Doing a thorough reckoning of the impact of AI on humanity is hard.
There’s the rapidity of the current tech wave, and the reality that we’re really kind of in the middle of a big change. Then there’s the complexity of trying to weigh the pros and cons of something so vastly powerful.
“With its growing influence comes an increasingly urgent question: Is artificial intelligence a blessing or a curse?” writes Muhammad Tubin at ScienceNewsToday. “Like all technologies, AI brings a mix of promise and peril. It can empower, liberate, and enrich lives—but it can also disrupt, displace, and manipulate.”
That fear of disruption, displacement, and manipulation is driving quite a bit of angst, globally. You don’t have to read the entire article to end up with a face like this one pictured at the top of “The Economics of Superintelligence” posted at The Economist last month. But there’s also promise and opportunity. I’ve recommended setting up some kind of SWOT analysis, like the devs do, to try to put things in neat categories (here’s something in this vein by our own Adrian Bridgwater).
Talking Tech
There’s some accounting of this in a recent panel at Imagination in Action’s Stanford Event earlier this month.
I talked with Anna Makanju of OpenAI and Erik Brynjolfsson, a Stanford professor, about their experience and views on the future of our human collaboration with intelligences that share our environment, but not our build, or origins.
Both admitted some amazement with what early AI systems can do.
Brynjolfsson talked about starting a company called Foundation Technologies, with a nod to Isaac Aasimov, and about using AI to write comments for a paper.
“That was sort of my aha moment, when it was like, these amazing poems about this economic paper,” he said of the latter foray. “And I was like, okay, that’s just too cool.”
“I was coming out of eight years in government, by no means a technologist,” Makanju said. “But even at that moment, I thought, this is incredible. It is coherent, it answers my questions.”
She also talked about her experience working in the U.S. White House, and how AI could apply to things like CIA briefings.
Ongoing Work with AI
Makanju then discussed the work that she does at OpenAI, and how that factors into all of this.
“OpenAI’s mission is to ensure that AI benefits all humanity, which is, you know, an incredibly ambitious approach,” she said, noting a UNICEF program that streamlines education as an example. “We have (a program) with UNICEF that builds digital, accessible textbooks all over the world, in many places where there is low connectivity or low penetration of digital devices, and they are able to create textbooks at 10 times the speed and reduce the cost by 10x, so basically, working with organizations that are serving these communities. How do you empower them and enable them to really serve the people they work with?”
Opportunity Surplus
Brynjolfsson went into what I would call an “opportunity surplus” of AI as opposed to a cost.
“A lot of the benefits from AI are not being measured in any way,” he said. “In economic statistics, (there’s) this concept of consumer surplus, how much value people are getting versus how much they actually have to pay. And when you’re paying zero, you know that consumer surplus can be a pretty big number.”
Indeed: he estimated the consumer surplus of OpenAI’s technologies at around $100 billion, or in other words, no small potatoes.
“I’m going to have to rethink our economic statistics to account for all these digital goods that are creating a lot of free goods,” Brynjolfsson said.
Disruption, he noted, would generally be in the “cons” category.
“There’s an emerging pattern there where a lot of job categories are changing course, and falling a lot,” he said. “And then … there was this report that just came out. … 911,000 fewer jobs created in the past year or so than we had previously thought. They did a huge, huge revision to the jobs numbers. … despite firing the BLS Commissioner, there are still changes and revisions in the job numbers.”
Waiting Impatiently
The duo also spoke to the idea that some business leaders are chomping at the bit to be able to factor actual AI value into business processes, noting the oft-cited “productivity paradox” where new IT improvements don’t get plugged into the business accounting right away.
“(There’s) a lot of disappointment and growing anger that the benefits aren’t turning into business value,” Brynjolfsson said. “But as you just mentioned, that’s actually a very common pattern with general purpose technology. Electricity (for example) … you have these eye-popping capabilities, but getting them to actually turn into business value often takes much longer. … I think we may be seeing significant productivity gains in … the second half of the decade. … I’m optimistic that the Congressional Budget Office and others have (greatly) underestimated how much productivity we’re going to see over the next three or four years, and that would obviously raise living standards a lot, but also create a lot of disruption. And so we have to prepare for that.”
Makanju cited user disconnects.
“When you actually dig into the study, what you see is that a lot of them had no idea how to engage with the tool, so they were just given the tool,” she said of a recent experiment to measure adoption. “They weren’t told anything about how would you use it, or how would you integrate it into this workflow.”
On the other hand, she provided an example of how a clinical AI copilot project led to an 18% improvement in diagnosis in a medical environment.
“I do think a lot of that even … at any … large scale sort of entity that’s trying to integrate it, (the goal) is figuring out what is that workflow going to look like to actually get those gains,” she said. if there is one field where I feel like it’s just going to be incredibly transformative, transformatively positive, it’s healthcare.”
Don’t Make Them Like Humans
There’s a lot more in the video, about innovation, about economics, and about past figures like Alan Turing, Marv in Minsky and John McCarthy.
But I wanted to focus in on a couple of points that we talked about near the end of the segment, where Brynjolfsson suggested that the community should be looking more at the specific strengths of LLMs and AI systems, rather than trying to make “replacement humans” or, as he put it, “imitative AI.”
“I think there’s too much of a focus on automation,” he said. “What new thing can we do that’s never been done before?”
He summed up some of the dangers of making imitative robots this way:
“You make a machine that perfectly imitates humans,” Brynjolfsson said. “From an economist perspective, that means it’s a substitute instead of a compliment, instead of (augmentative) cognition. What do (substitutes)do? They drive down the price of the competition. That tends to lead to lower wages, more concentration of wealth and power. … for most of us, that’s not the world we want, where humans don’t have much of a role, and the machines are able to make all the scientific discoveries.”
Three Questions
Leading on from this, the segment closed with three pointed questions to both panelists, the first being: What keeps them up at night?
Brynjolfsson and Makanju shared concerns about authoritarian governments using these tools inappropriately.
“I’m from Russia,” Makanju said. “I have many friends who are in jail or unable to return, and I have, in my mind, constructed many versions (of how) the Russian government could use (AI) to further scale the current level of oppression. And we know these tools are already being built.”
“I share Ana’s concerns about AI being used to concentrate wealth and power,” Brynjolfsson added. “(Artificial) intelligence may reverse or undermine some of what Friedrich Hayek wrote about when he wrote this classic paper called The use of knowledge in society: that awfully important knowledge is widely dispersed in society. That leads to power being widely dispersed in society. And he said, there’s no way one entity could have all that information and knowledge in one place. And he was certainly right for most of the past 70 plus years.”
Will that continue?
“It’s increasingly possible that big data centers … can get very detailed information about all consumer preferences, supply chains, etc., aggregate it, and make organized, coordinated decisions,” he added. “And that might be very efficient, but it also leads to this concentration of power.”
Check out the video for more. These are the kinds of conversations that we have to have, right now, as we ponder how new AI gets integrated into our societies.
