Matt's Thoughts In Between - Issue #21

Ian Hogarth (whose “AI nationalism” ideas we discussed before) and Nathan Benaich have an excellent Mary Meeker-style report out on the state of AI. It’s 156 slides, but worth your time to read in full. It’s a superb survey of what’s new in AI across research, hardware, talent, politics and more. There are also excellent deep dives into some of the areas where AI is already having real-world impact (which incidentally mention six EF startups)

It’s worth paying particular attention to the section on what is now possible with reinforcement learning (RL) - a particularly hot topic in a week in which OpenAI unveiled new and striking results in RL agents beating human players in DOTA 2, a popular and complex video game. It’s worth reading Stephen Merity’s thread on why this is a big deal. In short, this is a decisive demonstration of the ability of artificially intelligent agents to collaborate as a team. (OpenAI co-founder Greg Brockman also has a good interview on related topics)

The other must-read section is the final one on public opinion and AI, starting at slide 109. It’s striking both how little the public know (yet) - “Don’t know” is almost always the most popular answer to survey questions - and how little optimism there is among the segment that thinks they do know, which leads to quite radical policy preferences. Given how early we are in the evolution of AI, it’s clear that influencing public opinion is going to be one of the major policy battlegrounds in the years ahead.

A hugely important question in startups is how to maximise the productivity of a very small team. We know that it’s possible for very small teams to have a disproportionate impact on the world - Instagram and Whatsapp had 13 and 55 employees respectively when acquired - so how can we make these situations more likely? 

Tyler Cowen has an interesting, and typically Cowen-esque, post on the topic. I am a little biased, as he basically is espousing the EF philosophy of team building, but I find this idea particularly insightful:

It’s also worth reading in conjunction with this great paper on how DARPA works (via the brilliant Michael Nielsen). Nielsen points out several interesting features of DARPA, including its flat organisational structure and high tolerance for divergent views. The full paper is very good, as is this excellent podcast series on DARPA’s work. 

There are many great books on this broad topic (I like Organising Genius and Reinventing Organisations) and it’s certainly an underrated one for people who want to change the world.

Gizmodo has a surreal-but-true story about a strategy/board game, Machine Learning President, that was designed as a simulation of a US presidential election… and which, according to the New Yorker, got into the hands of Rebekah Mercer - major funder of Trump and Cambridge Analytica. 

Quite apart from the sheer bizarre fun of the story, it does shed light on the surprising real-world impact of board games. It’s not quite Player of Games (sci-fi novel about an empire ruled by games) territory, but it’s well known that the CIA design and play custom strategy games to simulate national security situations. One even got released on Kickstarter.

We’ve talked a bit before about the importance of games in imagining utopias. When you add to this the role of video games, mentioned above, in training and shaping AI, it’s not hard to imagine that we’re entering the golden age of the influence of games. It’s an area worth paying attention to.

As ever, thanks for reading. If you like this, please share it by forwarding to a friend. If you have links you’d like to share, just hit reply.

Until next week,

Matt