I'm trying to understand the difference between weak and strong AI. For an example, let's say we would pass the turing test - would it show strong AI or weak AI then?

I don't believe that this is standard terminology, but more philosophical. It was mentioned by John Searle in his "Chinese room argument". As I understand, strong AI is about computers really being intelligent such as having a mind and thus a conciousness, and weak AI refers more to computers being able to simulate the behaviour of human intelligence on only specific problems (think chess, etc.)

Now, the question is - if we would be able to pass the turing test, would it be called weak or strong AI then? Could it be strong AI due to the fact that the turing test is not limited to a certain area or a specific problem?

I came across it on wikipedia: http://en.wikipedia.org/wiki/Chinese_room

  • 1
    $\begingroup$ I have no idea if this is standard terminology. In case it's not (and maybe in any case, just to be safe) you should give a reference, that is where you found the words. Also, does this source contain a definition? $\endgroup$
    – Raphael
    Aug 21 '14 at 10:00
  • $\begingroup$ You are right, I don't think it's standard terminology. I've edited the post. $\endgroup$
    – Regnard
    Aug 21 '14 at 11:02
  • $\begingroup$ I am not sure everyone can pass the Turing Test. Are they weak or strong NI? $\endgroup$
    – babou
    Aug 21 '14 at 14:54
  • $\begingroup$ a worthwhile question here & also for Psychology & Neuroscience $\endgroup$
    – vzn
    Aug 21 '14 at 23:39

The usual distinction is that strong AI, also known as artificial general intelligence, refers to systems that can perform arbitrary tasks requiring intellect, whereas a weak AI is only able to perform specific intellectual tasks. So, for example, a computer chess program or a system that was able to compose interesting music would be a weak AI, since it only does one thing.

Passing the Turing test would indicate something close a strong AI. Although one could argue that it only requires performing a specific intellectual task (i.e., conversation), that conversation could range over just about anything that humans can intellectualize about. Indeed, the Wikipedia page I linked above mentions the ability to pass the Turing test as a component of an operational definition of strong AI.

  • $\begingroup$ @Regnard It's probably best not to accept an answer so quickly. My answer's is rather short so there's a good chance somebody will come along and post a better one. (On the other hand, if somebody does post a better answer than mine, I believe you can change your mind about which one you accept.) $\endgroup$ Aug 21 '14 at 13:28
  • $\begingroup$ At least your answer was clear. Then we have some humans answering more or less the weak AI definiion, And there is the issue of how many thing you should be able to do to be considered strong AI. Maybe open-endedness is the point. All I know is that whatever it can do, there is a TM that does better. But this may be considered metaphysics. $\endgroup$
    – babou
    Aug 21 '14 at 14:56

it is worth mentioning some of the historical/ philosophical debates behind AI that have persisted for over a half century. AI as a field has absolutely/ unequivocally made major progress and evolution in sophistication over many decades. however it has succumbed to early hype and optimism about its achievements. so in some ways it has not been an "even" advance.

there is a so-called moving the goalposts problem in AI. not sure who first originated this observation but it can be found in many papers. it is mainly levelled by AI proponents in favor of AI research and they have stated that critics argue that no matter what is achieved by AI technology, it is not human. there is clearly some validity to this observation. so there is some psychological component to attempting to measure AI progress that tends to normalize/ minimize significant advances.

so in short "weak" and "strong" AI are informal notions mainly used in philosophical/ meta analysis of AI. the problem with AI is that there are no hard-quantitative ways to measure it. even the Turing test with its dressings of impartiality (of so called "judges") involves a large amount of human subjectivity. unfortunately Turing did not seem to reference this subjectivity much in his original statement and its in some ways "haunted" the field ever since.

if any hard-quantitative way of measuring AI were formulated, it would probably fail as far as anyone can tell. there are quantitative ways of measuring the performance of AI related applications, eg say chess scores. and computers did very poorly at chess originally, and those scores improved to become superior to humans. in earlier philosophical writing, some AI critics did accept that chess or game playing was a sign of intelligence, but that concept melted away after computers successfully played chess and there seemed to be no "human like" aspect to their operation or play. on the other hand some chess experts felt that in the legendary Kasparov/ Deep Blue game, the computer seemed to exhibit a kind of "creativity".

as a particular case study, the "moving the goalposts" phenomenon happened on both sides of this match. Deep Blue became more powerful with software and hardware advances, and Kasparov increasingly complained that the match rules were unfair. Kasparov had long said that a computer would never beat him. when it did, he asserted it was through cheating. so similar caveats apply to the study of "weak" and "strong" AI.

  • $\begingroup$ fyi this IEEE article "Cars May Think, But Will They Achieve Artificial Stupidity?" attributes the "moving the goalposts" quote to Simon $\endgroup$
    – vzn
    Aug 21 '14 at 23:45
  • $\begingroup$ there is also some analogy to "hard vs soft sciences" where more quantitative sciences like physics are "hard" and less quantitative/ more subjective sciences eg like psychology etc are "soft"... $\endgroup$
    – vzn
    Aug 23 '14 at 14:53

I believe in order to really answer your question; it will take many more years of research in neuroscience, computer science, psychology, and chemistry. Even within this thread itself, how can you be so sure that there are intelligent beings parsing the thread who know english. One would have to clearly define what it truly means to know something. Well, we philosophically understand knowledge as something we have personally experienced empirically, which is known as a posteriori knowledge, or a priori knowledge , that is knowledge that is not from direct personal experience. But who can really deny that there is such a mechanism for a computer to experience things in their own way of experiencing, that which is to be experienced.

The true crux of the problem is that we can't experience another conscience empirically except that of our own, that is, with today's current technology, which is also to say a priori. And yet even our own consciences is mysterious to us, because it does not directly present itself to you as you would normally perceive an entity. When you do perceive an entity, you are using your very conscience, and so, how do you begin to perceive that which gives you the ability to perceive in the first place.

What is essential to remember that just because one has not experienced something yet, does not imply that it will not be experienced. Who is to say that there are not even greater forms of AI, then that of our own Intelligence? What would be so inherently special about our species, that we would experience the "greatest" kind of "intelligence" that we have, that could not somehow be emulated outside of the constructs of nature.

I recommend reading the following philosophical book, that I myself have currently started reading. Sein Und Zeit von Martin Heidegger, that is "Being And Time", and there is an english translation for it.

  • $\begingroup$ Down voters please justify why you are down voting. $\endgroup$
    – Gary D.
    Aug 25 '14 at 14:23

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.