I tried a few of these games, that apparently use human inputs to solve problems computers aren’t good at, and I am wondering why there aren’t neural nets built for them since these games aren’t very complicated? How do these games solve problems that computers can’t solve through human input?
In: 2
Building a neural network involves having lots of training data. Sometimes citizen science games are about gathering training data where there isn’t any. For example, if we want to make a neural network detect pauses in brain activity we need lots of examples of brain images that are labelled “pause” or “no pause”. We cannot train a neural network without this training data, so we need the citizens to do the heavy lifting _first_, and then the neural network takes over.
The human brain is also an exceptional system and there are some things it can do that we just don’t know how to replicate yet, even with our best neural networks. Language problems are one example of this – even the hugest and most accurate neural networks will still make mistakes that humans won’t on certain problems. This is also the case with protein folding – trained human citizens are still better than supercomputers. However, both of those tasks are seeing rapid progress in machine capabilities.
Latest Answers