How to Find the Truth in Our "Fake News" World
With fake news and AI “deepfakes,” how do we get to what's real?
Posted Aug 10, 2018
Increasingly, it's getting hard to watch the news. Paradoxically, it's also hard to tear myself away from it. Each day I either log in or turn on the TV to figure out what is going on in the world. Not uncommonly, I find myself less certain about events than when I started. Politicians are bemoaning “fake news.” Seemingly absurd conspiracy theories are being taken seriously. I feel like I have a hunger for meaty high calorie facts, but all I get are potato chips and cigarettes. It satisfies the urge to put something in my mouth, but I am still left wanting and malnourished. Hence, I return to the trough for another unsatisfying bite.
I am not optimistic the situation will improve. With artificial intelligence, people are starting the see the emergence of “deepfakes," the combination of machine learning algorithms with facial mapping software that enables the creation of artificial content that can hijack a person’s face, body, and voice. Beyond the simple and creepy example of transposing a person’s face into a porn flick, these new approaches can create content of people saying and doing things that never occurred. These emerging capabilities leads to the unsettling feeling that the world of truth—the assumption that we can have unassailable facts to make decisions—is shifting and buckling beneath our feet.
Historically, this is not a new occurrence. In the early 20th century, physicists dealt with the same problem with the advent of Quantum Mechanics. When trying to understand the behavior of particles at atomic scales, the classical models of physics were failing. Measurements and calculations were not absolute. A photon is both a particle and a wave. An electron doesn’t orbit a nucleus but rather is a probability field. An uncertain, non-deterministic particle smeared about the central core of protons and neutrons. The emergence of this field fundamentally revealed the statistical nature of our knowledge of reality. Everything is a probability and not a certainty. Even weirder is the observer effect—the theory that simply observing a situation or phenomenon necessarily changes that phenomenon. Physicists have found that even passive observation of quantum phenomena can actually change the measured result (e.g. double slit experiment). While philosophically unsettling, these insights have been hugely successful in advancing numerous field of technology ranging from electronics, cryptography, to quantum computing.
So what does that have to do with fake news? Rather than making decisions on what is true and not true (classic media model), we have to make decisions on what is most probable (quantum media model). To do this, we can't only look at a single event and ask, “did this individual perform the act that is being documented either in the news or through some digital content?” We must also ask, "does this act fit within a high or low probability of their previous acts as documented by other media?" It forces a more holistic approach. Given that the media and content is nearly ever present and pervasive, we can now create these media probability fields. As an example, if a politician is accused of sexual harassment (which he claims is fake news), but there are numerous prior examples of this politician bragging about groping women, then it is probably true. If there is media accusing someone of inappropriate behavior, but all the digital content prior to that is inconsistent with that behavior than there is reason to believe it is probably false. Now that being said, since we are talking about probabilities, it is also possible that the sleazy politician is innocent and that the upstanding individual has had a fall from grace. They are just less likely.
Ironically, the emerging solution may be the very technology that complicates the problem in the first place—artificial intelligence. With the vast amounts of digital information, it is impossible to have any one person go through it all and make a calculated assessment. An AI, on the other hand, may be perfect for just such a task. Imagine an analytic that creates a “behavioral vector” for a given individual in which all their online actions are statistically mapped (like, John has followed the Red Sox for 20 years). When something comes up that is called into question (John sends out a text cheering on the Yankees), the program can determine how probable that behavior is, based on past content.
While this may not be completely satisfying that we will get “2+2=4” certainty of events in this world, perhaps it reflects a deeper truth that the physicists discovered over half a century ago—that the observer is inextricably linked to the events that they are watching. Our minds and our feelings cannot be separated from the reality as they occur and are documented. Our perceptions have always shaped the “truth” and it inevitably creates ambiguity. In recent times that ambiguity is sky rocketing, but at least modern techniques have the potential to quantify those shades of gray so we can make the best decisions possible. At least, it is highly probable.