Content area
Full Text
Computer scientists should build devices to enhance and empower-not replace-humans.
In early 1997, before Alexa and Siri were conceived, I engaged in several public debates with MIT Media Lab professor Pattie Maes about the future of computing and artificial intelligence. The starting place for our disagreement was the question of control. Maes was advancing a vision of machine autonomy, of the proactive "software agent" that "can take initiative because it knows what your interests are." I offered a different vision: software designs that give users high levels of understanding and control over their AI-enabled devices to preserve human agency.
Although both views of AI design have their value, I believe that pursuing a vision of AI and technology design that starts with machine autonomy is dangerous for the future of human beings, our societies, and our environment. The consequences of an autonomy-first approach have begun to make themselves felt, in stock market flash crashes, deadly failures of autonomous missile systems in the second Iraq War, fatal accidents involving self-driving cars, and the two Boeing 737 MAX crashes. Autonomy-first elevates the dangers of hidden biases in algorithms for making decisions about who gets a mortgage, who gets a job, who gets paroled, and in the "surveillance capitalism" that turns surreptitiously collected private data into targeted advertising.
As Al-enabled devices proliferate in the economy and our daily lives, we are at a decisive moment. What will it take to put humanity on a path where humans steer AI to advance their own values and aspirations, rather than accepting control by autonomous decision algorithms that remain beyond our view and direct influence?
We should reject the idea that autonomous machines can exceed or replace any meaningful notion of human intelligence, creativity, and responsibility. The cliched images of a human hand shaking a robot hand or a humanoid robot pretending to think are archaic and misguided. People have remarkable abilities, and in contrast to future visions of humanoid robots and driverless cars that relieve us of our roles and responsibilities in daily tasks, I look forward to next-generation technologies that bring greater human control of ever-increasing automation.
My agenda is to help change the way we imagine, talk about, and design AI systems, starting with a vision of...