Artificial intelligence, or AI: how good is it?
In the March 7th issue of the journal Nature, Lisa Messeri at Yale and Molly Crockett at Princeton analyze how it can be used stupidly! AI can have the illusion of depth in explaining phenomena, while there are more dimensions to an explanation. AI can have the illusion of breadth, having tested many hypotheses, while actually there are many more, better-framed hypotheses it doesn’t test. It can have the illusion of being objective, testing various ways to view phenomena, when there are many more viewpoints to consider. They then look at four major uses of AI and the hope for each.
Let’s look at one. Can AI be a judge of the quality of scientific papers or proposals submitted for publication or funding? Such review by individual scientists, called peer review, is critical in keeping defective science from messing up the body of knowledge in print or online, as well as in mis-allocating limited funds at national agencies to support proposals for new research. Yes, there can be too many papers and proposals to review and individual reviewers can be biased. AI might do better, if its own illusions can be corrected. AI reviews can be in a “bubble” of supposedly global common thought.
Would Einstein’s relativity have made it past AI review, if homogenized AI had existed back then? The fixes for AI are waiting in the long term.
This has been an outreach activity of the Las Cruces academy, viewable at GreatSchools.org.