I'm seeing a lot of answers from people along the lines of "AI is just bits and bytes and ultimately cannot be smarter than its creator because its creator would have to use their brain to make something smarter than themselves, which isn't possible."

It's kind of baffling to me to see these answers, especially in regards to the singularity, on a forum dedicated to AI. There is already image recognition that can recognize objects more accurately than humans, IBM's Watson can diagnose lung cancer at a rate much more accurately than human physicians, and Google's Alpha Go beat the Go world champion, even while experts were predicting that AI wouldn't succeed at doing this for another 10 years.

At the same time, I am completely certain that any of the individual programmers of Alpha Go would not have succeeded in defeating the Go champion of the world. I'm also fairly certain that the Watson programmers would not do better than Watson or a human physician at identifying lung cancer. These are already cases of the AI being more intelligent than its programmer, albeit in domain-specific cases.

Therefore, it seems wholly lazy and uncreative for people to provide such answers that AI cannot be more intelligent than a single creator and therefore human-level AI and beyond is not possible. I think it does not contribute to the discussion.

  • $\begingroup$ It was my opinion against Hawking opinion which I don't agree. Define then 'more intelligent'. Fastest computation or every day problem solving? The 'intelligence' cannot be easily measured. If you think AI can be smarter than human, prove it then. Or disprove opposite. $\endgroup$
    – kenorb
    Aug 2 '16 at 18:46
  • $\begingroup$ I define more intelligent to be more capable. We already see examples of AI programs being more capable than their creators. While I can't prove that an AI can be wholly more intelligent than a human, there are multiple viable pathways already set forth for getting there within the next decade. Yours is also not the first such answer, I already saw multiple ones like it today. $\endgroup$
    – zavtra
    Aug 2 '16 at 19:20
  • 1
    $\begingroup$ I flagged this and cast a vote to close this question because this sort of thing is expressly forbidden by StackExchange rules. Not only is it totally opinion-based and certain to "solicit debate, arguments, polling, or extended discussion," but it clearly stacks the deck in favor of one particular point of view. It's even blatantly offensive, with loaded accusations that the opponents are "wholly lazy and uncreative." $\endgroup$ Aug 10 '16 at 3:09

If an answer is wrong, it should be downvoted, plain and simple. Clearly we want to discourage wrong information, and downvotes are designed to point out incorrect, irrelevant, or otherwise poor content. You seem to have really good examples that show such answers are wrong, so please feel free to mention them in a comment when downvoting!

  • $\begingroup$ Whether or not any A.I. can out-think its creator is a primarily opinion-based question that's likely to stir intense debate. StackExchange discourages those kinds of questions for good reason; however, if that policy is not going to be enforced here, we might as well not make matters worse by automatically downvoting one point of view. Philosophers, theologians, A.I. researchers, biologists etc. can't even agree on what "intelligence" consists of, so how can we define what "wrong information" is for this question? It smacks more of forcing an unpopular minority opinion to keep silent. $\endgroup$ Aug 10 '16 at 3:03
  • $\begingroup$ @SQLServerSteve Perhaps I should have been more explicit in that I don't know the answer to that specific question - I'm saying that if one believes an answer is wrong, the thing to do is downvote. $\endgroup$
    – Ben N
    Aug 10 '16 at 3:46
  • $\begingroup$ I don't think that's true of opinion-based questions. The problem here is a deep philosophical one; it's not like looking up what the right parameter is for a Python function, or how to translate a particular English word into French or Swahili, or the right conditions for conducting a goodness-of-fit test. Most questions on StackExchange have crisp, definite answers of that kind, so downvoting ones perceived to be incorrect is usually right. Here - or in any other opinion-based question - downvoting or closing unpopular answers silences the minority permanently. $\endgroup$ Aug 10 '16 at 4:02
  • $\begingroup$ That is one of the many reasons StackExchange bans opinion-based questions: it degrades the voting system into a popularity contest, which has the secondary effect of preventing minority opinions that could turn out to be the correct answer from ever being heard. if we lived in medieval times, doctors in favor of bleeding could downvote their opponents into oblivion, but we're discussing matters that are a lot harder to reason about than bleeding, which is just a matter of science.The definition of intelligence is a matter of philosophy that has stumped some of the smartest minds in history... $\endgroup$ Aug 10 '16 at 4:08
  • $\begingroup$ In this field, there's always going to be a dangerous bias towards overestimating the accomplishments of A.I.; the voting system should not get intentionally coopted in service of that default outlook. There has to be some form of checks and balances to guard against excessive enthusiasm, but policies of this kind would remove them even before the site is out of private beta. Just how often do you hear anyone in the A.I. community say, "This can't be done" or "We're far behind where we should be" or "this is not 'intelligence'?" Those viewpoints must be expressible without retaliation. $\endgroup$ Aug 10 '16 at 4:13

I'm seeing a lot of answers from people along the lines of "AI is just bits and bytes and ultimately cannot be smarter than its creator because its creator would have to use their brain to make something smarter than themselves, which isn't possible."

I think this argument is a bit unclear and needs some refinement. It is true that AI can indeed be smarter than the creator at certain tasks (AlphaGo being better at Go than the programmers of AlphaGo, for instance). What I think this argument is really saying is:

"AI is just bits and bytes programmed by its creator. The creator would be able to know how the AI works, otherwise he would be unable to create it in the first place. Therefore, the creator can be said to be superior to that of its creation, since the creator can understand its creation."

That seems like a more logical premise. Sure, AlphaGo is better at Go than the programmers of AlphaGo, but AlphaGo's programmers actually knows how AlphaGo operates. This type of argument was made in the paper Creativity, the Turing Test, and the (better) Lovelace Test, which specifically argues that AIs cannot be creative since programmers are able to figure out what their creations (AIs) are doing. Another paper "The Lovelace 2.0 Test of Artificial Creativity and Intelligence" saw this argument as so self-evidently true that it tried to create a weaker version of the Lovelace Test to identify and measure AI creativity.

The programmers, basically, know how their program works. That doesn't mean the program is less intelligent than the programmers. Just that the programmers can understand why their programs behave the way they do, given enough time and patience.

Either way, I would not support discouraging answers such as these, if only because this view does have support within the AI scholarly community. If you have experts who hold this view, then we should let this view be given exposure.

  • 1
    $\begingroup$ While I don't agree with the view, having made an AI using an animal's brain mapping and not fully understanding how its decisions were being made, I can support not discouraging it because the scholarly community doesn't yet see things that way. $\endgroup$
    – zavtra
    Aug 3 '16 at 14:44

Not the answer you're looking for? Browse other questions tagged .