I think the idea is to test (with the jury as a sample) if the evidence is compelling enough for society at large to agree with the verdict. It's rarely going to be 100% clear cut so the jury has to use their judgment.
I recently had a pretty bad injury and out of curiosity I asked Gemini what it thought based on some CT scan slice images (and no other information). Surprisingly it came to exactly the same diagnosis and treatment plan as my doctors, but the big advantage is that I could ask it follow up questions any time, whereas the doctors barely explained anything.
Nothing can beat the Python numpy/ML ecosystem. There's a lot of value in just being able to run a Python script as well without any compilation step. The typing isn't perfect right now but it's usable.
For vectorizable problems there also won't be huge performance gains from switching to a compiled language because all the hard stuff is already done in highly optimized native code. The only time it really makes a difference is if you have to write a custom for loop or traversal.
I think it's niche now because getting the hardware to run it is expensive and the quantized models don't work as well. If those improve then it would be a no brainer to pay one off for the hardware instead of a fortune for API calls.