How is it that a system billed as “Artificial Intelligence” (AI) can be so fundamentally stupid?

The screenshot above is Goggle version of AI reporting on the 2023 salmon harvest in British Columbia (B.C.) Canada. Canada most decidedly did not harvest 230.2 million salmon last year.

The official numbers from the province are not out yet. But given that more fisheries were closed than open last year, the catch is likely to be closer to the 692,000 salmon the Canadians caught in 2021 than the 2.1 million they caught in 2022, according to the data compiled by the North Pacific Anadromous Fish Commission (NPAFC), a treaty organization.

And no, there are no typographical errors there; no missing zeroes. The number for 2023 is not 20.1 million let alone 200.1 million. It is 2.1 million.

Canada has never seen a salmon harvest anywhere close to 230.2 million.

Forget about 230.2 million being the fourth-highest on record. A 230.2 million salmon harvest in Canada would be something beyond unheard of it. One might fairly consider it impossible.

The actual, fourth-highest salmon harvest in B.C., according to the NPAFC records going back to 1925, was 39.5 million in 1986.

The very biggest of the three larger was 41.7 million in 1985. That’s it.  The Canadians have never, ever caught more salmon than that. So the largest-ever Canadian harvest was less than a fifth of 230.2 million reported above.

This is not a difficult thing to look up. And not only that, Google’s AI itself links to a report from the Alaska Department of Fish and Game for details on the B.C. harvest.

Geographically challenged

Apparently, it has been taught or “learned” that Alaska is part of Canada and not the United States.

Oh for the good old days when computers were simply “programmed” to know the difference between countries, states and provinces.

OK, the ignorance here does admittedly make AI more “human” like.

There are more than a few poorly educated people also of the belief that Alaska is a Canadian province and even more whose understanding of geography is so limited they fail to recognize that Alaska is part of the “continental United States.”

One of the latter, Utah’s Sabriena Abrre managed to collect 850,000 views on TikTok by admitting she is among the foolish. One can only guess she’d never looked at a globe (do schools still have those?).

She blamed maps of the United States that put Alaska off in one corner with Hawaii (a group of islands for those who don’t know) and the American educational system for her ignorance. And her confession led many others to empathize by revealing their own blindness as to Alaska’s attachment to a giant land mass, according to the British website Unilad.

Euros were not quite as sympathetic to Abrre’s ignorance. “Like do you have internet there?” one asked. “Google maps? Are you not curious? Did you ever see a globe? So many questions.”

But hey, the lack of knowledge among humans is a sometimes good thing. Human fallibility has led to some great discoveries.

Scientists at the drug company Pfizer were trying to develop a drug for angina, ie. chest pain, when they created something called sildenafil, now better known as Viagra. It quickly hardened Pfizer’s drug sales and his since risen so far as to be described as a “blockbuster drug” by Science.

Still, the idea behind AI is that was supposed to be less fallible than humans.

Modern society doesn’t actually need AI-driven taxis dragging injured pedestrians along city streets when human drivers are perfectly capable of doing this on their own. 

And Google, for its part, seemed to have pretty good algorithms enabling one to search the giant library that is the internet before it went off on this AI kick, which now sort of pops onto the screen like a drunk librarian offering assistance.

Usually, I admit, I just scroll down past the nonsense and start going through the old-fashioned Google links looking for what I need. But sometimes, as in this case, the AI results are so messed up that they are hard to overlook.

OK, so maybe it is unfair to compare them to a drunk librarian. Drunk librarians only make information hard to find. They don’t twist it.

Google’s AI-generated responses are more like the work of a really bad journalist, one of those clueless about the subject matter on which he or she is reporting but happy to pull together random facts from here and there to create a good “story.”

Like in 2019 when The Nation magazine had on its cover: “THE LAST SALMON: Bristol Bay is one of Alaska’s few healthy salmon habitats. Can it survive the twin threats of a rapidly changing climate and a massive Trump-backed mine.” 

Three years later that changing climate delivered the Bay an unprecedented, record harvest of 60.1 million sockeye salmon – 15 million more than the previous record dating back to 1995 and 81 percent above a 20-year average inflated by global warming in the new millennium.

