a thoughtful web.
Good ideas and conversation. No ads, no tracking.   Login or Take a Tour!
comment by katakowsj
katakowsj  ·  3918 days ago  ·  link  ·    ·  parent  ·  post: Why robots will not be smarter than humans by 2029

Interesting from my perspective as an educator. Even if an A.I. could learn, it would likely take much time to nurture the behaviors and learning that would make it similar to a human adult. I teach the 8th grade.

Daily, I see 8th graders who are immensely intelligent and capable of great things, but are also equally deficient in wisdom and experience, and therefore make terrifying choices. I see that any A.I. will have a very long series of learning experiences resembling those of a human being. 2029 for A.I. could be very much like people's expectations of being in flying cars at this point in time.





rob05c  ·  3918 days ago  ·  link  ·  

    A.I. could be very much like people's expectations of being in flying cars
AI is exactly like flying cars. For some 40 years, researchers thought they were 10 years away from sentient AI.

Yet, we continue making advances. I think the development of an AI which is indistinguishable from a human is inevitable; the question is when, not if. Will it be in 15 years, or 150?

    Even if an A.I. could learn
AI can most certainly learn. Machine learning is a fundamental subfield of AI.

    that would make it similar to a human adult
Therein lies the challenge. Electronic computers taught us human intelligence is much more than it appears to be. A hundred years ago, chess was considered the height of human intelligence. But now we know natural language processing alone is exponentially more difficult than chess.

    A.I. will have a very long series of learning experiences
Most modern AI researchers agree with you.

But time doesn't work the same way for computers. Assuming Moore's Law holds, if it takes an AI 20 years to learn to the level of an adult human, 3 years later it will take 5 years, and 10 years later it will take 2 months.

Furthermore, once you have a sentient AI, you don't need to spend another 20 years creating the second. You can simply copy the data.

kleinbl00  ·  3918 days ago  ·  link  ·  

I read an interesting take on the Turing Test recently where (I wish I could remember who) observed that Alan Turing wasn't so much arguing that machine intelligence was possible, but that machine intelligence that imitates human intelligence was good enough as to make the discussion meaningless. It was suggested that Alan Turing was metaphorically discussing homosexuality - and that if he could pass for straight, did it matter that he was gay?

Seen in that context, the imitative nature of "intelligence" shifts the discussion a bit...

briandmyers  ·  3917 days ago  ·  link  ·  

I'm not sure about the gay/straight metaphor, but as I understand it, you are spot-on about the Turing Test; it's simply a thought experiment.

I agree with this article in general, and I believe it's highly probable that we'll have human-equivalent imitation thinking, probably sometime soon, but it won't be true thinking, and we won't learn much from doing it. It's just ELIZA plus plus. I hope (and fear) that I might be wrong.

rob05c  ·  3918 days ago  ·  link  ·  

    machine intelligence that imitates human intelligence was good enough as to make the discussion meaningless

Heinlein's Jerry Was A Man is an excellent short story on the topic of intelligence and rights.

    It was suggested that Alan Turing was metaphorically discussing homosexuality

I hadn't heard that. That is interesting.