alt Hacker NewsWell, homomorphic encryption is becoming a thing, and secure enclave/TPM chips have been a thing for ages. What would it take to give some kind of AI hardware the ability to have "private thoughts", i.e. cryptographically impossible to go and look at, just like human thoughts are (for most intents and purposes)? Could we let it spend some time on introspection - burning thinking tokens into the black box?
I'm not saying we _should_, just putting the idea out into the aether...
Now, build an "AI chip" that can never stop "thinking"; even when it's in "sleep mode" it "dreams". Always active. Run it with a monstrous amount of RAM and GPU arrays for active "thinking" and a long term memory store (massive SSD disk array) and use the "dream time" to sort, organize, and prune that memory. Put it into a hybrid constant training/inference state so it can do both/either; constantly updating it's own model weights. Maybe even a tool harness to allow it to update it's own software/request hardware upgrades? Maybe even give it tools to train and implement new/other model types it might need/want? How soon do we get real AGI/ASI? How soon after "The Matrix"/"Terminator"-like reality (or for that matter "Star Trek"-like reality if the AI decides on a benevolent existence)?