While I agree with the thrust of this post, it ignores the harm done to individuals and groups that are victimized by deepfakes. Eric Eiswert is a prime example.
I have felt since about 2018 that the end result of all of this ultimately has to be a return to and revaluation of some form of traditional arbitors of truth and institutional mediators. Eventually we'll all mostly realize that anything that happens beyond our "lying eyes" needs a trusted source of validation and authenticity. If trust is in short supply then those who prove themselves to be trustworthy over time should therefore be invaluable, yes?
Or perhaps we'll learn to distrust everything first until we have evidence of the contrary. At least that's what we naturally did when cameras and other truth-imprinting devices didn't exist
We've long been at the point where even the most evidenced, unarguable facts are questioned or blatantly ignored by many. As you say, people believe what they want to believe. Can't wait for the constant arguments of "that's AI", "No it isn't" that we've got coming...
Exactly. The deceptive aspect of deepfakes is real but given that people willingly ignore truths when they come across them, I believe the true innovative power of deepfakes is indeed their unbounded capacity for expression
I think this is too tilted toward the abstractly philosophical and delight in the paradoxical. Plato would have called it Sophistry, in the classical sense. (See, e.g., the character of Gorgias in the Symposium. [PS: OOPS: he's not in the Symposium! Memory fail in placing him there. See later in this thread.])
It might have some connection to confirmation bias for globally famous people, but what about those for whom the Internet is a way of finding out more information about a relatively unknown person -- like most of us?
I'm married to someone in local politics, a challenger for a seat in the national parliament. She's relatively unknown, and is running against a long-term incumbent who is one of the most powerful politicians in the entire country (son of a PM, brother-in-law of a former PM, himself a former Finance Minister, and now the #3 guy in the ruling party). A nightmare I have is of a deepfake purporting to show her saying something she never said, or doing something she never did. (To a lesser extent, I worry about this for myself, too, since there is publicly available video online of me speaking, albeit from 10+ years ago.) In a country with an older, non-tech-sophisticated electorate like Japan, this isn't a matter of people *wanting to believe* my wife said something, but rather of filling in more information about her.
The same is true for almost any political candidate or expert, other than the most famous who already have attained positions of power. E.g., suppose it were a video of a candidate purportedly giving a speech to supporters, in which the candidate advocates that Japan build nuclear weapons -- a real taboo here. What if it were purported video showing her leaving a "love hotel" with another man (one local politician had to resign her seat in the national Parliament here, because of evidence like this that was genuine.)
In these cases, the video would be fatal because it showed a person *betraying* expectations, or doing something generally frowned upon. I think you really underestimate what a tragedy that could be for most people.
Thanks for calling it sophistry, lol. I'm talking about the primary power and value of a deepfake. What you're describing might be tragic for an individual, but it doesn't carry the same weight as a deepfake of a famous person. My point is that most people assume deepfakes exist primarily to conceal and deceive, but that's just not true. Some are designed to deceive, sure, but deception isn't *the* defining power of deepfakes.
OK, well, if you believe that the power to destroy individual lives isn't important, I see your point. But I think your belief is mistaken, and will become more so the easier it becomes for anyone with a laptop to generate deepfakes.
There's also a paradox in the notion of "weight" of a deepfake of a famous person. If I understand you correctly, then a deepfake of someone famous is most often going to be taken as a kind of satirical commentary about that person, without much real-world consequence. In that sense, what "weight" does it carry compared to revenge porn, defamation, etc. of less well-known people, whose lives and careers could be destroyed?
Are you implying that someone who isn't famous is less important than a famous person, even though the impact of a deepfake is less severe on the latter?
Maybe turn it around: the defining power of deepfakes is to make it more difficult to prove defamation, and to make it easier to bully relatively powerless people, to destroy their lives and bias others against them -- with the exception that the powerful and famous are somewhat insulated from those consequences, and deepfakes about them tend to be for different purposes. Why allow the point of view of the powerful to define what deepfakes are? And why ignore the real danger that everyday people face -- except that doing so is so delightfully contrary to what most people think?
