Reality+ : Virtual Worlds and the Problems of Philosophy (9780393635812), page 27
So our answer to the Reality Question is mixed. In the short term, deepfakes may involve real digital entities, but these don’t have anything like the causal powers of standard virtual or physical objects. In the long term, virtual AI deepfakes may involve real digital entities with causal powers on a par with those of standard virtual or physical objects. However, in both cases we’ll probably say that a deepfake is not the real thing. Deepfake Obama is (probably) not the real Obama. A deepfake football is (probably) not a real football. This is enough to raise a serious question about knowledge.
How can we know whether an image is real?
Suppose we see a video that looks like Obama saying something. How do we know that this is the real Obama? Can we know that the video image is real? The problem generalizes to any photo or video. How do we know whether or not a photo of a waterfall is a photo of a real waterfall? How can we know that a video of a violent protest is authentic?
As the philosopher Regina Rini has observed, we are used to using images as reliable backstops for our knowledge. If we’re in doubt, seeing is believing, and a photo is proof! But in an age of deepfakes, images cannot be trusted so straightforwardly. Until now, it’s been possible to distinguish fake images from real images by a close-enough examination of giveaway signs. But as deepfake technology develops, the giveaways get harder and harder to notice. Soon they’ll be detectable only by advanced algorithms, and then giveaways may disappear entirely. At that point, there will be no way to distinguish deepfakes from real images by examining the qualities of the image.
The implausibility of an image may tip us off that it’s fake. If the image shows the Sydney Harbour Bridge turned upside down, it’s probably fake. If it shows Bernie Sanders endorsing the Republican Party, it’s probably fake. But if the image has enough antecedent plausibility—even just the plausibility of an everyday surprising news item—this method won’t work.
At the other extreme, if you encounter a mundane video of your cousin Sam telling you unimportant things, it’s most likely real—who’d bother to fake it? More generally, at present there are relatively few deepfakes circulating, so we have reasonable grounds for thinking most images are real. But as deepfakes become more common and easier to generate, they’ll become more and more of a crucial issue.
In the long run, the only way to know for sure whether an image is real or fake may be through authentication by a reliable source. If a trusted friend tells you that she took a certain photo, you have good reason to think it’s real. If a reliable media outlet puts out a video and says they recorded it, it’s probably genuine. By contrast, if we find a photo lying on the road or encounter a video on a partisan website, we’ll have less reason to think it’s real.
Relying on authentication is probably the best way to protect oneself from deepfakes in an environment where they’re ubiquitous. Of course, the method isn’t perfect: A trusted friend might play a trick on us, or his email account might get hacked. A reliable news outlet might sometimes be fooled. Or it might be taken over by bad actors without our knowing, or it might be convincingly spoofed. Still, a source can build up a track record that gives us reason to trust them. Some sources may be endorsed as reliable by other reliable sources, expanding our network of trust. Some hard issues remain: for example, what if all our sources are misleading? I’ll discuss these shortly.
In any case, photos and videos are only one source of evidence about the external world. If it turns out that we can’t trust any of them, this would undercut some of our knowledge but not all of it.
Once deepfake virtual realities are possible, the problem will multiply. Some cases concern VR environments themselves: How can I know I’m really playing multiplayer Beat Saber with my friends, as opposed to a deepfake emulation with bots acting my friends’ parts? Some cases concern ordinary perception: When I think I’m discussing an upcoming product with my corporate employer, how do I know I haven’t been kidnapped into a deepfake virtual reality by competitors eager to learn our secrets?
Again, one solution is authentication. To help avoid fakes, you should access only virtual reality software that you trust, and you should use only reliable augmented reality equipment. There will be some virtual elements, but within limits that are understood. The rules will say, “No fake friends and relatives without telling the users!” In some contexts, these rules will be difficult to enforce. In social VR, where people have control over the appearance of their avatars, a deepfaker might make an avatar that looks like your mother. But authentication should still be straightforward—for example, via a username known to belong only to your mother.
We still have to worry about extreme circumstances. What if the VR environment has been taken over or hacked? What if there are no trusted systems? What if someone has hacked into your brain, as happened to an unsuspecting airplane passenger in the movie Inception?
Or worse: What if you are yourself a deepfake? Perhaps your enemies produced a deepfake simulation of the original you, based on video records and the like, and now are using it to gain information about the original? In the Black Mirror episode “White Christmas,” police use a version of this method to induce confessions from suspects. Is there any way to know that this isn’t happening to you?
I don’t have a general answer to these questions. If the deepfake virtual reality is an imperfect simulation of the original, it should be possible to discover this through investigation. You can interact with your mother and see whether she knows things your mother should know. You can see whether the secret information in your secret notebooks is present. You can explore your world to make sure it’s all there. You can perform scientific experiments to see whether or not they give the expected results.
