In my last article I wrote about the exciting potential for AI to summarize what is going on in vast swaths of research that are too broad for human comprehension, as most research areas are.
By coincidence one of the world’s biggest scientific journal publishers — Wiley — has just published something on that. It includes a good looking computer generated research summary. The summary is about a medical condition about which I have some sad knowledge.
They call them “synthetic journal articles” which may be as good a name as any.
The article has the usual anti-AI caveats, including worrying about computer generated “fake news” (a term that has lost any meaning). But the point is that the algorithm in question got it mostly right.
The summary looks good. And this is a scaled down version because the creators say they are afraid of releasing the real thing, lest it flood the world with research summaries. I am not sure that would be a bad thing.
The confusion is that they think a review article is research. Research is very specific, normally a tiny result. Trying to say what is going on in thousands of journal articles is not research, technically it is more like journalism.
Mind you journalism is not a popular word in scientific communication, even though there is a huge movement on now, how to increase what is called impact, which sure sounds like journalism. In fact they have view metrics for social impact of journal articles. They even measure tweets. It is called altmetrics. This is very funny in its way. Scientists competing for public attention.
What will be truly interesting is when these review algorithms start to argue with each other. Science is not a pile of facts. It is an ongoing debate about how the world works. One wonders what it will look like when the algorithms debate.
Climate change is a good candidate for algorithm debate, given that there are tens of thousands of articles published on it each year, many contradicting one another.
Energy policy is another obvious candidate. But given that something like three million scientific journal articles are published annually there is no lack of candidates for algorithmic debate.
It might be fascinating to see computers that ingest thousands of articles debating each other. Bring it on.