Temptation , passion , any ambition and emotions

Recognise a spoken word?

Your tests are not that hard for machines. Look at what a system like IBM Watson can achieve (http://www.ibm.com/smarterplanet/us/en/ibmwatson/).

I would distinguish 2 cases:

Case 1 : You provide a list of typical questions and answers and you allow a programmer to implement an algorithm of his choice to generalize the solutions. Banks and other critical websites are in this case. They want problems that cannot be solved by computers even if programmers make tests and try to implement algorithmic solutions. Pattern recognition problems are often proposed because they are known to be very difficult.

Case 2 : You do not provide any example. This is the one shot exercise. In this case, most of your examples would be appropriate, because programmers don't have time to establish a list of all possible problems and it is difficult to generalize, though some theoretical background exist (see "general problem solving" and "automated reasoning" in Wiki).

Now here is the typical question I would ask for the Turing test:

"Tomorrow I will have a dinner with Napoleon Bonaparte. Is that true or false and why?"

@JMS IBM program Watson can solve this easy since he won in Jeopardy!

There are many question answering systems like IBM Watson, Wolfram Alpha, Bing, and many others for specific domain... they would answer very easily, for eg:

www.wolframalpha.com

It is vital to understand what it means to say that systems/programs like IBM's DeepQA implementation Watson, Wolfram|Alpha, Google translate, etc., can answer anything. To assert, for example, that computers comprehend anything whatsoever is no different than asserting that a pocket calculator understands math. Every question, problem, challenge, etc., that we are able to use a computer for we can do so only by stripping any need for comprehension and reducing the issue to the equivalent of using a calculator for arithmetic. Computers are physical instantiations of an abstract algebra (Boolean algebra) that is itself a particular formal system equivalent or reducible to classical (symbolic) logic. The logic gates of a computer are physical instantiations of logical operators from one's intro to mathematical/symbolic logic course or (at least some of) material covered in a mathematics of computer science course,

It is no coincidence that many of the names of computational intelligence paradigms relate to dynamics of living systems, from swarm intelligence to genetic algorithms. We have become very adept at imitating the kind of adaptions plants, slugs, single-celled organisms, etc., are capable of and exploiting what computers are so fantastic at: computing. In other words, we formalize what might in cognitive or psychological sciences be called procedural memory/learning (i.e., without any awareness, such as the kind of memory that enables one to drive/walk/bike to work while thinking about the presentation one has to give and unthinkingly making a left turn here and a right turn there because the route is so familiar). It is the kind of "learning" that enables a Venus fly trap to react to a fly landing in its "jaws", or for the sea slug to become habituated to stimulation of it's gill/siphon. In other words, it is absolutely and completely bereft of anything remotely recognizable as comprehension. Mice vastly outstrip our best AI systems. The reason it appears otherwise is because humans set up vast databases specially programmed so that systems like Watson can be fed input and, just like a calculator, compute an output with no more understanding of the question than a calculator is of algebra.

The central problem is that we do not yet know what it is that enables some tiny number of living systems (namely, those with brains) to be capable of conceptual processing rather than simply syntactical. It was sometimes tacitly but often explicitly assumed that the difference between comprehension and algorithmic/syntactic manipulation was quantitative, not qualitative (and this is still a position some hold). However, after 60+ years of more of the same, it seems apparent that there is a qualitative difference. Yet defining or determining this difference is much like defining meaning itself: we understand concepts by relating them to things we understand, but the entirety of mathematics, all computer programming languages, and the very design of computers themselves are built around formal languages which remove to the extent possible any and all semantic content reducing everything to syntax. How do we formally define meaning or any concepts when formal definitions are designed to do the very opposite?

I would say, understanding a joke (hence ambivalence and ambiguity, plus double-meaning…), and, as a consequence, laughing.

We can name other things as understanding metaphors and poetry.

I believe that it is possible for a computer artificial Intelligent agent (AIA) to answer almost all of the questions posed above, with the possible exceptions of a hard N_P problen, and perhaps some of the Hilbert Problems. I will have to check them out further. Jokes and laughter, and recognitions of them could be included in an AIA but it would be a waste of time to do it.

What a language-based AIA cannot do is reason or infer by analogy. That is "Since A is like B, the A may react like B when B is operated on. The problem is in the "A is like B" In human thougt, the like items may be quite dissimilar, but still be like is some way, and is seems impossible to put codable limts on what items are alike.

A language-only AIA would not be able to deal with spatial or graphical concepts because a picture is worth more than 1000 words. However, if a AIA had a way to input graphical input; camera, scanner... then it could manipulate that input. Facial recognitions, construction of 3D space given offset 2D images (two cameras), motion detection. These are all currently doable.

Another possibility would be curiosity. How can an AIA exhibit curiosity. I think this could be done by including imperitives, or drives, at the upper level of control. One drive would be "I want more information." This could be keyed when it came across a word in it's input that was not in it's working vocabulary, so it would start searching the internet for more information about it. I if were dealing with a user, it could say "I'm sorry, I don't know that word or concept. Could you explain it to me."

In it's drive for information, it could search the web, continuously as a beckground activity, even have multiple search-bots working at the same time, and when it came across something new, it would concentrate one bot on that topic.

Another possibility would be creativity. Can an AIA create a picture, a new concept. In can create new knowledge by inference from its knowledge base, putting together two itmes that are not generally related and but finding there there is a meaningful relationshiip. It could be solving a murder mystery type problem, as Hecule Poirot did, by putting together disparate facts. But this is not creating something totally new, such as inventing a better mousetrap totally unrelated to prior knowledge, based on the characteristics of mice, a knowledge of mechanisms. In the field of design software, there may be some features that can be transferred to an AIA, but it is not immediately how it could be implemented. The big problem is setting up the specifications for a design. This is normally done by a human agent; could it be done by an AIA?

Winograd Schemas are a good example on questions that are hard for a computer to answer but easy for a human, but I see some obvious problem that needs to be fixed if we want to use those kind of problems in HIP technologies, 1- problems are generated by hand. 2- the answer-space is to small, (leaving the option for the computer to guess)

I fear that any kind of questions generated from a specialized domain will sooner or later be outperformed by a computer, and generating questions about concepts in a wide range of domains is too hard to achieve. This will probably lead us to resolve this issue using other means. Like, for example, removing anonymity from the web. (changing the problem to verification of ID). And having a registry for bots that operate on the web (non-human ID)