Google Earth fakery and Hanlon's Razor
On assumptions of human decency, Nano Banana 2, and the problem with failing to think through basic questions of what harm might come of things
Most of the time, it’s best to extend grace and assume the best in others. Even though everyone among us is imperfect, it can be safely assumed that most people are generally trying their best to be decent human beings. Failures run amok, of course: Hanlon’s Razor (“Never attribute to malice that which is adequately explained by stupidity”) exists to account for precisely that high frequency of failure.
■ However, it is important for people in certain positions to realize that the exception is also true: Even though most people are trying their best, some people are crooked. Some will do wrong, knowingly and willfully. And to protect the vast majority who are trying to be good, we need those who create, manage, and regulate powerful tools to account for the really villainous who walk among us.
■ Google excitedly announced an integration of its “Nano Banana 2” AI image generator with Google Earth’s mapping software. Mere hours later, it retracted the announcement because they had “seen people sharing screenshots of generated imagery that appear to violate our policies”.
■ It should generally be safe to assume that most people using an integrated tool like this would do so for above-board reasons of curiosity, imagination, and exploration. But it’s also reasonable to expect Google to take prudential precautions ahead of time in anticipation that someone would try to do something harmful with the same tool. External researchers didn’t have to try hard to get the tool to generate content that it absolutely should not have produced -- disinformation of the first order.
■ You don’t keep good people safe by assuming that everyone is going to be good all the time. You protect good people (who are, after all, the vast majority) by asking, “How would a malicious actor use this and how might we deter them?” Don’t assume that most people will use a thing badly, but do assume that a small number quite certainly will.