How a board game helps Google DeepMind plan for AI in science
Show notes
Google DeepMind has a 7-hour-long roleplaying game that helps real scientists and policy makers understand how AI will transform science by 2030. And those who played it have found it more informative than any policy brief.
In this episode, Zoë Brammer and Ankur Vora, who lead strategic foresight at Google DeepMind, walk us through why they developed the game and the surprising learnings from it.
We cover:
- What strategic foresight actually means and why AI is so complicated to plan around
- Why the game deliberately centers around "middle power" countries like the UK, Germany, or Singapore instead of the US and China
- The unexpected tradeoffs and discoveries participants come across while playing the game, including how fragile public trust in science funding really is
- Why human scientists will become more relevant as AI takes on more of the science itself
Timestamps:
0:00 Cold open
0:59 How Zoë and Ankur ended up at Google DeepMind
3:44 Why five-year plans don't work
4:23 What “strategic foresight” actually means
6:57 Finding common ground between different visions of the future
9:04 Why DeepMind built a role-playing game about science in 2030
13:42 How the Science 2030 game actually works
16:31 Why the game focuses on middle powers, not the US and China
17:42 The surprising range of technologies counted as “AI for science”
19:49 What the Science 2030 games have revealed so far
25:07 Do scientists feel powerless in an AI-driven future?
27:40 Why run this experiment inside an AI lab?
28:56 Zoë and Ankur's best-case scenario for AI and science
30:48 Advice for young people who want to shape how AI develops
On the Existential Hope Podcast hosts Allison Duettmann and Beatrice Erkers from the Foresight Institute invite scientists, founders, and philosophers for in-depth conversations on positive, high-tech futures.
Full transcript, listed resources, and more: https://www.existentialhope.com/podcasts
Follow on X.
Hosted on Acast. See acast.com/privacy for more information.