UV blues in ChatGPT

Trust the new AI App at your peril ....

The new ChatGPT app reminds me of the old “Ask Jeeves” site, which met its demise 2 decades ago. I was younger and (more) foolish then. For fun, I once asked Jeeves “Who is the smartest out of Richard McKenzie and B** L**** ”. Spoiler alert: the correct answer is NOT Richard McKenzie. My contender for the prize was a work colleague who now occupies the office next-door. I did supply Jeeves with his/her name without asterisks, but I won’t divulge it here. He/she might be embarrassed about their anonymity at that tender age - because the simply brilliant ‘answer’ Jeeves came back with was “Who is B** L****?” I confess that I’ve made great mileage out of it - and here we go again!

Hopefully ChatGPT will be just as illuminating as my old friend Jeeves! But will it give the correct answer?

I tested it with the question What is the world’s highest UVI?

It’s a deceptively easy question to ask. But not so easy to answer.

The graph below that I made many years ago is helpful. It shows the calculated clear-sky UVI at sea level (for the average Earth-Sun separation distance) as a function of sun elevation angle for several ozone amounts. As you can see, for ‘average’ ozone amounts (around 300 DU), the UVI reaches a maximum of about 12 for overhead sun, as shown by the blue curve. For ozone amounts near 100 DU, as seen under the Antarctic ozone hole, the UVI would far exceed 20 at large sun elevations; except that at those polar latitudes, the sun never rises high enough in the sky for that to occur (black curve). For ozone amounts near 200 DU - which do occasionally occur at low latitudes - the UVI at sea level can exceed 20 (red curve).

Further, in unpolluted clear-sky conditions, the UVI increases by about about 5 percent per kilometre of altitude increase. So it should reach 25 or so for overhead sun at altitudes greater than 5,000 m (16,400 feet) above sea level.

And indeed it does ….

The largest plausible UVI for clear skies should occur at tropical high altitudes, with the Sun close to being directly overhead, and with low ozone amounts. And probably in the southern hemisphere, where times of overhead sun occur in the same season as the closest Earth-Sun separation. As expected, the correct answer for the peak UVI measured so far - is somewhere around UVI = 25, as seen in South American Altiplano (and also at the Mauna Loa Observatory in Hawaii, where ozone amounts are sometimes less than 200 DU). No doubt those records will be broken in future.

The highest mountains at latitudes close enough to the equator for overhead sun, have peak altitudes around 6 km. According to my uv-calculator, the corresponding highest UVI over non-reflecting surfaces under clear skies at that altitude and at closest Earth-Sun separation would be around 27.

But the highest UVI doesn’t occur for clear-sky conditions. It occurs when scattered clouds that don’t block direct sunlight are present. The clouds are brighter than the surrounding sky, and reflections from them can increase the UVI by ~20%, thereby giving a cloud-enhanced maximum UVI greater than 32.

Above surfaces of high reflectivity (albedo) like snow, it can be even higher. Although snow-covered surfaces are unlikely under these peak UV conditions, highly reflective cloud decks, which are sometimes present beneath mountain observatories, like Mauna Loa in Hawaii, can increase the UVI by ANOTHER 20%. Taken together, short bursts with UVI greater than 38 should be possible. There may already have been (as yet unobserved) occurrences of UVI greater than 35.

Now, back to ChatGPT. I thought it might (erroneously) report back that the answer is 43, from a now-discredited study that reported data from a high altitude site in Peru where record values may be expected. But, as we later showed, their reported value was far too high for the solar elevation angle at the time. For those conditions, it’s unlikely that the real UVI would have exceeded 20 - less than half their reported value.

Thankfully GPT didn’t give that incorrect answer, but its answers were just as bad. Here are a few variations to the question, and it’s responses.

Q: What is the highest UVI ever recorded?
A: “The highest UVI ever recorded was at the Arona Observatory in Tenarife, Spain on December 30th, 2003. The UVI was 11.8”
Keh? (as Manuel of Fawlty Towers fame would have said)

Q: World’s highest UV?
A:“The highest known ultraviolet radiation index is 11+”
(this answer probably merely reflects the fact when the UVI was first mooted, it was highly ‘Eurocentric’, so the maximum reported would be 11+ (the highest seen in Europe), regardless of by how much it really exceeded that number).

Q: World’s highest UVI?
A: “The highest recorded UVI is a score of 16.8 in the Chilean …”
Keh? again …

Q: What is the world’s highest UVI?
A: “The highest recorded UVI recorded was 13.9 in Brazil on November 12, 2003.
Keh? again, again ....

Take your pick! All are way too low.

There were a few silly answers to other variants of those questions. The app advises that by investing US$39.99 per year, I can get a more complete answer. Perhaps I can too, but on the basis of what I’ve seen so far, but I’m not prepared to test it at that price.

None of those answers for the peak UVI bear even a passing resemblance to the truth. It’s like the example of ‘verbal’ promises not being worth the paper they’re written on.

Troublingly, the app gives an answer without any information about the source. That’s a serious shortcoming and an invitation for promulgating false news. As we’re all aware, there are plenty of sources of unreliable information out there. I prefer being able to check my sources.

Wikipedia is usually reliable, as long as the information has been there long enough to have been peer-reviewed. The current answer it gives to my question isn’t too bad. It’s a lot better than the answer it gave in 2014 (before our corrections to the bogus result were published), but it’s still a bit generous to the outlier. More details can be found here, here, and here.

Just for fun again, I tried ChatGPT on my first question above to Jeeves. The answer: “It is impossible to answer this question as it is subjective and depends on many different factors”. Boring! And I suspect, less true than what Jeeves came up with two decades ago (though times might have changed 😊). But at least the new app appears to have heard of both of us. That’s progress B**!