Two of the three so-called “godfathers” of artificial intelligence (AI) have recently expressed alarm about it, and the prospect that “this stuff could actually get smarter than people,” reacting to advances with chatbots whose suddenness shocked them. They also signed onto a group warning that “Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.”
But one of them, Geoffrey Hinton, former head of AI at Google, came in for sharp criticism for failing to support critics of more immediate discriminatory harms, most notably Timnit Gebru. She was fired as co-leader of Google’s ethical AI team after refusing to scrap a co-authored research paper on the dangers of large language models perpetuating harmful racial, gender and other biases.
The focus on speculative threats of extinction are a distraction from already present harms, critics say.
And discrimination and disinformation are just part of the picture. There were also sharply clashing reports of massive job loss (300 million full-time jobs in the US and Europe according to Goldmn Sachs) and dramatic economic growth ($4.4 trillion annually for the global economy according to McKinsey Global Institute) Massive job loss amidst exploding wealth. What could possibly go wrong?
Or how could it be set right?
A promising set of answers, organized into eight categories can be found in a 2019 report, “Towards an AI Economy That Works for All,” from the Keystone Research Center, a Pennsylvania think tank, that draws on a broad range of historical, economic and technological research. The report’s title exemplifies the first category “Advance the narrative that we can have an economy that works for all” as a counter to the misleading narrative of “free enterprise” that casts government as a regulatory menace, rather than a protector and the source of foundational funding that nurtured the birth of the tech sector.
Other categories include re-balance the relationship between business and labor, “trust-busting 2.0,” promote stakeholder capitalism, promote work-relevant skills and continuous learning, create enough work to go around, share the AI productivity bonanza through taxes and social insurance, and restore responsive democracy. This broad range of policy goals reflects the comprehensive nature of technological change.
One key point the report makes is that while some economists attribute today’s income inequality to so-called “skill-biased technological change,” favoring more highly educated workers, the relationship between education and skills is murky at best. “We conclude that a far stronger case can be made that public policy (deregulation, including of labor markets) and institutions (e.g., the decline of collective bargaining) explain more of the increase in inequality.”
The report’s premise is that AI is not novel, but rather a new iteration of processes long underway—as old as the industrial revolution in some respects, and the computer revolution that gestated during WWII in others. Indeed, today’s AIs are largely driven by neural networks, a technology first explored in the 1960s that was impractical then for lack of computing power.
Although released four years ago, the report seems more timely than ever. Two observations seems particular apt in light of the current furor. First, it notes, “People tend to overestimate the effects of technological change in the short run and underestimate them in the long run.” Second, it refutes a crucial confusion about AI. “Humans are general-purpose thinkers, problem solvers, decision makers. AIs are not, certainly not today,” it states. “As machines, they remain specialized. It is not so much that a general purpose AI is inconceivable, it is that no one at this point has much idea how one might be built.”
Chatbots like ChatGPT might appear to be general purpose, because they generate language about virtually anything. But generating language is itself a specialty. Ask ChatGPT to take out the garbage, and it’s lost. So it’s no surprise that it hasn’t changed the authors’ thinking.
“The first thing to say is that ChatGPT does not, in my opinion think,” co-author John Alic told Random Lengths. “Nobody quite knows how cognition in humans or for that matter, animals works, but we know how AI’s work. They just digest digital information and regurgitate it according to patterns that they quote ‘learn.’ And that improves their ability to reach prescribed targets,” he said. “But it does not mean that they think.”
Humans and animals “have agency,” he said. “It’s unlikely that AI’s in the foreseeable future will have agency, because that would imply that they actually think something like animals do. But as I say nobody knows how cognition works in detail. We just see it from the outside. We see the results…. The mechanisms are invisible to us.”
He called alarm about AIs threatening extinction “a distraction… a red herring. It’s the everyday damage that AI’s are doing and are poised to do in workplaces that we should really care about.”
Alic and co-author Stephen Herzenberg, Executive Director of Keystone, spoke about their recommendations with Random Lengths. “The recommendations that we argued was most important was rebalancing the relationship between business and labor,” Herzenberg said. “I would say that’s that’s one of the areas where that has been some change since 2019,” although not dramatic. “There’s been organizing at Starbucks, there was some activity of Amazon, there’s the whole Biden administration’s ritual repetition of ‘good union jobs,’ in the context of infrastructure and climate infrastructure,” he noted.
He also pointed to a Gallup poll showing 71% support for unions, the highest since 1959. “I interpret that as there being a broad understanding among many Americans that in the economy and politics we need that rebalancing,” Herzenberg said. But, “You have to translate into action,” Alic noted. The Protecting the Right to Organize Act would go a long way toward doing that. Its labor law reforms would enable millions of Americans who want to join unions to be able to do so. But, “Were not yet close to the stage where we have a scale of organizing and or a political lineup at the federal level where we can actually pass progressive legislation,” Herzenberg underscored, which is why “We’ve still got to do a combination of fighting the people’s hearts and minds, and preparing to try to get to the point that we were at 1935 when we passed the National Labor Relations Act.”
Even so, there’s some notable state-level progress, such as rising fare wage laws eliminating sub-minimum wages for tipped workers. In states like California an Washington, this results in “a pretty profoundly different policy regime in the restaurant sector” Herzenberg pointed out. “15 bucks an hour for everybody is pretty different than $7.25 and $2.13,” but “despite this profoundly different policy regime employment growth in this industry has remained robust.”
The report devotes significant attention to skill development—or lack thereof. Other countries have proactive retraining programs, set up ahead of time to retrain displaced workers. It’s not that we don’t know how to do the same, however. “The success of the US military in building the skills of service women and men on tasks ranging from troubleshooting complicated electronics systems to cyberwarfare to emergency medicine” is an excellent model that, like apprenticeship “integrates classroom learning and applications,” mostly with non-college graduates.
Things are looking up in some respects, Herzenberg observed.
“There’s a bunch of activity in different states around apprenticeship, including to grow apprenticeship and occupations in industries where it hasn’t existed much in the US, and also bring it back in manufacturing, where it was somewhat more significant decades ago, but is not very significant now,” he said, “But, “What’s depressing frankly is … there’s not that much of a creative vision the least I can see around the diagnosis of the weakness of our training and learning infrastructure connected to actual jobs and careers. The challenges are not that hard to document,” he said. “But there’s not much big picture policy thinking or experimentation around fixing that problem .”
Still some progress has been made, in contrast to the category of meaningful job creation. “There was there was a pretty live congressional debate around a revived Civilian Conservation Corps,” Herzenberg recalled, but it was killed by West Virginia Senator Joe Manchin, even though “it would’ve benefited West Virginia, which has a horribly low prime male employment rate, more than anyone else.”
This points directly to the importance of the last category: restoring responsive democracy, summed up in the report as “Reverse slide toward oligarchy; build grassroots political power to implement policies in other categories.” Almost all the policies outlined in the report have broad public support, or at least build on principles that do so. But without organization, public support is, unfortunately, meaningless. It takes years to build, but it has been done in the past. Which means it can be done again.