Seven Weeks on Artificial Intelligence Progress: Real, Questioned, Disappointing, and Baked into the Investment Cake

What recent substantial contributions has AI recently made?  What big weakness does it still have?  What has happened to its great prospects?  Can we know its true inherent advancement?  And what forecasts do today’s AI-related stock prices include?

The first report is “A sequence of zeroes” (The Economist, July 6^th), subtitled “What happened to the artificial-intelligence revolution?”  “Move to San Francisco and it is hard not to be swept up by mania over artificial intelligence… The five big tech firms – Alphabet, Amazon, Apple, Meta and Microsoft… this year… are budgeting an estimated $400bn for capital expenditures, mostly on AI-related hardware.”  However, “for AI to fulfil its potential, firms everywhere  need to buy the technology, shape it to their needs and become more productive as a result,” and although “investors have added more than $2trn to the market value of the five big tech firms in the past year… beyond America’s west coast, there is little sign that AI is having much of an effect on anything.”  One reason for the non-progress is that “concerns about data security, biased algorithms and hallucinations are slowing the roll-out” – an example here is that “McDonald’s… recently canned a trial that used AI to take customers’ drive-through orders after the system started making errors, such as adding $222 worth of chicken nuggets to one diner’s bill.”  Charts here show that the portion of American jobs that are “white collar” has still been marching steadily upward, and, disturbingly, that share prices of “AI beneficiaries” have stayed about even since the beginning of 2019 while others have on average risen more than 50%.  Now, “investors anticipate that almost all of big tech’s earnings will arrive after 2032.”

“What if the A.I. Boosters Are Wrong?” (Bernhard Warner and Sarah Kessler, The New York Times, July 13^th), and not even premature?  MIT labor economist Daron Acemoglu’s “especially skeptical paper” described how “A.I. would contribute only “modest” improvement to worker productivity, and that it would add no more than 1 percent to U.S. economic output over the next decade.”  The economist “sees A.I. as a tool that can automate routine tasks… but he questioned whether the technology alone can help workers “be better at problem solving, or take on more complex tasks.””  Indeed, AI may fall victim to the same problem which got 3D printing out of the headlines in the 2010s – lack of a massively beneficial, large-scale application.

In real contrast to common concerns, especially from last year, “People aren’t afraid of A.I. these days.  They’re annoyed by it” (David Wallace-Wells, The New York Times, July 24^th).  One issue “has inspired a… neologistic term of revulsion, “AI slop”: often uncanny, frequently misleading material, now flooding web browsers and social-media platforms like spam in old inboxes.”  Some delightful examples cited here are X’s and Google’s pronouncements that “it was Kamala Harris who had been shot… that only 17 American presidents were white… that Andrew Johnson, who became president in 1865 and died in 1875, earned 13 college degrees between 1947 and 2012… that geologists advise eating at least one rock a day,” and “that Elmer’s glue should be added to pizza sauce for thickening.”  Such “A.I. “pollution”” is causing “plenty of good news from A.I.” to be “drowned out.”  With Google’s CEO admitting “that hallucinations are “inherent” to the technology,” they don’t look like they’ll be going away soon.

Even given the disappointments above, “Getting the measure of AI” (Tom Standage, The Economist, July 31^st) is not easy.  One way “is to look at how many new models score on benchmarks, which are essentially standardised exams that assess an AI model’s capabilities.”  One such metric is “MMLU, which stands for “massive multi-task language understanding,”” contains “15,908 multiple-choice questions, each with four possible answers, across 57 topics including maths, American history, science and law,” and has been giving scores “between 88% and 90%” to “today’s best models,” compared with barely better than the pure-chance 25% in 2020.  There will be more, and it will be useful to see how they improve from here.

On the constructive side, “A.I. Is Helping to Launch New Businesses (and Not Just A.I. Businesses)” (Sydney Ember, The New York Times, August 18^th).  A Carnegie Mellon University professor who for 14 years has been having “groups of mostly graduate students start businesses from scratch,” said, after advising the use of generative AI extensively, that he’d “never seen students make the kind of progress that they made this year.”  The technology helped them to “write intricate code, understand complex legal documents, create posts on social media, edit copy and even answer payroll questions.” As well, one budding entrepreneur said, “I feel like I can ask the stupid questions of the chat tool without being embarrassed.”  That counts also, and while none of these are, as a Goldman Sachs researcher quoted in the Wallace-Wells article asked about, a $1 trillion problem that AI could solve, they collectively are of real value.

Is it reasonable to think that AI stocks will roughly break even from here if lofty expectations go unrealized?  No, according to Emily Dattilo, in Barron’s on August 19th: “Apple Is Set to Win in AI.  How That’s ‘Already Priced In.’”  Analysts at Moffett Nathanson, for example, pronounced that, although Apple was “on track to win in artificial intelligence,” the “bad news” was “that’s exactly what’s already priced in.”  I suspect that’s happening with the other AI stocks as well.  If the technology not only grows in scope but does so more than currently expected, share prices may rise, but if it only gets moderately larger, they could drop.  That can be called another problem with artificial intelligence – if enough investors realize this situation, the big five companies above, Nvidia, and others may have already seen their peaks.  Small-scale achievements such as startup business help will not be enough to sustain tremendous financial performance.  What goes up does not always come down, but here it just might.  And the same thing goes for AI hopes.