A debate over whether to pause AI development has drawn a challenge from four prominent AI ethicists. They argue that focusing on hypothetical future dangers can distract from harms already linked to the use of AI systems, and call for regulation that makes companies more transparent and accountable.
A different focus for the pause debate
Earlier this week, thousands of people signed an open letter from the Future of Life institute proposing a six-month pause on developing AI models like GPT-4. Among the signatories were Steve Wozniak and Elon Musk. The letter cited risks including “loss of control of our civilization.”
Timnit Gebru, Emily M. Bender, Angelina McMillan-Major and Margaret Mitchell did not sign it. The four, who work together at the DAIR Institute, have instead published a response criticizing the letter for not engaging with problems caused by AI systems today.
They describe the focus on hypothetical risks as part of “a dangerous ideology called longtermism” that overlooks present harms. Their examples include worker exploitation, data theft, synthetic media that props up existing power structures, and the further concentration of those structures in fewer hands.
Present risks are already visible
The response contrasts fears of a robot apocalypse with concerns about how existing technology is used. Its authors point to a report about Clearview AI being used by police to “essentially frame an innocent man.” The article also mentions Ring cameras accessible through online warrant factories.
The point is not that future risks deserve no attention. The group agrees with some aims of the pause letter, including identifying synthetic media. But it says that discussion of future threats should not take attention away from harms for which remedies are available now.
This distinction changes the question from whether AI development should stop for a set period to what rules should govern systems already being built and deployed. The ethicists say companies should have obligations around how their systems are developed and how their outputs affect people.
Transparency and accountability as remedies
Their proposed response centers on regulation that enforces transparency. They say it should be clear when people encounter synthetic media. They also argue that organizations building AI systems should document and disclose their training data and model architectures.
They place responsibility for making tools safe to use on the companies that build and deploy generative systems. In their view, those companies should be accountable for the outputs their products produce. These proposals focus on making the development process more visible and connecting responsibility for outputs to the organizations behind the systems.
- Make clear when media is synthetic.
- Require developers to document and disclose training data and model architectures.
- Hold companies accountable for outputs from the generative systems they build and deploy.
The response also challenges the idea that ever larger AI experiments are inevitable. The authors characterize the race as a series of decisions shaped by the profit motive, and argue that regulation should shape corporate choices to protect people’s rights and interests.
Who gets to shape AI development?
At the heart of the disagreement is where public concern should land. The pause letter warns about powerful future systems. The ethicists argue that the immediate focus should be exploitative practices by companies claiming to build those systems, alongside the centralization of power and growing social inequities.
The article also recalls a statement from Uncharted Power founder Jessica Matthews at an AfroTech event in Seattle: “You should not be afraid of AI. You should be afraid of the people building it.” Her proposed solution, as described in the article, was to become one of the people building it.
The article says it is vanishingly unlikely that a major company would agree to pause its research in response to the open letter. Still, the letter’s engagement suggests that AI risks, both real and hypothetical, concern many parts of society. The ethicists’ counterpoint asks that concern to translate into action on current harms, with transparency and accountability at its center.