Illustration redesigned by Devin Thorpe
Last Thursday (Nov. 14, 2019) at Cambridge University, IBM’s “Project Debater” artificial intelligence program facilitated a debate over whether AI will do more harm than good to humanity in the future.
It began with a website, where anybody could submit an argument supporting one side–more harm than good–or the other–more good than harm. Over 1,100 responses were logged and processed by the algorithm. It found that 570 supported the “more harm than good” position, 511 “more good than harm,” with the remaining comments deemed irrelevant and discarded.
The algorithm then synthesized each side’s responses into five overarching arguments–for example, that AI will take away jobs from humans, or for the opposing side, that it will automate tedious tasks. At the beginning of the Cambridge debate society’s event, it presented each side’s five arguments, then left the debaters to debate.
Why it Matters
It may be called “Debater,” but IBM’s AI has applications far broader and more practical than facilitating philosophical debate. Really, what it did at Cambridge was simply condense human arguments, contextualize them, and re-package them into a coherent form.
Project Debate’s lead engineer said IBM is planning to allow certain customers of their cloud computing service to use Debater for business purposes.
Something to Consider
This is not Project Debater’s first, or even its most ambitious use to date. In February it debated Harash Natarajan, a worldwide debate champion. Only fifteen minutes prior to the contest, each side was given the topic–whether preschool costs should be subsidized by the government–and what position they were to take.–AI for, Natarajan against. The winner would be determined by what percentage of the event’s audience they were able to sway with their arguments. Before the contest began, 79 percent of audience members reported supporting subsidized preschooling, 13 percent not supporting.
By the end of the debate, 62 percent of audience members reported supporting subsidized preschooling, 30 percent not supporting. Natarajan converted 17 percent to his side and won.
Featured Header Image Source: SingularityHub