Shane Legg

Overview

Shane Legg is a New Zealand artificial intelligence researcher and a co-founder of DeepMind ^[1]. He is one of the three people who started the London lab in 2010, together with Demis Hassabis and Mustafa Suleyman ^[2]. Much of his later work has centred on the question of how to keep very capable AI systems safe ^[1].

Education and early research

Legg completed a PhD at IDSIA in Switzerland under Marcus Hutter; his 2008 thesis was titled "Machine Super Intelligence" ^[1]. With Hutter he worked on theoretical models of super-intelligent machines, a line of work known as AIXI ^[1]. That research was formal rather than engineering work: it aimed at mathematical descriptions of super-intelligent machines rather than at building working systems ^[1].

He is also credited, together with Ben Goertzel, with re-introducing and popularising the term "artificial general intelligence" (AGI) in the early 2000s ^[1]. The phrase describes AI systems whose abilities are meant to be general rather than confined to one narrow, single-purpose task, and it later became the ordinary label for that goal across the field.

From University College London to DeepMind

Legg first met Demis Hassabis in 2009 at University College London, where Legg was a postdoctoral researcher; he later held a fellowship at the Gatsby Computational Neuroscience Unit ^[1]. In 2010 he co-founded DeepMind with Hassabis and Mustafa Suleyman in London ^[2].

DeepMind's own account of its history says the three of them started the company in 2010 with the mission to "build the world's first artificial general intelligence", and that Google bought the company in 2014 ^[4]. The stated mission put the research term Legg had helped popularise at the centre of a commercial laboratory, rather than leaving it as an academic label.

Safety and AGI

A 2017 article described leading DeepMind's AI safety work as a significant part of his job, and as of July 2023 he was Google DeepMind's Chief AGI Scientist ^[1]. On his own website he describes himself as Google DeepMind's Chief AGI Scientist and as co-chairing the AGI Safety Council ^[3]. DeepMind's blog says that the AGI Safety Council, which Legg leads, analyses AGI risk and recommends safety measures ^[5].

His public roles therefore sit on both sides of the same problem: the company states that it is working towards general AI, while a council led by one of its co-founders is tasked with assessing the risks of that work and recommending how to handle them ^[4][5].

Significance

Legg belongs to the small group of researchers who pushed the idea of general-purpose AI from a theoretical topic into the stated goal of a major industrial lab ^[1][4]. Wikipedia gives his birth year as 1973, or possibly 1974 ^[1].