If the deepfake virtual reality is a perfect simulation of the original, then those sorts of investigation won’t work. We’ve seen that it’s impossible to know you’re not in a perfect simulation. This returns us to familiar issues. If you’ve lived your whole life in the simulation, it’s your reality, and your beliefs about your world are still true. But what if your VR world has recently been hijacked, or you’ve been kidnapped or uploaded into a perfect simulation? In this scenario, much of what you believe about the world around you may be false, and there’s no way to know for sure.
In a world in which VR hijacking, kidnapping, or uploading into perfect simulations is common, perhaps the best we can do is take precautions to avoid getting into those situations. Once deepfake virtual realities are perfected, we can expect that both computer security and brain security will be growth industries.
What about fake news?
Let’s return to Earth in the early 21st century, where we face the very real issue of fake news—misleading news stories produced and circulated without regard for truth.
Fake news has been around for as long as there has been news. One could argue that the Roman triumvir Octavian engaged in a fake news campaign against his rival Mark Antony in 31 BCE, portraying him as a traitor to the Roman Empire. In 1782, during the American Revolution, Benjamin Franklin produced a hoax issue of a newspaper with a false story about American scalps being sent to the king and queen of England.
The term “fake news” exploded around the time of the US election in 2016. The paradigm case of fake news is perhaps the Pizzagate story, which circulated before the election. The report accused Hillary Clinton and other officials of the Democratic Party of running a child sex ring out of a pizzeria in Washington, DC. The story appears to have started on Twitter and was circulated widely through social media and alternative news media. Investigation (to say nothing of plausibility) revealed that there was no truth to it.
The explosion of social media has served to amplify fake news. Social media allows readers to circulate misleading news stories to like-minded people on all parts of the political spectrum. The term “fake news” itself has now become controversial, in part because public figures often use it to delegitimize the news media by calling any unfavorable story fake news. But like deepfakes, fake news stories are a worrisome phenomenon.
Fake news isn’t the same thing as false or inaccurate news. If a journalist trying to report the truth makes mistakes, that’s false news but not fake news. Fake news requires intent to deceive, or at least a disregard for the truth. A clickbait news site will promote itself by making up news stories without caring whether they’re true or false.
We can raise a version of the Reality Question for fake news. If simulations and virtual worlds are real, then are the worlds conjured by fake news real, too? Fake news stories are akin to depictions of a fictional world. There’s often an underlying proposition (Hillary Clinton is a crook! Barack Obama was born in Kenya!) that crops up from story to story, suggesting a single underlying fictional world. But this world isn’t a virtual reality world: It isn’t immersive or interactive, nor is it computer-generated. So the virtual-realism argument can’t support it.
Perhaps we could imagine a computer simulation (Sim Pizzagate?), set up to spin off one fake news story after another. Inside the simulation would be a simulated entity, Sim Hillary, engaged in nefarious activities in a simulated pizza parlor. In that case, there would be a digital reality corresponding to the world of Pizzagate. But when we say “Hillary,” we’re talking about the original Hillary, so Sim Hillary’s nefarious deeds do nothing to make the Pizzagate allegations against Hillary true. In a far-out case where our world has been a Pizzagate simulation from the start, then we may be talking about Sim Hillary who is really a criminal, and the Pizzagate news will not be fake. But short of our inhabiting the simulations ourselves, there’s no danger that simulations of news stories will make fake news stories true.
The Knowledge Question for fake news is more pressing. How can we know whether any given news story is fake news? If we can’t know this, is the news media a source of knowledge at all?
We certainly treat the news media as a source of knowledge. In modern society, most of us rely on news stories for much of what we know about the wider world. We know about the political situation this way. We know about what’s going on in other countries and other cities this way. We know about crises and disasters this way. If we couldn’t trust the news media, then we’d know much less than we think we do.
Fortunately, there are ways to distinguish real news from fake news. As with deepfakes, glitches and inconsistencies serve as giveaways. We can sometimes rely on implausibility, or on mundaneness that no one would bother to fake. As before, the most important method is authentication by a reliable source. If a story is presented by a reputable source, it’s probably not fake. Stories by reputable news outlets often contain mistakes, but it is rare for them to be made up completely. Verification by other reputable sources, such as independent fact-checking outlets, can also bolster confidence in the accuracy of a news story. Because there are widely recognized reliable sources, the onset of fake news and deepfakes has not yet led to complete chaos, where no one knows what to believe.
How do we know that an apparently reliable source is indeed reliable? Mere consistency isn’t enough. Fake news can be consistent and presented as if it’s true. If a source is endorsed by other reliable sources, that’s helpful, but we can still imagine that the whole network of sources might be unreliable. Some political subcultures involve networks of media sources interconnected in webs of mutual endorsement but which are nevertheless unreliable.
In these subcultures, many people may be deceived about many things. But for a sufficiently critical subject, there will be limits. If you have full access to the internet, it will soon become clear that many sources contradict the information in your bubble. At this point, further inquiry will often reveal which aspects of one’s news are fake.
