I'm a big believer that AI will transform the Earth Sciences, in the way it has Life Sciences. This is a very good essay by
@orbuch, who has been great fun discussing this topic with
Imo this is what good venture investing should look like— mobilising imagination, thought and words to articulate what the world is capable of looking like tomorrow.
> The big claim underlying Ryan's thesis is that
@RichardSSutton's Bitter Lesson can be applied to understanding the Earth System as well. I agree that compute and data can help us reach generalization in Earth Sciences. However, I have a minor but important disagreement-- unlike LLMs, this *will* require specialised human knowledge (given how complex & specialised Earth Systems are, its geochemistry, oceanographic complexities, geophysical quirks & more)
> I like that Ryan talks about all that generalizable AIs can potentially achieve when it comes to monitoring and moulding our planet. While also making a very subtle point about the dangers of AI: that this a fundamentally new form of intelligence previously unbeknownst to the planet. They don't breathe (!) They are not part of the natural world that makes up our planet. How do we understand this from an AI Safety lens?
> Liu Cixin has a brilliant short story Moonlight, about how human ingenuity that engineers the planet eventually created unintended consequences seen only decades later. I think this essay tackles that criticism head on constantly keeping in mind the sentiment of the butterfly effect, which is refreshing.
> The visualisations of sensor networks shown in the essay are a very good reminder of the large difference that exists in mapping Earth Systems, between the Developed and the Developing countries. ERA5 is a good example. It's barely decent in prediction for South Asia becuase of sparse ground truthing. We will need creative ways to finance such gaps— from global philanthropic support, to multi governmental initiatives, to citizen science networks, to some potentially very profitable startup ideas. I think one great source could be frontier labs and the new windfall hitting them.
> The essay talks about the Ocean being a major blind spot for us. I would propose another— the subsurface. We barely know what's beneath our feat, despite geophysical tools now finally becoming advanced enough that we could peer to the centre of the earth without having to drill (see image below for how sparse our coverage of the underground has been).
> Lots of new learnings from sensor carrying seals to how somali pirates make monsoon tracking difficult!
I hope this inspires a new generation of natural scientists. Taking the Gaia hypothesis seriously, studying the Earth System, making it programmable will be a remarkably difficult but also a very exciting journey. It will take us to volcanoes, forests, rivers and canyons. We will bring these natural phenomena back to the lab, and carry the labs into the planet. Along the way, we will solve & discover new mysteries about our planet, study and learn about the long geological history that has given birth to and sustained life, and maybe, just maybe, we will also understand where we as humans have come from, and how are we floating on this pale blue dot, suspended in a vast galaxy around us.
If you find this exciting
@AltCarbonIndia is always hiring (you can read a short note from last year about our aim to create Planetary Intelligence):
altcarbon.notion.site/Alt-Ca…
I wrote an essay about why the next frontier for AI is the world outside the data center
Nature is a system more complex than anything we've ever created
AI can learn how the system fits together, so we can finally understand how to shape it
How?
naturalgeneralintelligence.a…