THE GLOBE AND MAIL. SOURCES: GETTY IMAGES
Daron Acemoglu is Institute Professor of Economics at MIT. Simon Johnson is the Kurtz Professor of Entrepreneurship at MIT’s Sloan School of Management. They are the co-authors of Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity, which is a finalist for this year’s Lionel Gelber Prize, presented by the Munk School of Global Affairs & Public Policy.
In Neal Stephenson’s 1995 novel The Diamond Age, a technological breakthrough allows young people everywhere the opportunity to acquire the best possible education at essentially zero cost (at least to them). The consequences, as you might imagine, are profound. Instead of being limited by broken-down school systems with underpaid teachers, all human knowledge is available to anyone who wants it – with implications for productivity, democracy and, most of all, what humans can achieve.
Now, almost 30 years later, some people argue we are on the brink of what Mr. Stephenson so brilliantly imagined. The latest and next versions of generative AI are interactive knowledge engines. Ask any question and you will get a detailed and, in the most recent versions, well-documented answer.
Unfortunately, there are many flies in this ointment.
The first problem is obvious. More people around the world need access to the technology. The hardware component of this revolution is non-trivial and will cost real resources.
The second issue is even more important, but tends to be completely ignored by tech visionaries. Part of what we know can be written down, and these are the things we learn from books and from discussion with our teachers and peers. But a lot of what we know comes from our experience in the world. The idea that brain development is entirely detached from physical interactions with others will strike any parent as unrealistic. We learn by doing, by making mistakes and by watching others.
“Experiential learning” is not a nice-to-have or something that can be left for college study or even graduate school (e.g., work with companies, intern with government). Rather, we need school systems everywhere to pivot to provide these experiences, at the same time as integrating real-world lessons with what AI-powered tools can help students further explore.
Consequently, there is (unfortunately) no Stephenson-esque “scaling up for free” on the horizon (and there probably never will be). If you grow up in poverty, if teachers don’t come to work, if there is violence around you, it will remain hard to escape those surroundings. At the same time, kids who grow up in secure homes, have engaged teachers, and engage in stimulating activities will likely learn more things faster and better than ever before.
Rather than equalizing opportunities, next-generation AI may well widen gaps in reading and math skills. These gaps were already huge before COVID-19, and the pandemic widened them further.
The third problem is the most profound. Generative AI has greatly multiplied the power that’s already in the hands of the people who run the world’s dominant tech ecosystems. For many of the executives in charge, the primary focus is on developing versions of AI that can replace humans in as many tasks as possible. And this focus on automation is made possible through using massive amounts of people’s data substantially without permission.
In the past, new technologies were more socially beneficial when they took a “pro-human” path. In the production process, this means creating new tasks for workers. It also means not sidelining workers who don’t have a college degree. It is possible to make workers of all skill levels more productive. If we fail to do so, we can neither build true human capabilities nor genuinely share prosperity.
In education, we need greater personalization without sacrificing student-teacher interactions and peer learning. It would certainly not mean abandoning the reliability of information at the altar of large-language models.
Advances in digital technologies over the past 20 years have made this pro-human path more feasible, but tech leaders have consistently preferred automation and manipulation by social media. Now the misplaced emphasis is on automated teaching and replacing teachers and skilled human resources as credible sources of information.
It is not too late to pivot the development of AI in a more pro-human direction. We can start anywhere, but perhaps we should start in education, where hundreds of millions of children around the world are hungry for useful knowledge. AI can help us prioritize children and human learning-through-interaction, but we need to start by recognizing the folly of our current ways and wrest control of technology away from its self-appointed luminaries.