Is any AI program ever going to pass the Turing test?

Question

Is it possible, likely, and what are some links to the best AI work online? All I really know about is http://www.20q.net and that's not really Artificial Intelligence exactly. I'm talking about something that can ask an answer question during a conversation in a relevant way.

Answer 1

How do you know your not communicating with a computer AI right now? :)

Answer 2

Yes. Eventually. However, a machine will pass the test long before AI ever exists or before it actually understands what it's talking about. All it has to do is be able to fool a human for a certain amount of time. It doesn't have to understand what it's talking about. Although, as long as the machine doesn't understand what it's talking about, a human will always be able to expose it, given enough time. So the question is really how long it can keep a human fooled.
There are programs out there that can appear human for a little while, and say things that are "relevant", but again, these are just computer programs, and generally use such simple techniques as picking out keywords and then forming basic sentences using those keywords, so it seems like it's "talking about" the same thing you are, or whatever.

Answer 3

Given the Turing test as well as its common objections, and given your specific requirement ("ask and answer questions during a conversation in a relevant way"), the answer is "Yes, it is possible." When it will become available depends upon the quality of conversation. Drill-down menus already ask and answer relevant information in a pseudo-conversational way. But having the quality of "Data", as in Star Trek: the Next Generation, such high communicative abilities remain in the future (though possibly within the 21st century).

First, the ideal system will need to be a "control system" with a feedback loop. (Even evolution is essentially a "feedback loop".) Next, the ability to store and retrieve feedback would be crucial (memory). Then a "value system" with the ability to apply varying degrees of positive or negative value for all stored memories would be critical for making decisions and risk assessment. The decision process would hinge on competing values against each other. (Effectively, this is what is occurring now in the nRT of human brains - see ERTAS by Baars, Newman, & Taylor as well as Emotive Energy by Keeran.) The rudiments of such a system have been created by Ron Blue & colleagues with the Ricci robot (see http://u2ai.us/ricci.htm). However, Ricci is a far cry from passing your stipulations. Ricci is only a first step in the right direction. With your requirements, you are asking for virtually a nuclear submarine or a space shuttle. Ricci is a mere rowboat.

Even so, remember that we all started this journey as single-celled organisms. Thus, Ricci is a critical and successful first step.

Answer 4

There are already instances of the turing test being passed. Chess games over the internet more or less qualify for the turing test.

Also, book list recommendations from amazon might qualify, or at least could if they dressed up the lists a little bit,

There are a lot of programs in use that can pass the test most of the time.

For sites try other search engines, not just google, Also try sites at MIT, CTI and so on.