PS: I goofed in my previous comment about the Sophist Gorgias: he's not in the Symposium! I inadvertently placed him there because long ago in college I'd written a paper about him in the same tutorial where the Symposium was our main focus as a group. (My memories of college not as fresh as they used to be: apologies🙇🏻♂️ .) The paper was actually about one of G's own works, The Encomium of Helen -- which really fits here.
Back in ancient Athens, Helen got the rap for causing the Trojan War and countless deaths. Gorgias went out of his way to contradict the conventional wisdom, using dazzling rhetoric and clever arguments to defend and praise her. (This speech was a set piece meant partly as entertainment, and partly as a marketing tool, to gather students who wanted to make flashy legal arguments and such.) Kind of like saying deepfakes aren't really what most people think they are, starting a post by saying we shouldn't be reading it, etc. 🙂
I think you understood me but are intentionally misinterpreting what I mean in the article. Reducing a stated hierarchy of importance to "so you mean that's not important at all" is simply engaging in bad faith. I never denied deepfakes can deceive. I explicitly said it's the wrong framing to say that's their main purpose.
I'm sorry that's happening to your wife and hope the best for you both. I guess when things get personal it's harder to make an unbiased assessment of the topic. I really understand that.
No bad faith at all, nor intentional distortion at all. Please think about the impact of saying that a famous person has a greater "weight" than a private individual -- when you also say that the impact on the famous person is less. It's unexpected and intellectually piquant to frame it your way; but ethically, there is a real problem to say that a famous person has more "weight" in this situation. BTW, the language in quotes in your comment is more extreme than what I actually said.
Re-reading your piece, you start out by acknowledging the potential for harm, but ultimately conclude that the deepfakes persuade us by expressing "what we already want to be true":
"That’s what they are—apt at expressing what could happen; what we'd want to happen but hasn’t. Instigators of viral counterfactual worlds instead of passive catalysts of unwanted realities we unwittingly buy."
This is simply not relevant to the vast majority of people whose lives could be ruined by deepfakes. As I mentioned, with regard to such people there isn't a pre-existing desire on the part of the public for that person to be one way or another; rather, the deepfake documents how they are, in the eyes of people who know the victim hardly at all.
Then you engage in a non-sequitur by saying that their greatest danger is in the hands of a charismatic leader, not a compulsive liar, even though you don't give an example of this. As to DJT, who is both of those things, he is a victim of deepfakes in your example. And the deepfake of Principal Eiswert, discussed near the same passage, wasn't deployed by a charismatic leader. (BTW, in saying that it's unclear what the context or evidence was for your statement about charismatic leaders, that doesn't necessarily mean I disagree with you about the potential of charismatic leaders to abuse deepfakes.)
Fortunately, the situation regarding my wife is still hypothetical; the point about bias is incorrect. It's just that her exposed situation is shared by countless other people, most of whom are totally private people. I am challenging your claim that it's the wrong framing, on the basis of the number of people potentially affected, the degree of harm they could experience, and the injustice of prioritizing people whose fame at least partly insulates them from such harm -- especially because your framing makes the potential harm of deepfakes seem much lower than it is, in my judgment.
It is a characteristic of our era for people to gravitate toward escalation of controversy, including to be a little trigger-happy about alleging bad faith. I'm old enough to be able to report that this wasn't always the case, even in political discourse with people of very divergent views. Certainly, it's not my MO to casually allege bad faith, nor to use it. Please relax: we're just having a conversation here.
I've always said the real power of deepfakes is not in creating something false, but in obscuring true things, and a deepfake doesn't even have to have been created or shared in order to do that for the reasons you mention above. It only has to be a possibility.
Great insight into the human condition! Human bias heaped on human bias arranged to maximize viewer dopamine. This also supports the idea that AGI must not replicate all of the human intelligence mechanisms that are the basis of our intelligence. Human biases cut both ways; they are the shorthand to our intelligence and the gateway to our manipulation. It will be difficult to only implement the positive valences of thought and create AGI.
While I agree with the thrust of this post, it ignores the harm done to individuals and groups that are victimized by deepfakes. Eric Eiswert is a prime example.
