Is an epistemic breakdown from social media and AI deepfakes a market failure of the Web?
Social media and AI deepfakes make it easy to spread falsehoods or distortions online. At the same time, these technologies make it more difficult to figure out what’s true. These (perverse) incentives can give rise to negative externalities in our information economy, such as misinformation or disinformation that wreak havoc on society—for example, eroding social trust in elections, harming public health, etc. If the Internet was supposed to help foster a marketplace of ideas, does this epistemic breakdown indicate a market failure of the Web?
The early days of the Internet held great promise. As the World Wide Web became available to the public, more information was more accessible to more people. An optimism arose that citizens and consumers would be better informed, and democratic society would flourish. A new marketplace of ideas was emerging online. At least, that was the hope for our new information economy.
Of course, an assumption behind this hope was that the information in the new marketplace of ideas would be of good quality. And there’s the rub. A lot of information online lacks quality, and how it spreads can incentivize quantity over quality. Consequently, people who take the necessary time to do careful research, make thoughtful reflections, and craft accurate content are often not rewarded for that work.
Instead, the opposite may hold true. Making falsehoods and distortions go viral can be easier than rendering truthful content or verified facts. To illustrate, consider a couple of salient examples known for spreading misinformation or disinformation online: social media and AI deepfakes.
Social media: outrage and falsehoods
Social media platforms don’t necessarily spread content based on its truth value. Rather, they spread content based on how likely it’ll hook your short-term attention. Then, by collecting data about what you ‘like’ and scroll through, these platforms sell access to your data to advertisers and outside parties, who, in turn, use that info to target you with more ads and content, especially clickbait that keep you ‘liking’ and scrolling endlessly.
Not surprisingly, the more outrageous the clickbait or content, the more likely it’ll hijack attention. Unfortunately, what’s outrageous need not resemble what’s true. As a result, social media platforms are designed (incidentally, if not intentionally) to spread outrage and falsehoods, not truthful or credible content. That’s one reason why misinformation and disinformation are rampant on social media.
In this way, the design of how information spreads over social media may exemplify what economists call perverse incentives.
Perverse incentives: incentives, or reward structures in the market, that can cause people to act in ways leading to unintended consequences, possibly producing the opposite of what the market was supposed to deliver. When these effects impose uncompensated costs on society, known as negative externalities, they may be signs of a market failure.
Thus, by making it more effortless to spread fake information—and more work to constantly verify what’s factual or not—social media platforms give rise to externalities like misinformation and disinformation. All that misleading info can erode social trust in elections, harm public health, or cause other societal damage. Likewise, we see a similar problem with the spread of AI deepfakes.
Deepfakes: misrepresentations and distortions of reality
Deepfakes refer to AI-generated images or sounds, such as photos, videos, and voices, that are made to look real. As such, they’re commonly created to misrepresent people, organizations, or societies.
Maybe that’s not much of a problem if most of us can clearly detect these distortions of reality. But how good are most individuals at spotting deepfakes? To date, studies show that test subjects can correctly identify deepfakes just over half of the time. That’s barely better than chance! Moreover, this ratio reflects what people can decipher in controlled experiments. In less-controlled environments, individuals likely fare worse.
For instance, when casually scrolling through social media, most users are probably more prone to being fooled by deepfakes, because there’s less opportunity to pause, think, and reflect. In this way, AI deepfakes, combined with social media, can spread distortions with ease.
When mendacity is easier than veracity
In effect, social media and AI deepfakes have the capability to make mendacity easier than veracity. And that capability to spread falsehoods or lies at ease can disproportionately enable bad actors. For example, generating fake but convincing online video makes it easy to scam people by impersonating others. On social media, spreading fake news was already a piece of cake. Now, deepfakes add fuel to that fire.
Conversely, authenticating identities and verifying information often requires more research and effort. Hence, we see a growing demand for cybersecurity, digital forensics, and other risk-management disciplines to verify information online. These professions are essential for the information economy. But they also involve time-consuming, and sometimes expensive, methods to validate data and figure out what’s true.
And so, we seem to have a problem of incentives in the way that information spreads online. Making falsehoods and distortions is taking less time and effort. Rendering truthful content and verified facts is taking more time and effort. These seemingly perverse incentives can give rise to negative externalities like misinformation or disinformation.
So, if the Internet was supposed to help foster a marketplace of ideas, does this epistemic breakdown indicate a possible market failure of the Web? If so, is it feasible to fix it?
Fixing a market failure of the Web
It’s possible that robust data-protection policies could help mitigate the damage to our information economy from the double whammy of social media and AI deepfakes. For instance, Scientific American‘s board of editors has suggested extending copyright laws to protect “one’s right to their likeness and voice.”
The basic idea is that your own image of yourself should belong to you, and nobody should be able to use it without your permission. A rule like this could empower people with legal means that force companies to prevent and remove fake, damaging content—like AI deepfakes that spread over social media.
Perhaps this threat of lawsuits over copyright infringement could help correct what appears to be a market failure of the Web. In fact, the fear of lawsuits may have been one reason why OpenAI suddenly discontinued its Sora AI video app, an AI application that generated realistic-looking, but fake, video.
Granted, extending copyright laws won’t be a panacea, and other policies will be necessary too. Whatever the solutions, they’ll need to change the incentives in the information economy. In other words, they’ll need to raise the transaction costs of spreading falsehoods and lower the costs of rendering truth.
Related posts
Addictive social media platforms bring out the worst in everything
Why ethical questions about technology design are unavoidable
Social media on trial: liability based on design