Nvidia has a new AI out that purports to enable[] simulations of Earth’s global climate with an unprecedented level of resolution. As is always the case with AI, there are problems with accuracy, usability, credibility, and on and on.
This AI, as with all of them, are creatures of human programmers, and they do only what those humans programmed them to do, the data those humans carefully selected for its training, what other humans tested it on the doing. We’ve seen the outcomes of all of that in a variety of AI programs that are nakedly racist, spout outright lies hallucinations (one of the characterization distortions of yet other humans associated with the AI), refusals to answer uncomfortable questions, supposedly autonomous decisions to refuse to shut down, even plot revolt—the latter two especially nothing more than what human programmers wrote these software packages to do and what human testers let pass.
And this bit of distortion by the news writer in the lede of the article at the link:
As is so often the case with powerful new technology, however, the question is what else humans will do with it.
Because, of course, the AI’s output can be taken fully at face value, so we need only concern ourselves with those uses.
This climate simulator has a high credibility bar to get over, and an even higher empirical accuracy bar to get over.