Yes, I'm choosing to underscore the counterintuitive aspect of deepfakes, not the part everyone knows. That's the point of the post
I have felt since about 2018 that the end result of all of this ultimately has to be a return to and revaluation of some form of traditional arbitors of truth and institutional mediators. Eventually we'll all mostly realize that anything that happens beyond our "lying eyes" needs a trusted source of validation and authenticity. If trust is in short supply then those who prove themselves to be trustworthy over time should therefore be invaluable, yes?
Or perhaps we'll learn to distrust everything first until we have evidence of the contrary. At least that's what we naturally did when cameras and other truth-imprinting devices didn't exist
Excellent unpacking of how/why deepfakes appeal, and why we want to believe them.
Thank you Grant!
We've long been at the point where even the most evidenced, unarguable facts are questioned or blatantly ignored by many. As you say, people believe what they want to believe. Can't wait for the constant arguments of "that's AI", "No it isn't" that we've got coming...
Exactly. The deceptive aspect of deepfakes is real but given that people willingly ignore truths when they come across them, I believe the true innovative power of deepfakes is indeed their unbounded capacity for expression
Brilliant unpacking of the AI genie! I wonder what legal systems are going to do with 'we have a video to prove that you...'
I think this is too tilted toward the abstractly philosophical and delight in the paradoxical. Plato would have called it Sophistry, in the classical sense. (See, e.g., the character of Gorgias in the Symposium. [PS: OOPS: he's not in the Symposium! Memory fail in placing him there. See later in this thread.])
It might have some connection to confirmation bias for globally famous people, but what about those for whom the Internet is a way of finding out more information about a relatively unknown person -- like most of us?
I'm married to someone in local politics, a challenger for a seat in the national parliament. She's relatively unknown, and is running against a long-term incumbent who is one of the most powerful politicians in the entire country (son of a PM, brother-in-law of a former PM, himself a former Finance Minister, and now the #3 guy in the ruling party). A nightmare I have is of a deepfake purporting to show her saying something she never said, or doing something she never did. (To a lesser extent, I worry about this for myself, too, since there is publicly available video online of me speaking, albeit from 10+ years ago.) In a country with an older, non-tech-sophisticated electorate like Japan, this isn't a matter of people *wanting to believe* my wife said something, but rather of filling in more information about her.
The same is true for almost any political candidate or expert, other than the most famous who already have attained positions of power. E.g., suppose it were a video of a candidate purportedly giving a speech to supporters, in which the candidate advocates that Japan build nuclear weapons -- a real taboo here. What if it were purported video showing her leaving a "love hotel" with another man (one local politician had to resign her seat in the national Parliament here, because of evidence like this that was genuine.)
In these cases, the video would be fatal because it showed a person *betraying* expectations, or doing something generally frowned upon. I think you really underestimate what a tragedy that could be for most people.
Thanks for calling it sophistry, lol. I'm talking about the primary power and value of a deepfake. What you're describing might be tragic for an individual, but it doesn't carry the same weight as a deepfake of a famous person. My point is that most people assume deepfakes exist primarily to conceal and deceive, but that's just not true. Some are designed to deceive, sure, but deception isn't *the* defining power of deepfakes.
OK, well, if you believe that the power to destroy individual lives isn't important, I see your point. But I think your belief is mistaken, and will become more so the easier it becomes for anyone with a laptop to generate deepfakes.
There's also a paradox in the notion of "weight" of a deepfake of a famous person. If I understand you correctly, then a deepfake of someone famous is most often going to be taken as a kind of satirical commentary about that person, without much real-world consequence. In that sense, what "weight" does it carry compared to revenge porn, defamation, etc. of less well-known people, whose lives and careers could be destroyed?
Are you implying that someone who isn't famous is less important than a famous person, even though the impact of a deepfake is less severe on the latter?
Maybe turn it around: the defining power of deepfakes is to make it more difficult to prove defamation, and to make it easier to bully relatively powerless people, to destroy their lives and bias others against them -- with the exception that the powerful and famous are somewhat insulated from those consequences, and deepfakes about them tend to be for different purposes. Why allow the point of view of the powerful to define what deepfakes are? And why ignore the real danger that everyday people face -- except that doing so is so delightfully contrary to what most people think?
