Despite these myriad risks, industry professionals seem to have turned a blind eye to the oncoming specter of A.I., likely because they are optimistic about its commercial potential. The leading internet firms are offering free A.I. courses for their brightest engineers; are developing plans to integrate A.I. across their leading brands and products; and are staffing up with brilliant philosophers, ethicists and technocrats to deflect nosy regulators and win over the merchants of information. In his interview with The New York Times last week, Mark Zuckerberg even went so far as to describe A.I. as the antidote that will cure the internet of such negative externalities as hate speech and election interference. What internet firms are not transparent about, though, is the degree to which they plan to integrate A.I. into their principal profit-generating engine: digital advertising.
Consider the scale and complexity of the advertising ecosystem. Global internet firms like Facebook, Twitter, Snapchat and Google collectively enjoy billions of active users who on average spend more than two hours on these platforms every day. Each user might be shown hundreds or thousands of digital display ads — or “sponsored” content — within that time. People in the industry leverage an intricate web of ad agencies, exchanges, networks, demand-side platforms and supply-side platforms to manage the delivery of those ads around the clock.
But if you could remove those people from the equation, you could quietly turn a $100 billion digital ad industry into a $1 trillion persuasion machine.
To be clear, algorithms have long been part and parcel of the industry. Organizations like the Russian government and Cambridge Analytica were taking advantage of them simply by virtue of using social media for political communications. But the sharp uptick in industry research and development in A.I. over the past year strongly suggests that this new technology will soon be brought to bear in digital advertising. This will increase the speed of ad mediation, inundating users with content finely tuned to their personal desires. It will abet the seamless and accurate development of “look alike” audiences, enabling advertisers to upload their customer lists and automatically send ads to like-minded people that they do not already know. And it will enable automated contingency-based marketing, allowing clients to programmatically trigger certain kinds of content to be shared in the moments after real world events transpire.
For students of disinformation — including the Russians who to date have not even had to leverage such sophisticated web technology to mislead American voters — this new information ecosystem presents a vast land of opportunity. One could imagine that the Internet Research Agency could set up automated, machine learning-informed content-targeting systems so that minutes after North Korea’s leader references a hypothetical I.C.B.M., the Russians send inflammatory A.I.-produced messages and imagery to classes of the American population that A.I. has predicted will be susceptible to disinformation. The scalability of such activity is what makes such tactics especially fearsome.
Deep-rooted societal tensions will likely be exacerbated by the irresponsible integration of A.I. into digital advertising services — not to mention, into the ranking and curation of “organic” content on social media news feeds and search results. We’re already seeing this. For example, just a few hours before Senator Marco Rubio took the stage to speak with students from Marjory Stoneman Douglas High School, ill-taught algorithms were responsible for the spread of YouTube videos that mocked and shamed the students.
Even in light of the Cambridge Analytica revelations, there is time yet to act. Internet firms should aggressively work to limit disinformation on their platforms by developing algorithms — perhaps driven by A.I., as suggested by Zuckerberg — that can detect disinformation and flag it for fast human review. Strong one-off actions against widespread disinformation tactics, such as Twitter’s recent move, can also help. They also must be more transparent about their algorithmic software and data practices with researchers, journalists and consumers. Further, the regulatory community must continue its aggressive review of the industry’s practices. The Federal Trade Commission’s announcement of its forthcoming investigation into Facebook’s privacy practices represents excellent progress.