Has an AI ever solved a detective mystery?

Not exactly a detective mystery, but according to a slide dated June 2012 from a NSA PowerPoint presentation (see: Glenn Greenwald’s site), NSA used some kind of Skynet AI technology to analyze and detect suspicious patterns from location and communication data in order to create a watch list of suspected terrorists. This helped to track associated members of Al-Qa’ida as well as the Muslim Brotherhood. And I'm sure their AI solved a lot of mysteries and found some controversial figures.

Source: U.S. Government Designated Prominent Al Jazeera Journalist as a member of AI Qaeda

For more details check: SKYNET: Courier Detection via Machine Learning for courier detection data and charts generated by analyzing GSM metadata using machine learning algorithms. Also Applying Advanced Cloud-based Behavior Analytics.

I might be wrong, but I do not believe that something of the scope you describe would be possible with the current state of technology. It would require a lot of things which are still in relatively early stages of research.

For one, just extracting relevant information from text is a huge task by itself. Doubly so with a novel which contains a large amount of unimportant details.

It might perhaps be easier if the input was presented in the form of some sort of list of important facts. But it would still be rather difficult for the AI to connect them and find a solution.

As an example, let's say that we have these two facts:

• Alice died of a snake bite.
• Bob was seen buying a couple of mice recently.

To a human, it seems obvious that the mice were bought to feed a venomous snake. However, it would probably require a tremendous effort to teach an AI to make such connections.

Disclaimer: I don't do text processing myself, so I'm not quite up-to-date on the current state-of-the-art. It's possible that some of these things have already been done in some form. If anyone knows more about the subject, please correct me if I'm wrong.

Generally agree with @Inquisitive Lurker, but I think we also have a wide range of potential abilities/requirements. As with computer chess or Go, where there's a big difference between "beating an honest novice human " and "beating all humans"; there's a big difference between solving a simple kids' mystery and a complex adult novel.

So I don't think there would be any problem writing a program that could solve a problem that is listed as a list of statements, or laid out as a (very young) children's book. However something like an Agatha Christie or John Le Carre's "Tinker Tailor Soldier Spy" (relatively simple solution, but the story is told in a complex manner) are far in the future.

Sometimes an alternative approach might work. For example a neural network could probably solve all Colombo mysteries at the "Who did it?" level without a full "Why?" explanation, after only reading a few Colombo mysteries. The same is true for most kids!