PS: I goofed in my previous comment about the Sophist Gorgias: he's not in the Symposium! I inadvertently placed him there because long ago in college I'd written a paper about him in the same tutorial where the Symposium was our main focus as a group. (My memories of college not as fresh as they used to be: apologies🙇🏻♂️ .) The paper was actually about one of G's own works, The Encomium of Helen -- which really fits here.
Back in ancient Athens, Helen got the rap for causing the Trojan War and countless deaths. Gorgias went out of his way to contradict the conventional wisdom, using dazzling rhetoric and clever arguments to defend and praise her. (This speech was a set piece meant partly as entertainment, and partly as a marketing tool, to gather students who wanted to make flashy legal arguments and such.) Kind of like saying deepfakes aren't really what most people think they are, starting a post by saying we shouldn't be reading it, etc. 🙂
I think you understood me but are intentionally misinterpreting what I mean in the article. Reducing a stated hierarchy of importance to "so you mean that's not important at all" is simply engaging in bad faith. I never denied deepfakes can deceive. I explicitly said it's the wrong framing to say that's their main purpose.
I'm sorry that's happening to your wife and hope the best for you both. I guess when things get personal it's harder to make an unbiased assessment of the topic. I really understand that.
No bad faith at all, nor intentional distortion at all. Please think about the impact of saying that a famous person has a greater "weight" than a private individual -- when you also say that the impact on the famous person is less. It's unexpected and intellectually piquant to frame it your way; but ethically, there is a real problem to say that a famous person has more "weight" in this situation. BTW, the language in quotes in your comment is more extreme than what I actually said.
Re-reading your piece, you start out by acknowledging the potential for harm, but ultimately conclude that the deepfakes persuade us by expressing "what we already want to be true":
"That’s what they are—apt at expressing what could happen; what we'd want to happen but hasn’t. Instigators of viral counterfactual worlds instead of passive catalysts of unwanted realities we unwittingly buy."
This is simply not relevant to the vast majority of people whose lives could be ruined by deepfakes. As I mentioned, with regard to such people there isn't a pre-existing desire on the part of the public for that person to be one way or another; rather, the deepfake documents how they are, in the eyes of people who know the victim hardly at all.
Then you engage in a non-sequitur by saying that their greatest danger is in the hands of a charismatic leader, not a compulsive liar, even though you don't give an example of this. As to DJT, who is both of those things, he is a victim of deepfakes in your example. And the deepfake of Principal Eiswert, discussed near the same passage, wasn't deployed by a charismatic leader. (BTW, in saying that it's unclear what the context or evidence was for your statement about charismatic leaders, that doesn't necessarily mean I disagree with you about the potential of charismatic leaders to abuse deepfakes.)
Fortunately, the situation regarding my wife is still hypothetical; the point about bias is incorrect. It's just that her exposed situation is shared by countless other people, most of whom are totally private people. I am challenging your claim that it's the wrong framing, on the basis of the number of people potentially affected, the degree of harm they could experience, and the injustice of prioritizing people whose fame at least partly insulates them from such harm -- especially because your framing makes the potential harm of deepfakes seem much lower than it is, in my judgment.
It is a characteristic of our era for people to gravitate toward escalation of controversy, including to be a little trigger-happy about alleging bad faith. I'm old enough to be able to report that this wasn't always the case, even in political discourse with people of very divergent views. Certainly, it's not my MO to casually allege bad faith, nor to use it. Please relax: we're just having a conversation here.
Is it reasonable to imagine a future where we trust an AI agent to filter out all the fake content we encounter?
It could happen. But I believe eventually it'll be easier to create a fake than to detect it (I mention that's not the case just yet)
of course. you gotta trust something. it is one reasonable possibility among many
I've always said the real power of deepfakes is not in creating something false, but in obscuring true things, and a deepfake doesn't even have to have been created or shared in order to do that for the reasons you mention above. It only has to be a possibility.
Great insight into the human condition! Human bias heaped on human bias arranged to maximize viewer dopamine. This also supports the idea that AGI must not replicate all of the human intelligence mechanisms that are the basis of our intelligence. Human biases cut both ways; they are the shorthand to our intelligence and the gateway to our manipulation. It will be difficult to only implement the positive valences of thought and create AGI.