Scientists had fingered warming as a plus for the Bay early in the 2000s and by 2019 were reporting in the peer-reviewed Nature Ecology and Evolution that the numbers of Bay salmon were increasing, not decreasing, due to “climate-enhanced growth opportunities.”

This could, of course, change at some time in the future. Almost anything can happen in the future, making the future impossible to predict. But you’d never have guessed this from reading The Nation.

The reporter who wrote The Nation story made the classic mistake, either by ignorance or design, of judging environmental events on a one-year basis instead of on a decadal or multi-decadal basis. Drawing broad conclusions from what might be a one-time, isolated event is just plain foolish.

“Scientists have been cautious about saying what happened with reds (sockeye salmon) last year, insisting they need more time to study it,” the reporter wrote. “But many suspect the anomalies may have to do with rising ocean temperatures.”

The scientists weren’t being “cautious;” they were being realistic. Natural ecosystems have large degrees of annual variability. Any one year is meaningless. It might mark the start of a trend; it might not.

To date, those rising ocean temperatures have, overall, spelled nothing but good for Alaska salmon, although the same cannot be said for Canadian and Pacific Northwest fish. Their numbers have gone down, down, down as Alaska salmon returns have driven North American coastwide numbers up, up, up.

There are now more salmon than ever in the North Pacific. The only problems are that most of them aren’t the salmon humans most desire, and they are no longer distributed to the places where most people live.

AI is supposed to figure this shit out.

The entire idea behind AI is that it can convert all information to numbers and that by then doing math at the hyperspeed of computers it can eliminate the simple mistakes humans make in in not knowing how to sort information and put it in context.

But Google’s AI sometimes seems even worse than humans in sorting information, and it likes to draw conclusions with which knowledgeable people are likely to argue. To wit:

“Iditarod wouldn’t be considered Alaska’s most famous ghost town. Here’s why:

• Still on the Map: While Iditarod’s population has dwindled significantly since its gold rush days, it still has a handful of year-round residents and functions as a checkpoint during the Iditarod Trail Sled Dog Race. This keeps Iditarod somewhat active and on the minds of people following the race.
• Focus on the Race: The Iditarod Sled Dog Race overshadows the ghost town aspect of Iditarod. The town is primarily known in connection with the race, not its history as a gold rush boomtown.

“Places like Kennicott, with its collection of preserved buildings, or remote Ukivok with its haunting presence, better represent the classic image of a ghost town. However, Iditarod does hold a unique place as a ghost town touched by modern activity due to the famous race.”

As factual matters, there are no year-round residents in Iditarod, and Kennecott (not Kennicott) was once a ghost town but is now a National Park Service-maintained tourist attraction.

Meanwhile, few people, including many if not most Alaskans, will have heard of Ukivok, which is the Inupiat name for King Island, a mile-wide block of rock jutting up in the Bering Sea about 60 miles northwest of Nome. 

The island, though treeless, does still have a long-deserted village on stilts originally built of driftwood. The island and the still-standing village have an intriguing history and prehistory. Human artifacts found atop the island’s plateau have been dated to 1,000 years ago, but the stilt houses appear to date back to only the mid-19th century, according to the work of ethnoarchaeologist Claire Alix.

“…The remaining house structures (on the island now) show the transition from an economy based on driftwood to one based on lumber supplemented with driftwood. In fact, because of its geographical isolation and the King Islanders’ relocation to Nome some 50 years ago,  the old village offers a unique look into this major but poorly documented transition,” she observed in a 2012 paper. 

Ukivok is an amazing place, but here is the problem: It is hard to be virtually unknown and “famous” at the same time. The Merriam-Webster dictionary actually defines “famous” as “widely known.”

The need to be widely known to be famous is a connection Google’s AI – Gemini – doesn’t seem to get. Still, it does do better with most questions than Google’s “experimental” “Generative AI” which now regularly pops up when using the original Google search engines dating back to 1998.

Generative AI might be better described as Degenerative AI. It sometimes seems dumber than toast.