This is of course one of the all time great (geek) topics: who will rule in the future – humans or machines? I was reminded of it by discussion of Techmeme’s story selection. It should not be surprising that a single algorithm – in Gabe’s case link analysis – is not always successful at picking out the important stories. One of the things that makes our human brains so powerful is that they can more or less in parallel apply many different algorithms and synthesize the results. When we look at a tech story, for instance, we can quickly relate it to past stories, we almost instantly infer its sentiment, we triangulate with our perception of the source, etc.
I have a few believes about where we are at now:
First, to improve automated news systems we have to combine many different algorithms. Link analysis, semantic analysis, sentiment analysis, and so on. Doing so is not easy because it requires a probabilistic framework that can process the inputs from multiple algorithms.
Second, even once we have done that, we are likely to still fall short of those human algorithms that are based on deep understanding. But we may do well enough to let individual readers take that last step rather than editors.
Third, in order for the second point to apply, I believe we have to change our attitudes toward automated systems a bit. This will happen naturally for the generation growing up with such systems but will require some work for us older folks. Instead of “dissing” a system when it makes a “silly” mistake (which is really a fairly lame way of asserting our human superiority), we should learn to simply ignore those.
I personally do not see an ultimate barrier to deep understanding becoming part of what systems can do, but I believe it will still take a long time. Until then we can asymptotically approach what humans can do, but probably with significantly higher outcome variance.