter for understanding it. network from DeepMindDeepMind, prop. n.: A London artificial intelligence company, bought by Google in 2014, famous for programs that mastered the game of Go and predicted the shapes of proteins. In 2023 it merged into Google DeepMind., with 1.2 billion parameters, that handled 604 distinct tasks without any change of weights — among them conversation, image captioning, Atari video games and block stacking with a physical robot arm. Every input and output, whether text, image patches, button presses or joint torques, was flattened into one token sequence, and the model learned by supervised learningsupervised learning, n.: Teaching a computer by example: you give it many inputs paired with the right answers, such as photos labeled 'cat' or 'dog', and it learns to answer correctly for new inputs. to imitate expert behavior, using the context to work out what kind of output was wanted. Gato performed most tasks at a respectable fraction of expert level but rarely matched specialized systems, and its small size was dictated by the need to control a robot in real time. It was the best-known early demonstration that one sequence model could act as a generalist agent across modalities and bodies.
Gato's recipe was simple: tokenize everything and let one transformer predict the next token.
Origin
Described by Scott Reed, Konrad Żołna, Emilio Parisotto and colleagues in 'A Generalist Agent', posted on 12 May 2022 and later published in Transactions on Machine Learning Research.
Usage note
Its release prompted the DeepMind researcher Nando de Freitas to declare that the game was over on the road to AGI, a claim widely disputed because Gato showed breadth rather than strength.
Related
- artificial general intelligenceartificial general intelligence, n.: AI that could learn and do almost any mental task a capable adult can, rather than excelling at one narrow job. Experts disagree on the exact bar and on when, or whether, it will arrive.
- generalist robot policygeneralist robot policy, n.: A single robot-control program trained to do many different tasks in many places, often on several kinds of robots: the opposite of a specialist trained for one job.
- transformertransformer, n.: The design behind today's chatbots and much of modern AI. It handles text by letting every word look at every other word and decide which of them matter for understanding it.
- AI agentAI agent, n.: An AI system that does not just answer but acts: given a goal, it decides on steps, uses tools such as a browser or code editor, checks the results and keeps going until the job is done.
- foundation modelfoundation model, n.: A very large AI model, built once at great cost from vast amounts of text, images or code gathered mostly from the internet, that many products then reuse as their starting point, from chatbots to coding tools.
- DeepMindDeepMind, prop. n.: A London artificial intelligence company, bought by Google in 2014, famous for programs that mastered the game of Go and predicted the shapes of proteins. In 2023 it merged into Google DeepMind.
Sources
- Reed, S., Żołna, K., Parisotto, E. et al. (2022). A Generalist Agent. Transactions on Machine Learning Research.
Mentioned in
DeepMindDeepMind, prop. n.: A London artificial intelligence company, bought by Google in 2014, famous for programs that mastered the game of Go and predicted the shapes of proteins. In 2023 it merged into Google DeepMind.