Unlike AI, Human Beings Weren’t Born Yesterday
Why the promise and peril of artificial intelligence may be vastly overblown.
Here’s an interesting headline, courtesy of Wired: “Facebook’s Head of AI Says the Field Will Soon ‘Hit the Wall.’” As we shall see, that wall, paradoxically enough, is the upward limit of human ability—and maybe that’s the way it should be.
The article features an interview with one Jerome Pesenti, Facebook’s vice president of artificial intelligence. According to Pesenti, AI researchers have made great progress on vision—that is, on “understanding scenes and images.” And so experts can now readily identify violence, child pornography, and so on.
We might add that the underlying process here is the familiar strength of computers; the machine can imbibe millions, billions, even trillions of images, and thus create an overall template of what’s considered okay and not okay. So if a given image falls outside of that template, well, there it is; it can be identified, censored, and perhaps turned over to authorities.
In addition, Pesenti said, AI-ers have made progress on language: “We can understand if people are trying to bully, if it’s hate speech, or if it’s just a joke. By no measure is it a solved problem, but there’s clear progress being made.”
Here we might pause to question that claim. Yes, it’s possible that AI researchers can do the same for speech as they can do for images: that is, put every accessible word or utterance into a database, crunch it around, and come up with an algorithm that decides what constitutes “hate,” and what’s merely a joke.
Yet we might ask ourselves: what are the chances that AI researchers, no matter how intelligent and well-intentioned, are going to succeed in nailing down the quintessence of human nature, which is found in language, thought, and communication?
After all, words and meanings have been evolving for thousands of years. Yes, back in the days when “silicon” just meant “sand,” we were already layering our lingo with irony, sarcasm, humor, inference—and oh yes, double meanings.
For instance, we might consider the word “bad,” which, in certain circles, at certain times, has meant “good.” Back in 1971, the American actor-director Melvin Van Peebles released a film, Sweet Sweetback’s Baadasssss Song, which introduced many Americans to the idea that “bad” might mean “good.” Now, over the last half-century, how many permutations have there been on this one word, “bad”?
In fact, we can begin to answer that question by consulting the Urban Dictionary, a sort of funky Wikipedia. The top-ranked suggested usage of “bad” in a sentence uses the word in reference to sex appeal. It reads, “Dam she hella bad.” For most of us, that’s an unfamiliar sentence construction.
Of course, any bad AI program will know all about the Urban Dictionary, and will have absorbed its terabytes of content in a single digital gulp. However, the AI program will always have trouble with subtexts and situations—what does this mean? What does that mean?
Furthermore, neither the Urban Dictionary nor AI knows what word, or meaning, or usage, will be added next.
In the meantime, we can all think of words and concepts that are deemed politically acceptable when coming from one mouth, and politically unacceptable if they come from another.
In fact, across society, these rules—mostly invisible and yet often career-altering—are already solidly in place. That is, we don’t need a computer to tell us what’s safe, and not safe, to say. But if so, it begs the question of what we need AI for, other than as a facilitator for mass surveillance.
Indeed, by this reckoning, AI is just a radically extended version of human sensibilities and judgments—and so the technical science is a lot less important than the political science.
Of course AI has vast uses in myriad other fields—everything from molecular modeling to consumer marketing—and yet in the end, these uses too are political. If we allow law and economics to elevate certain technologies, and certain technologists, to some empyrean plane from which they can look down on the rest of us, we then have to answer Juvenal’s timeless query, “Who will guard the guardians?” And today, if we’re not satisfied with the answer to that question, maybe this is the time for some guardian-adjustment.
Meanwhile, among the hottest topics in politics are the controversies over Facebook’s role in the 2016 presidential campaign: what did Putin do? And what did Cambridge Analytica know, and when did it know it? Many also are questioning Facebook’s role in the ongoing 2020 campaign.
As we all know, Facebook chief Mark Zuckerberg has been pounded, but good, on these very questions. A Republican cynic might say that Democrats are “working the ref,” such that Zuck is made more amenable to tilting the company in a Dem direction.
After all, given the labyrinthine complexity of all this tech stuff, it’s at least possible that Facebook, or some key subset of Facebookers, could make an algorithmic tweak that would help the Democrats carry, say, Michigan. To be sure, there are plenty of tech experts watching Facebook closely, on both sides of the aisle, but we might ask: if something were to happen, electorally, what are the chances that the experts would all agree on what it was, exactly, that happened?
Indeed, in the wake of any possible hanky-panky, the forensic debate over clicks, eyeballs, micro-targeting, and all the other arcana would rage in the way that every other debate rages these days—with the two sides looking at diverging, even dueling, fact patterns.
Of course, no matter what it does, Facebook will be part of any future feud. Yet maybe Pesenti, the AI guy, is realizing that it’s time to pull back as best he can. That is, the more the company bites off, AI-wise, the more it will have to chew.
Thus it was interesting to see Pesenti using the Wired interview to preach a new and perhaps unaccustomed humility for his company. Whereas Google, to this day, declares that its mission is “to organize the world’s information and make it universally accessible and useful”—no little hubris there!—the Facebook man says that AI has “hit the wall.” He adds:
Deep learning and current AI, if you are really honest, has a lot of limitations. We are very, very far from human intelligence, and there are some criticisms that are valid: It can propagate human biases, it’s not easy to explain, it doesn’t have common sense, it’s more on the level of pattern matching than robust semantic understanding.
We can pause over some of Pesenti’s words: “propagate human biases,” “doesn’t have common sense.” If even an AI enthusiast says this, what should the rest of us think—including those who, from the get-go, are profoundly mistrustful of this technology?
In fact, the skeptics are right to be skeptical, because in the long run—over the longest run—AI can’t tell us what’s true and what’s not. Why? Because it can’t foretell the future. As former Nixon defense secretary James Schlesinger said about arms control verification in the 1970s, “You cannot photograph an intention.”
Similarly, there’s no way to know the forward-looking veracity of any campaign pledge. In both the 1916 and 1964 presidential elections, for instance, the incumbent chief executive, seeking a second term in the White House, pledged not to embroil the U.S. in a foreign war. Yet in both cases, within months of the election, we were in the very war that we were told we wouldn’t be in.
On the other hand, plenty of campaign promises have been kept—and sometimes those kept promises worked out, and sometimes they didn’t. As always about the future, one never knows what will be what.
Thus we see the limits of AI, or any kind of technological nostrum. So long as we humans get the final say in actual matters, we will be free to tell the truth, to lie, to change our mind—and to engage in the googol of mental twists, tropes, and tricks that make us human.
Of course, some believe that humans won’t always get the final say, because computers will eventually take over—and perhaps even kill us all, Terminator-style.
Such an eschaton of humanity is possible, but as it says in our name, Homo sapiens, we aren’t dumb. Indeed, many people, from many vantage points, are keeping a close eye on AI. After all, unlike computers, we weren’t born yesterday.