There are already impressive efforts in AI/ML for climate change (oops, I said the words!) and I would be shocked (maybe not surprised) if this doesn't get included in this effort in some way.

www.realclimate.org/index.php/ar...

The use of AI to describe everything from LLMs to regression on big data sets is a bit problematic. It's used by the folks creating AI slop to claim credit for ML successes based on totally different architectures & we are now at the point where normal people hear AI & think chatbot (argggh!).

Nonetheless, if the budget is there, this mission could be useful because some of the biggest roadblocks to big ML efforts is really bringing the training data onto one accessible system. Note that this will really be ML, and not based on GenAI efforts such as ChatGPT etc.

There are some downsides. The DOE is a very expensive place to do 'normal' (not national security related) science. Overhead rates that pay for the security at national labs are ~300 to 400% which means that doing something within their firewall is roughly 3 times as expensive as doing it outside.

While earth science is specifically not mentioned in the EO, my expectation that there will be an earth science thrust - based likely on the successes of ML weather forecasting (FourCastNet, DeepCast, NeuralGCM etc.) and building on the E3SM infrastructure that DOE has already developed.

DOE has resources for this - and some very impressive high-performance computing resources already. These are developed mostly for nuclear weapons testing simulations, but have been used for other applications - protein folding, earth system models etc.

It's not obvious to me that creating a single platform for genomics, pharmaceuticals, earth science, materials science etc. is really a good idea since you'd need a big increase in scale for not much benefit. My guess is that they will spin off specific topics to mostly separate systems.

This is something that people have been advocating for in the US for some time, but the scale of such an effort is greater than any one of the science agencies could really do on it's own. The Genesis Mission seems to be even more ambitious - bringing multiple scientific areas together.

First off, this 'mission' is not something totally out of the blue. In Europe, a main pillar of the Destination Earth project is a 'data lake' that would bring all Earth science data sets together (satellite data, models, reanalyses, in situ) to provide a foundation for machine learning.

We got a Charlie earlier this year - game changing. Even Kate comes over to watch me boil water and bake at the same time!

Yes. Same concept

Similarly, it's different from weather forescasting - even on seasonal time scales - which does provide information on climatological risks (i.e. the probability of extremes).

It is motivated by the (obvious) non-stationarity in observations and the need to predict risks for today, not for 20 or 30 years ago. This is distinguishable as well as climate trends going forward 20 or 30 years whch depend on future emissions.