Don't Worry About Singularity, We Can't Even Copy an Amoeba

[EricBarbour]

Artificial Intelligence, Really, Is Pseudo-Intelligence

By a UC Berkeley philosophy instructor. Replete with raging nerd comments, mostly from "Gravitas Shortfall", such as:

I don't have words to express just how fundamentally wrong this statement is. Can we please have articles on here from people who actually have even the faintest clue what they're talking about?

It was posted on Slashdot, where similar snotty putdowns ensued.

It never fails. Point out that the AI field has resulted in very little in the way of actual "intelligence", and that the wonderful Singularity still is nowhere in sight, and nameless manchildren will crawl out of their mancaves and scream at you. I suspect that "transhumanists" tend to be very smart, but broken, minds.

[lilburne]

hahaha! I'm amused that they always bring up chess and the Kasparov game in these discussions. Deep blue managed to win because they gave up on AI and resorted to brute force. By the mid 1990s processor speeds were such that positions could be analyzed in depth, no AI involved at all. Let them produce a chess program that plays like Mikhail Tal, or Emanuel Lasker.

[Zoloft]

Back in the day, I felt the two things you needed to do were to encourage your AI routine to make mistakes, evaluate against survival and environmental exploitation statistics, and if a mistake worked well, have a subroutine incorporate new code in the AI's own programming once per generation to make that mistake more likely. I did this in C. It offended the classic AI types. Oh, I always intended to create more than one instance and have them interact socially, but could not get that part to work. The hardware we have now is SO much better I bet if I picked this up again, I could get it to do some weird stuff. *mulls* They have Swarm OS (no Wikipedia article?) and mesh networking (T-H-L) now. I could implement this in small mobile devices.

[Ming]

Old CS joke: "What do you call AI that works? 'Programming.'"

[Notvelty]

Back in the day, I felt the two things you needed to do were to encourage your AI routine to make mistakes, evaluate against survival and environmental exploitation statistics, and if a mistake worked well, have a subroutine incorporate new code in the AI's own programming once per generation to make that mistake more likely. I did this in C. It offended the classic AI types. Oh, I always intended to create more than one instance and have them interact socially, but could not get that part to work. The hardware we have now is SO much better I bet if I picked this up again, I could get it to do some weird stuff. *mulls* They have Swarm OS (no Wikipedia article?) and mesh networking (T-H-L) now. I could implement this in small mobile devices.

Newly (barely?) formed intelligences communicating in pseudo information via mobile devices?

Did you just invent Twitter?