I wonder whether the current technology is feasible for a replication of HAL.
- Very good chess player -> Check (And probably can be a top notch Go player too, although I don't play chess or go)
- Very good knowledge of history, science, technology, everything -> I'm not sure if it's possible. And I'd assume we need to send scans of all books, because for example it should know many fine details and educated guesses about a certain historical fact.
- Understand human language pretty well -> I actually think ChatGPT is good enough if I choose my wording clearly. But voice recognition might prose some challenges. It also depends on how good the microphone is.
- Have pretty good voice generation. I'd like to choose the voice too! -> I guess it's OK-ish nowadays? I listened to a few AI generated clips and they are pretty good, not sure how practical it is, though.
Anything else I missed? I know HAL also has its multi-decade objectives in mind, so this is different from the LLMs which don't seem to have a very long term of memory.