Erik Brynjolfsson: Why Using AI to Augment Human Capabilities Has More Value Than Automation

Stanford Professor Erik Brynjolfsson speaking on the value of artificial intelligence (AI) technologies being used to augment human capabilities:

If you are simply taking what is already being done and using machines to replace what the humans are doing, that puts an upper bound on how good you can get. If you simply automate the process of making clay pots so that clay pots can be done very cheaply then you have a lot of clay pots. But you don’t have new things. Whereas the bigger value comes from creating an entirely new thing that never existed before. You know, a supersonic jet or a nanoscale actuator or a new way of solving protein folding. These new things are where most of the benefits come from. We have iPhones now because somebody invented something new, they didn’t just make a cheaper telegraph.

Professor Erik Brynjolfsson discussed the perils of focusing AI development on systems that outmatch human capabilities as opposed to systems that complement humans with Brookings Fellow Anton Korinek during a fireside chat as part of the Brookings Institution series on "The Economics and Regulation of Artificial Intelligence and Emerging Technologies.”

In this talk, “The Turing Trap: A conversation with Erik Brynjolfsson on the promise and peril of human-like AI,” Professor Brynjolfsson explains why he believes that the Turing Test is a bad test of intelligence in machines. [In the Turing Test, a computer program replaces a man; and the idea was that if the questioner could not tell the difference between human and machine, the computer would be considered to be thinking.] Professor Brynjolfsson said, “if you have a machine that closely imitates humans, that can make human labor superfluous. More importantly, it can have some very negative economic effects.” He goes on to say:

A lot of people think that by definition, tech progress means just substitute for humans. The reality is that most tech progress is not substituting for humans. Most tech progress is amplifying humans, or complimenting them. A shovel or bulldozer allows the person to do more work, it doesn’t replace the worker.

Below is a summary of the Brookings talk, “The Turing Trap: A conversation with Erik Brynjolfsson on the promise and peril of human-like AI,” hosted by the Brookings Institution on November 2, 2022.

Summary

Increasingly, artificial-intelligence (AI) systems are hard to distinguish from human beings. If these systems enable humans to perform new tasks, we will be better off than if they merely substitute for human workers.

Main Points

• Technologists are approaching the point of creating machines that imitate humans so well that one cannot tell human from machine; the inability to distinguish humans from machines may be a test of human gullibility rather than a good test of machine intelligence.
• Technological progress that augments human abilities can increase wages; using machines as substitute for human beings will ultimately concentrate wealth in those who own the machines.
• We can encourage development of technologies that augment human abilities rather than substitute for human workers; currently, however, the entirely economic system is skewed to favor automation over augmentation.
  • Technologists are overly focused on making machines that replicate human abilities.
  • Businesses are overly focused on using machines for tasks that human workers already do.
  • Public policy taxes labor heavily and subsidizes investments in equipment, favoring automation over augmentation.
• One good example of a technology that augments human abilities is a call center system that advises a human operator how better to communicate with customers; the combination of human and machine is more effective than either working alone.
• Many performance metrics are geared to elimination of human workers; an alternative system should be developed to measure how well technologies augment human abilities.
• Some argue that the economy is too tied to a system that requires people to perform paid labor, and that technology should free people from the need to work; for now, this is inadvisable:
  • People find work fulfilling.
  • Those who depend on others’ generosity to support something like a universal basic income would have little bargaining power.
  • When machines do make humans dispensable, new institutions will have to be developed to protect the bargaining power of nonworkers.

Conclusion

Technologists, businesses, and policymakers have focused on the development of machines that serve as substitutes for human workers. Substitution will lower wages and could result in the concentration of wealth in the hands of those who own the machines. Technology that complements and augments human abilities, allowing humans to perform entirely new tasks, could raise wages and have more widely distributed benefits. Current tax policies tend to discourage labor and encourage capital investments in equipment, skewing the economy towards automation. To promote a more even distribution of wealth policymakers could consider a tax policy that is more balanced, or favors augmentation.

Watch the entire conversation: “The Turing Trap: A conversation with Erik Brynjolfsson on the promise and peril of human-like AI,” hosted by the Brookings Institution on November 2, 2022.

Related Reading

• The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence by Erik Brynjolfsson (Daedalus, Vol. 151, No. 2, pp. 272-287, 2022)
• TAP Article Summary for The Turing Trap
• Race Against the Machine by Erik Brynjolfsson and Andrew McAfee (Digital Frontier Press, 2011)
• The Second Machine Age: Work, Progress, and Prosperity in a Time of Brilliant Technologies by Erik Brynjolfsson and Andrew McAfee (W.W. Norton & Co., 2014)
• Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics by Erik Brynjolfsson, Daniel Rock, and Chad Syverson (in The Economics of Artificial Intelligence: An Agenda, University of Chicago Press, 2019)

Erik Brynjolfsson is the Jerry Yang and Akiko Yamazaki Professor and Senior Fellow at the Stanford Institute for Human-Centered AI (HAI), and Director of the Stanford Digital Economy Lab. He also is the Ralph Landau Senior Fellow at the Stanford Institute for Economic Policy Research (SIEPR), Professor by Courtesy at the Stanford Graduate School of Business and Stanford Department of Economics, and a Research Associate at the National Bureau of Economic Research (NBER). Professor Brynjolfsson’s research examines the effects of information technologies on business strategy, productivity and performance, digital commerce, and intangible assets. One of the most-cited authors on the economics of information, Professor Brynjolfsson was among the first researchers to measure productivity contributions of IT and the complementary role of organizational capital and other intangibles.