It seems like you're looking for a Turing test for a range of internet activity. I don't have much to add except that I'd advise you to include the phrase 'Turing test' in any general searches you do to start your research.

The difference between the two is the quality of classification of various situations, including their similarities (or what Johann called "the context of the activity"). "Is that situation similar to this one, i.e. belongs to the same class, and to what 'extent'?"

Hello Good sirs,

The Turing test indeed. When Kasparov played deep blue, each move is a sort of turing test. However, kasparov mentioned how the machine moves like it's conscious.

So what's the next test? I read Do Androids Dream of Electric Sheep, by Philip K. Dick. In his book he mentions how human beings have an extra capability that AI doesn't have, namely empathy. Nevertheless, how do we ever know that the machine isn't programed in such a way as to add the right response programmed into the AI.

This is very much like the problem of other minds, isn't it? How do we even distinguish your best friend is human?

Here are a few more imediate thoughts on the matter:

-Tell a joke to see if it can understand the abstraction of the absurd world of comedy. This one reminds me of Searl's Chinese room argument. Saerle couldn't under stand the joke, because he was only given instructions to follow. While it was the people that put him up to it that understood what was conveyed by the instructions. The program is all dark inside, unlike those that programmed it.

-Test to see if it can frame rules. Rules, in general are socially designed. Let's say that no one is allowed to smoke in the bar. With all the response possibilities an AI might be calcuating through, what if he merely opened a window, stuck his head outside of it, and puffed on the cigerette? This is the framing problem. A human would know to step outside rather than opening a window. or are we going through a million possibilites. Our thought seems to be fluid and pretty much precise on such simple tasks. Deep blue probably htought of a thousands ways to move its chess pieces. Kasparaov probably only thought about what? 50 moves maybe?

-How are words defined? Humans often define words by pointing to this or that, and so it is named. Would a robot understand the multiple meanings of the word "bank"? There are banks where you go to get a loan, there are piggy banks, there are river banks, there are bank shots in pool.

I think the only hope here is to be pragmatic what is actually conscious and what isn't. We witness people and ai everywhere. So how do we sift through which are conscious? Perhaps the best is to just ask and watch these entities and see if they behave in such a way that makes them seem to be conscious... It isn't a once and for all true statement, rather it's one based on a collection of data on how I think, and relating it to other entities around me who might be consoius, too. What do you think about that? Is it just a way to cop out?

Another difference is on the perception of the environment which determines what input the human or bot will extract before deciding on how to behave i.e. bots will not recognize Captcha whereas humans will have no difficulty.

Amir Karniel, who is on Research Gate, has done some work on the developing a Turing test for movement. Here is his profile (I wish ResearchGate would let us tag researchers!) : https://www.researchgate.net/profile/Amir_Karniel/publications/

The paper you want to look at is: "Towards Perceiving Robots as Humans - Three Handshake Models Face the Turing-like Handshake Test" and was published in IEEE transactions on Haptics.

Most of the suggestions here are based on cultural knowledge as well as physical or cognitive abilities. Would a baby pass any of the tests? If someone has no sense of humor does that make them less human? How about someone who is blind or deaf?

Domestic cats and particularly domestic dogs have empathy and can distinguish their environment. I guess that's why many of us treat them as members of the family.

What is a more interesting question is how an AI starting with a limited set of knowledge (instincts we are born with) is able to independently process new information including behavior feedback into its own version of new knowledge.

The difference between contemporary “bot” decision-making and biologically-based decision-making is that the “bot” will exhibit only a capability for classical logic, while the biologically-based decision maker can utilize both classical and quantum logic. The key to blindly distinguishing between the two will lie in being able to test for the use of quantum logic. Such a test shouldn’t necessarily involve rapid factorization of large random numbers, but rather should look more like a Bridges of Königsberg problem at the level of a complex adaptive system. Many of the tests that are mentioned here seem to be so related.

Fabian,

the Bridges of Königsberg Problem is a mathematical problem. A machine will not have problems even it is not solvable. Machines have no problems with logic but with emotions and feelings. It is hard to manage fuzzy problems and it is not trivial to find analogies. So I would seek in such directions.

I think a lot of people in this forum are addressing 1/2 of the original question . The original post was about "behaviour in web surfing/chat", but most posters are discussing only about the chat part (Turing test, empathy, etc). Regarding the "web surfing" part, I would say that most bots are mostly deterministic and based on a finite state machine (or an equivalent model). You can model past behaviour as a Hidden Markov Model, user teh Baum–Welch algorithm to get the most probable HMM, and test future behaviour. If past and future behaviour is consistent with the most probable HMM, then it is a bot, else it is a human.

I think web surfing is different than driving. It is more like shopping at a supermarket... you go with the goal of shopping milk, but you end shopping other things (impulse buying) because they are in sight. Replace products with links, and you get "impulse" web navigation. Anyway, human web surfing could be more like a probabilistic state machine , while bot surfing would be more deterministic. Studies like "Strong regularities in World Wide Web surfing" were done in the 1990s, when there was no social networks and web contents were less rich in interactivity and more task oriented (email, online forums)

Wilfried,

I was thinking more of a sort of test that requires shifting frames of reference to look at a complex system from different perspectives, and I think that this ties together the above thoughts of Johann Kelsch and Stephen Pedersen. I’m not a mathematician, but in trying to define such a test I was considering something along the lines of a “#P” complexity class problem and then on top of that adding a concomitant determination of the constraints of that problem. It seems to me that you would need quantum logic to do something like that.