Cambridge Analytica built a weapon. They did so understanding what uses its buyers had for it, and it worked exactly as intended. To help clients manipulate voters, the company built psychological profiles from data that it surreptitiously harvested from the accounts of 50 million Facebook users. But what Cambridge Analytica did was hardly unique or unusual in recent years: A week rarely goes by when some part of the Internet, working as intended, doesn’t cause appreciable harm.
I didn’t come up in computer science; I began my career as a physicist. That transition gave me a specific perspective on this situation. That the field of computer science, unlike other sciences, has not yet faced serious negative consequences for the work its practitioners do.
Chemistry had its first reckoning with dynamite; horror at its consequences led its inventor, Alfred Nobel, to give his fortune to the prize that bears his name. Only a few years later, its second reckoning began when chemist Clara Immerwahr committed suicide the night before her husband and fellow chemist, Fritz Haber, went to stage the first poison gas attack on the Eastern Front. Physics had its reckoning when nuclear bombs destroyed Hiroshima and Nagasaki, and so many physicists became political activists — some for arms control, some for weapons development. Human biology had eugenics. Medicine had Tuskegee and thalidomide. Civil engineering, a series of building, bridge, and dam collapses. (My thanks to many Twitter readers for these examples.)
These events profoundly changed their respective fields, and the way people come up in them. Before these crises, each field was dominated by visions of how it could make the world a better place. New dyes, new materials, new sources of energy, new modes of transport — everyone could see the beauty. Afterward, everyone became painfully aware of how their work could be turned against their dreams.
Each field dealt with its reckoning in its own way. Physics and chemistry rarely teach dedicated courses on ethics, but the discussion is woven into every aspect of daily life, from the first days of one’s education. As a graduate student, one of the two professors I was closest to would share stories of the House Un-American Activities Committee and the antiwar movement; the other would talk obliquely about his classified work on nuclear weapons. Engineering, like medicine, developed codes of ethics and systems of licensure. Human biology, like psychology, developed strong institutional review boards and processes.
None of these processes, of course, prevent all ethical lapses, and they neither require nor create agreement about which choices are right. Many physicists, for example, began avoiding working on problems with military applications in the years after the McCarthy hearings and the Vietnam War. But many others do such research, and the issue is frequently and hotly debated.
Computer science is a field of engineering. Its purpose is to build systems to be used by others. But even though it has had its share of events which could have prompted a deeper reckoning — from the Therac-25 accidents, in which misprogrammed radiation therapy machines killed three people, up to IBM’s role in the Holocaust — and even though the things it builds are becoming as central to our lives as roads and bridges, computer science has not yet come to terms with the responsibility that comes with building things which so profoundly affect people’s lives.
Software engineers continue to treat safety and ethics as specialities, rather than the foundations of all design; young engineers believe they just need to learn to code, change the world, disrupt something. Business leaders focus on getting a product out fast, confident that they will not be held to account if that product fails catastrophically. Simultaneously imagining their products as changing the world and not being important enough to require safety precautions, they behave like kids in a shop full of loaded AK-47’s.
* * *
What would a higher standard of care look like? First of all, safety would be treated as a principal concern at all stages, even when “just trying to get something out the door,” and engineers’ education would equip them to do so. If safety came first, the Facebook Graph API used by Cambridge Analytica, which raised widespread alarm among engineers from the moment it first launched in 2010, would likely never have seen the light of day.
Tech companies focus intensely on preventing crashes. A rigorous effort to anticipate what could go wrong is already standard practice for specialists in system reliability, which deals with “what-ifs” around computer failures. A higher standard for safety would simply do the same for “what-ifs” around human consequences. This would not imply that all systems should be built to the same safety standards; nobody expects a tent to be built like a skyscraper. But the civil engineer’s approach would require a substantial shift of priorities.
Such a shift would sometimes be resisted for business reasons, but working codes of ethics give engineers (and others) more power to say “no.” If breaking ethics rules would mean the end of someone’s career, an employer couldn’t easily replace someone who refuses to cheat. If the systems for enforcement are well-built, a competitor couldn’t easily work around those standards. Uniform codes of ethics give engineers more of a voice in protecting the public.
Underpinning all of these need to be systems for deciding on what computer science ethics should be, and how they should be enforced. These will need to be built by a consensus among the stakeholders in the field, from industry, to academia, to capital, and most importantly, among the engineers and the public, who are ultimately most affected. It must be done with particular attention to diversity of representation. In computer science, more than any other field, system failures tend to affect people in different social contexts (race, gender, class, geography, disability) differently. Familiarity with the details of real life in these different contexts is required to prevent disaster.
There are many methods by which different fields enforce their ethics, from the institutional review boards that screen life-sciences experiments on humans and animals, to the mid-career certification of professional engineers who then oversee projects used by the unsuspecting public, to the across-the-board licensure of doctors and lawyers. Each of these approaches has advantages, and computer science would need to combine ideas and innovate on them to build something suited to its specific needs. What would not be acceptable is the consequence of inaction. The public would lose trust in technology, and computer scientists would face a host of practical, commercial, and regulatory consequences.
Computers have made having friends on the other side of the world as normal as having them next door, have put the sum of human knowledge in our pockets, and have made nearly every object we encounter more reliable and less expensive. Yet their failure, whether by accident or by unthinking design, can have catastrophic consequences for individuals and society alike.
What stands between these is attention to the core questions of engineering: to what uses might a system be put? How might it fail? And how will it behave when it does? Computer science must step up to the bar set by its sister fields, before its own bridge collapse — or worse, its own Hiroshima.
Yonatan Zunger, now at the startup Humu, is a former distinguished engineer in security and privacy at Google. Follow him on Twitter @yonatanzunger.