Ask ChatGPT... How many AI stories can El Reg create in one day?
I guess it must be getting engagement, but it's getting a bit tiresome now.
The hype cycle must be about to burst?
Using AI in science promises to add to problems in reproducing important results, the UK's highly prestigious Royal Society has warned. In its report "Science in the age of AI," the 350-year-old institution argues that introducing AI in scientific research created barriers to reproducibility – the idea that a particular result …
Depends. If it's simply a common public delusion, then it persists until a notable number of opinion-formers announce that the Emperor is in fact wearing no clothes. If it's more deeply embedded, and being driven by those who are throwing money at it, then we're two years into an investment mania, and they last 5-6 six years.
"The hype cycle must be about to burst?"
One can only hope ... *but* ... I think it is too late, too much money has been spent to walk away at this stage !!!
A good disaster would be just as good as walking away *but* as usual 'Joe Public' will pay the piper ... the hard way !!!
[My 'AI' metaphor mixer has just 'choked to death' .... sort of proves my point !!!]
:)
The reproducibility problem is not new. Until research institutions are compelled to make available _all_ software and other information used to generate their results, this issue will never be solved. I can see _many_ obstacles to this, but that is precisely what needs to be done.
The basic obstacle is that even a half-hearted attempt at this (and I have tried) is huge PITA. It can be achieved in high-profile cases if you throw lots of resources at it. But everyday scientific poking and tinkering would basically halt. And that would be the end.
I am much more afraid of the deluge of ‘papers’ which are basically faked using generative AI. Many have adopted the shotgun approach, betting on getting some of their nonsense past peer review by chance, long before current LLMs appeared. And now they are able to produce nonsense in much large quantities (and finally also correct grammar).
In the meantime, there's always The Journal of Irreproducible Results ... to definitely be re-instated for just this sort of AI research, that the high-and-mighty Annals of Improbable Research may outright reject!
is surely to just require that anything determined by 'black box'- whether AI or other unknown process- has to be verified with reproducible means.
So chuck a bunch of data into an AI and let it generate a guideline to help steer research. But then actually DO the research. You might miss findings off to the side of the track that AI has suggested, but that's the risk you take by using a shortcut.
This is Science, not Religion; understanding the working is the important part, not just getting the 'right' answer.