Artificial Intelligence in General – Hopes and Expectations Both Behind and Ahead Of Reality

In some ways, AI is excelling, with a shoulder-high pile of successful specific applications.  Yet in others, such as matching human cognitive abilities, it seems to be motionless.  What do I mean?

First, “It’s Smart, But for Now, People Are Still Smarter” (Cade Metz, The New York Times, May 25^th).  This Sunday piece by this paper’s most prominent AI writer, subtitled “The titans of the tech industry say artificial intelligence will soon match the powers of human brains.  Are they underestimating us?,” addressed the coming of artificial general intelligence (AGI), which OpenAI CEO Sam Altman had told President Donald Trump “would arrive before the end of his administration,” and Elon Musk “said it could be here before the end of the year.”  This AGI “has served as shorthand for a future technology that achieves human-level intelligence,” but has “no settled definition,” meaning that “identifying A.G.I. is essentially a matter of opinion.”  As of the article’s time, “according to various benchmark tests, today’s technologies are improving at a consistent rate in some notable areas, like math and computer programming,” yet ”these tests describe only a small part of what people can do,” such as knowing “how to deal with a chaotic and constantly changing world,” at which AI has not excelled.  There is no clear reason why it should be able to jump from huge specific competences to matching overall human intelligence, which “is tied to the physical world,” and “that is why many… scientists say no one will reach A.G.I. without a new idea – something beyond the neural networks that merely find patterns in data.”  In the four months since this article came out, I have seen no signs of any such notion. 

On a related point, De Kai wrote on “Why AI today is more toddler than Terminator” (freethink.com, June 9^th).  That scarily predictive 1980s movie series provided the blueprint in many people’s minds for how AI could be relentlessly effective at what it called “autonomous goal-seeking,” at which it was amoral and lapsed into evil, and has underlaid people’s fears about it ever since.  Yet AI “relies less on human labor to write out digital logic and much more on automatic machine learning, which is analog,” meaning that “today’s AIs are much more like us than we want to think they are,” and “are already integral, active, influential, learning, imitative, and creative members of our society.”  Overall, “AIs are our children.  And the crucial question is:  How is our – your – parenting?” 

A new major ChatGPT release is gigantic news in the AI industry.  Soon after the last one, we read “How GPT-5 caused hype and heartbreak” (Alex Hern, Simply Science, The Economist, August 13^th).  It did well in some ways, “hitting new highs” by “showing improvements over its predecessors in areas including coding ability, STEM knowledge and the quality of its health advice,” but some users of its older 40 system “have bonded with what they perceived as its friendly and encouraging personality” which GPT-5 did not share, making the change they were steered to “a very personal loss.”  Apparently, the company did not expect that.

A wider-scope issue appeared in “MIT report: 95% of generative AI pilots at companies are failing” (Sheryl Estrada, Fortune, August 18^th).  “Despite the rush to integrate powerful new models, about 5% of AI pilot programs achieve rapid revenue acceleration; the vast majority stall, delivering little to no measurable impact on P&L.”  The study blamed “flawed enterprise integration,” as the software doesn’t “learn from or adapt to workflows.”  Working with vendors functioned better than “internal builds,” and “back-office automation,” such as “eliminating business process outsourcing, cutting external agency costs, and streamlining operations,” fared better than “sales and marketing tools,” which were absorbing half of generative AI budget money.

How reasonable is it to consider that “A.I. May Just Be Kind of Ordinary” (David Wallace-Wells, The New York Times, August 20^th)?  The author contrasted 2023 thoughts by “one-third to one-half of top A.I. researchers” that “there was at least a 10 percent chance the technology could lead to human extinction or some equally bad outcome,” with “A.I. hype… passing out of its prophetic phase into something more quotidian,” similar to what happened with “other charismatic megatraumas” such as “nuclear proliferation, climate change and pandemic risk.”  An April paper by two “Princeton-affiliated computer scientists and skeptical Substackers” claimed that we should see artificial intelligence “as a tool that we can and should remain in control of, and… that this goal does not require drastic policy interventions or technical breakthroughs.”  Given what it has done already, though, “the A.I. future we were promised, in other words, is both farther off and already here.”  With so much available right now and so much going nowhere, that is a better summary of this post than I can write, so I’ll stop there.