Product1 publisher3 min readPublished
DeepMind puts its chair and chief AGI scientist in charge of its new AGI essay platform
Google DeepMind has launched the DeepMind Institute, whose first four essays arrive with a note that they are not Google's official view even though its three directors sit at the top of Google and Alphabet.
The Product Desk · Product desk
What happened
- Google DeepMind has launched the DeepMind Institute, a platform for AGI research and debate, with Shane Legg, Demis Hassabis and Google research president James Manyika as its three directors.
- Its launch essay names cybersecurity, biorisks and the potential for loss of control in future self-improving systems as concerns, and expects today's capability gaps to close soon.
- An accompanying economics essay by Julian Jacobs and Alex Imas rates eleven policy responses using literature reviews, surveys and 51 AI agent raters built from survey data on 51 real economists.
- Legg called Dario Amodei's 12 September call for the industry to slow frontier releases "interesting directionally" and "worth considering" in an interview with the Financial Times.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- constraint Anyone citing Hassabis's frontier AI essay in a consultation response has to carry the site's disclaimer with it, so the argument arrives with the standing of Alphabet's chief scientist and no company position behind it.
- capability Scored and ranked policy options are the format legislative staff can lift straight into a brief, and the institute now supplies eleven of them written inside the lab whose product the policies would cushion.
- decision Teams weighing the eleven ratings have to decide whether scores partly produced by AI agents standing in for economists are evidence they will put their own name behind.
- contradiction The institute's managing editor holds that declaring AGI achieved is premature while Nvidia and OpenAI have already claimed it, which leaves regulators no agreed definition to write rules against.
A policy analyst who wants to quote "A framework for frontier AI and the dawning of a new age" next week has to decide what to put on the attribution line. The author is Demis Hassabis, who co-founded Google DeepMind, chairs it, and is Alphabet's chief scientist [2][9]. The site that published it says its pieces reflect their authors' ideas and "should not be read as Google's official view" [4].
Shane Legg is one of the three directors and also the managing editor [2][3]. The institute says contributors from Google DeepMind, Google and the wider research community "will not always agree" [8]. Bilal Chughtai, who left DeepMind's AGI safety team in July, posted a warning this week, and he is not among the four launch authors [17][9].
The essay with numbers in it is "Economic policy for AGI", by Julian Jacobs, an economist and research scientist at Google DeepMind, and Alex Imas, its Director of AGI Economics [9][18]. They match three "least-regret" responses to three scenarios [20]. Mild disruption gets expanded unemployment insurance, a wider Earned Income Tax Credit and employer-led retraining. Moderate displacement turns that credit into a negative income tax. A sustained fall in labour's share of GDP triggers a universal basic capital backstop [20]. Universal basic income they call "an expensive and blunt instrument that may fail to concentrate sufficient relief where it is needed most" [21].
The policy held back for the worst case is the one their ratings score lowest on getting done. Universal basic capital came first of the eleven on agency, at 76.3 out of 100, and 33.1 on feasibility, second lowest after a federal jobs guarantee [23]. The two scores are 43.2 points apart [24]. Their survey found 54% support for universal basic capital against 85% for publicly funded retraining [22].
Legg's own timing is more cautious than the claims made this month. He told the Financial Times it was premature to declare AGI had been achieved [13]. Jensen Huang said on 7 September that AGI has arrived, and OpenAI made its own AGI claim on 4 September [14][15]. The FT reported that Legg remains comfortable with a 50% chance of "minimal" AGI by 2028 [16].
TNW's report on the launch does not mention regulation or lawmakers as the institute's audience; the launch essay describes a platform for research and debate on AGI [27].
Two questions sort a lab-published essay for the person who has to cite it. Does the author hold operating authority over the product? Does the piece recommend a policy someone could legislate? Stephen Cave's "Principles for a new utopianism" answers no to both, and the disclaimer costs nothing there [9]. Jacobs and Imas answer yes to the second [20]. Hassabis answers yes to both, so anyone quoting him is deciding whether a sentence by Alphabet's chief scientist can be read apart from Alphabet [2].
What to watch
- Whether DMI publishes an essay that contradicts a stated Google or Alphabet position, and whether it stays up.
- Whether the eleven ranked policies turn up in a legislative submission or hearing cited to DMI.
- Whether Legg revises the 50% chance of minimal AGI by 2028 as rival AGI claims accumulate.