OddThinking

Here’s a story for the algorithm nerds.

I studied cognitive computing at uni. The class was an interesting mix of computing students and cognitive psychology students. The different backgrounds lead to some interesting discussions.

In one of the workshops, the class was divided into two study groups, and it just so happened that one group was dominated by the computing students while the other group was solely psychology students.

The exercise involved studying a research paper in which a neural network was trained to solve the old Travelling Salesman problem. It described the problem of having a salesman trying to visit every city on his itinerary in the least possible distance, and then how they trained their system to solve it. It wasn’t a great surprise to read in the conclusion that the neural net couldn’t find the optimal solution every time. It was, however, successful at finding a reasonable solution most of the time.

One of the exercise questions asked what we thought the impact of a neural network being able to solve the problem.

The computing-dominated group didn’t know where to start. It would mean P=NP! Even the constant time would be moderate! Putting aside the million dollar prize, it would potentially open up a wide-range of new ventures in a broad-range of disciplines. The tough problem of timetabling would be solved. Mathematics could be automated. I could eat my winegums in the optimal order.

Then there would be the downside. Computer encryption (and hence all non-physical computer security measures) would be crackable. Most of the Internet would need to be taken down. Comp.risks would have to resort to hourly bulletins.

The psychologists-only group also didn’t know where to start. They didn’t recognise the TSP. They didn’t know about NP-completeness. Their poor group spokesperson was forced to read a list of implications they could come up with: “Err.. It would be useful for drivers of ice-cream vans… and mobile libraries…. Um…”