What if you’re a citizen in a regime where information is tightly controlled? Perhaps you’re in North Korea with only state-run media available. Perhaps you’re in an American subculture that restricts access to most media. In these cases, it’s harder to know the truth. One check is consistency with other sources of information: what you learn through your senses and what you hear from trusted others. People who have lived under these regimes often report signs that the news is misleading; perhaps the news says that everyone is well-fed, but people around you are starving. Deception is easier when stories concern matters far away. Perhaps state-run media can deceive people about what’s happening in other countries. But even here, most people know that news outlets are tightly controlled, which raises questions about their reliability. This may not tell you what you should believe, but it might lead you to suspend judgment about what you seem to learn from the news.
Even suspending judgment is hard work. Reasoning about conflicts in the news may take a great deal of critical thinking. But if a sufficiently observant and reflective thinker can avoid being taken in, then there’s at least a potential path away from deception by tightly controlled media. Of course, while suspending judgment about fake news is better than being deceived, it isn’t as good as knowing the truth. It’s arguable that the purpose of much fake news is simply to sow doubt. In her 1951 book The Origins of Totalitarianism, the German-born American philosopher Hannah Arendt said, “The aim of a totalitarian education has never been to instill convictions but to destroy the capacity to form any.” If everyone suspends judgment as a matter of course, then that strategy will have largely succeeded.
For me and for many readers, media are not so tightly controlled. A typical democratic citizen with an internet connection has access to countless news sources and many different viewpoints. There are many biases and blind spots in the media, but there are also, typically, dissenting sources that reveal those biases and blind spots. For example, the widespread American media biases that economist Edward S. Herman and linguist Noam Chomsky document in their 1988 book Manufacturing Consent may be real, but they’re also the subject of a high-profile book. Large chunks of the media may be in a position to deceive large chunks of the population, but the media as a whole are not tightly controlled enough for seamless deception of an entire population. (As Abraham Lincoln is supposed to have said, “You can’t fool all the people all of the time.”) When the media as a whole get something wrong, this is more often due to media ignorance than media deception.
Of course you can spin a scenario in which there are puppet masters behind the entire media complex, giving the appearance of openness when in fact all is intended to deceive. But across-the-board deception would require an enormous and complex conspiracy. Either the conspiracy would have to deceive an entire population, as in John Carpenter’s classic 1988 movie They Live, or the population would have to conspire to deceive an individual or a group, as in The Truman Show.
Such enormous conspiracy theories can’t be ruled out entirely, but Bertrand Russell’s appeal to simplicity (see chapter 4) seems appropriate: It’s reasonable to count this hugely complex conspiracy as more improbable than far simpler ordinary-world hypotheses. Perhaps the only reasonably simple version of the fake-news hypothesis is one in which we’re part of a giant computer simulation, but that brings us back to familiar ground.
Let’s sum up. For people whose news sources are tightly controlled, it may not be possible for them to know whether their news is fake or real, but with some effort they may often be in a position at least to know that something is wrong and so to maintain doubt. For people like me and many readers, who seem to have access to a huge array of news sources, it is often possible to determine whether news is fake or real by using the network of sources as a whole. We cannot completely exclude extreme scenarios in which almost every source of information is fake, but short of a computer simulation, these scenarios are so complex as to be unlikely.
Part 5
MIND
Chapter 14
How do mind and body interact in a virtual world?
IN FEBRUARY 1990, I SET OUT ON A ROAD TRIP TO SANTA FE. I WAS a 23-year-old graduate student in philosophy and cognitive science at Indiana University. My fellow students and I had heard about the new field of artificial life in which researchers try to create or at least simulate living systems inside a computer. Ten of us—philosophers, psychologists, and computer scientists—rented a van and drove to New Mexico across the great plains of Kansas, Oklahoma, and Texas. Our destination was the second-ever conference on artificial life, put on by the fabled Santa Fe Institute for the study of complexity.
It had been just three years since the first artificial life conference at Los Alamos National Laboratory, but there was already a plethora of approaches to creating life inside a computer. The approach that intrigued me most was designed by the pioneering computer scientist Alan Kay. The idea behind Kay’s “Vivarium” was to simulate a whole ecology in a computer. There was a simple physical environment consisting of a two-dimensional grid. Each square in the grid could be inhabited by an object, or it could be inhabited by an animal. Animals had simple bodies that could be oriented in various directions, move from square to square, and pick up objects.
The Vivarium world had a “physics.” Simple rules governed the two-dimensional grid and ordinary objects. It also had a “psychology.” The animals had separate rules governing their behavior. It was especially interesting to me that the physics and the psychology were distinct. This separation was and is standard practice in a virtual world. Inscribed in code, there is one set of rules governing ordinary objects in the world’s environment and another set governing how creatures behave in the world. Nonplayer characters in a video game are like organisms in the Vivarium. Their behavior is determined by rules special to them.
