I left Google DeepMind. This is a particularly odd decision because I love the team I worked with at DeepMind. Since joining DeepMind, I’ve been constantly amazed by this team’s compassion, focus, intelligence, optimism, and desire for real-world impact. I’ve learnt a huge amount from DeepMind, and I will miss this team dearly. Leaving this awesome team is enormously difficult.

So, why did I leave?

I’m terrified by climate change; and I’m increasingly convinced that climate change mitigation is a ‘go big or go home’ thing. I have two ideas for how to help. These ideas might be wrong. But I want to give them a shot.

The first idea is motivated by the observation that some business norms (like hoarding intellectual property) get in the way of achieving climate impact. My hope is that it’s possible to design an organisation which is optimised through-and-through for climate impact; and short-circuits all business norms which get in the way.

So, I will build a non-profit product development lab, obsessively focused on building things which reduce CO2e emissions by at least one million tonnes per year; and hopefully billions of tonnes per year. The aim is to catalyse systemic change and help the entire community (including DeepMind!) to have as much climate impact as possible.

We’ll be very collaborative; everything will be open source; and we’ll encourage others to commercialise our work and/or out-perform any systems we build. We’ll aim for maximum CO2e reduction per dollar. The non-profit will combine ideas from open-science and startup culture, and will focus on interventions which are practical and scalable.

This idea is probably wildly naive. But I want to test it out.

The second idea is the first project that the non-profit will work on: forecasting solar PV power production a few hours ahead by using machine learning to combine real-time satellite data with numerical weather predictions, in collaboration with the UK Met Office and National Grid.

Please don’t interpret my resignation as a comment on DeepMind’s likely climate impact. We all need DeepMind to have huge climate impact. And, if anyone can have impact, it’s the astounding folks in DeepMind. Heck, they already are having impact.

I’m extremely excited to start this non-profit; and can’t wait to start building things!

I’ll post progress updates to twitter and my blog. Please feel free to email me at jack@jack-kelly.com if you’d like more info. I probably won’t get much coding done in January - Jan will mostly be spent setting up the non-profit, building a website etc. And settling on a name!

UPDATE Please see these blog posts for more info: