AI-generated News Articles

This article in the New York Times yesterday points out that machines are now generating summaries of sports games and financial data. StatSheet (now Automated Insights) has also been doing this for years, apparently. Comments of the singularity approaching abound...

My question is: is this really all that newsworthy?

Yes, it's a leap forward in AI-generated content. It reads fairly well, without the repetition and madlibs style of other generated content. But is it really all that different from what we've already been doing? Algorithmically, of course it is. But, merge-sort isn't any more an AI performing a sort than wait-sort. Sure, it's a faster (and, arguably better) algorithm, but it's still just a divide-and-conquer sorting algorithm. See also encryption: same problem; different algorithms to solve it.

What these companies have really done is taken sports and financial reporting and distilled them into an algorithm that works well in most cases. They're taking games with well defined and understood sets of rules, and analyzing past data to summarize them. It's no more approaching an AI singularity than the madlibs are. This isn't to belittle their success, but I doubt we'd be as impressed if they had these algorithms "sportscasting" checkers. I'm sure it could be done ("In a stunning 18th play move, black managed an astonishing triple-jump to retake the lead"), but it would just be an academic paper, not an industry.

So, the question of success becomes whether or not this is useful. The main two features of auto-generated news are timely reporting and SEO. Timely reporting seems to be a red herring for the sports world; if you care, you'll watch the game or listen to it on the radio. Having a summary up in under a minute is a neat trick, but it's only a side-show to the main event. Which leaves the arms race of SEO. Sure, good SEO is important, but it's only a matter of time until Google and Microsoft revise their algorithms to make these contributions less important. All it would take is making sure that content is generated in a "reasonable" amount of time for a human writer. Then the algorithm will have to have wait time added to it, and you lose the speed argument in favor of auto-generated content. At that point, this becomes solely a cost-cutting measure, and that's generally a race to the bottom. So, eventually we'll have companies pop up with madlib algorithms that are dirt cheap, and it's easy to see watching the game as a better use of time for consumers.

Get me an AI that can write a compelling Science Fiction novel and I'll start paying attention.


No comments:

Post a Comment