Comment by jauntywundrkind

From the challenges outlined in Section 4, Identification of National Science and Technology Challenges:

  advanced manufacturing;
  biotechnology;
  critical materials;
  nuclear fission and fusion energy;
  quantum information science; and
  semiconductors and microelectronics.

I have no clue what AI is going to do for most of these sectors/challenges & to me the problem-spaces feel largely non-virtualizable. Massive training & hallucination systems aren't going to happen into great yields regularly, not in a way that justifies the cost.

One place I do think that virtual science can make a huge difference is the last one, semiconductors and microelectronics. IMO there's already enormously promising work, that just needs more funding. Maybe bigger clusters would help some, but mostly, this field needs more works like the incredibly successful OpenROAD work that DARPA begot in 2018 with the Intelligent Design of Electronic Assets (IDEA) program. https://www.darpa.mil/research/programs/intelligent-design-o... https://en.wikipedia.org/wiki/OpenROAD_Project

We should fund better AI powered chip synthesis tools, better open source chip design tools, more open source cpu design, and fund Silicon Foundry projects to build & test promising open source hardware cores. Some focus/personel/money here would go far far further IMO than a seemingly non-funded executive order to get the DoE to scrounge together some systems & command them to make an innovation or two.