Meta (Facebook), Alphabet (Google), Microsoft, Twitter and a few other tech companies have come to dominate what we see and hear on the internet, shaping hundreds of millions of people’s perceptions of the world. In pursuit of advertising revenue, their algorithms are programmed to show us content that will hold our attention – including extremist videos, disinformation and material designed to stimulate envy, insecurity, and anger. With the rapid development of “large language models” such as ChatGPT and Bard, Big Tech’s hold on impressionable minds will only strengthen, with potentially scary consequences.
But other outcomes are possible. Companies could deploy the latest wave of artificial intelligence (AI) much more responsibly, and two current court cases serve as warnings to those pursuing socially destructive business models. But we also need public policy interventions to break up the largest tech companies and to tax digital advertising. These policy levers can help change Big Tech’s pernicious business model, thereby preventing the platforms from inflicting so much emotional harm on their users – especially vulnerable young people.
The legal cases include Gonzales v. Google, which is currently before the US Supreme Court. At issue is the tech industry’s insistence that Section 230 of the 1996 Communications Decency Act exempts platform companies from any liability for third-party content that they host. If platforms are acting more like news outlets than mere online repositories when they recommend videos, tweets or posts, they should be held to the same standard as established media, which, under existing defamation laws, are not allowed to publish what they know to be untrue.
Hence, in a US$1.6 billion lawsuit filed against Fox News, Dominion Voting Systems has uncovered ample evidence that Fox’s top on-air hosts and executives were well aware (and told each other) that former President Donald Trump’s claims of election fraud were entirely false. Dominion thus has a strong claim to damages if it can show that Fox knowingly spread falsehoods about Dominion’s voting machines in the 2020 election. Shouldn’t online platforms whose algorithms disseminated the same lies be held to the same standard?
Addressing such questions has become even more urgent now that programmes like ChatGPT are poised to reshape the internet. These sophisticated algorithmic recommenders could potentially be trained not to promote extreme content or deliberate lies, and not to encourage extreme emotions. If an algorithm is exploitative or manipulative toward children (or anyone else, for that matter), the responsibility for such harm should lie with the humans in charge. After all, AIs at this level are not operating autonomously of human decision-making. To claim otherwise is to grant their creators legal immunity.
Tech companies should no longer be able to excuse their own inattention or negligence by arguing that “there’s too much data” for them to monitor. That wealth of data is the source of their profits, and the sheer abundance of content on their platforms is what makes their AIs so potent. While they should enjoy a reasonable degree of protection against liability for what someone else posts on their site, this should apply only to passive content that the platforms do not in any way recommend to other users. Active content that is algorithmically pushed out to millions of people to generate revenue is a different matter. Indeed, it is just like traditional publishing, only much more powerful.
If a daily newspaper publishes a commentary by a terrorist, some readers will probably stop subscribing. But since most individuals do not want to walk away from their existing online social networks, we need government regulation to re-empower consumers.
First, the largest platform companies should be broken up to create more intense competition between recommendation algorithms and their trainers. But for this to work in the public’s interest, platforms also must be required to allow a user’s social network to be transferred to a different platform. The same “interoperability” rationale allows you to keep your cellphone number when you change carriers. Social media and digital content consumers should be able to vote with their feet when they don’t like what a platform is promoting.
Second, and even more importantly, we need to force an adjustment in the prevailing Big Tech business model, which is based on harvesting vast amounts of user data and monetizing it through digital advertising sales. This business model explains why disinformation, outrage and insecurity are so prevalent online. Emotional manipulation maximizes user engagement, enabling more intrusive data collection and higher profits.
A tax on digital advertising is one of the only practical ways to change this extraordinarily destructive business model. It would reduce platforms’ temptation to maximize user engagement through emotional manipulation; and, if coupled with limits on data collection, it would provide incentives to develop alternative approaches, such as subscription-based models.
Another advantage of a digital advertising tax is that it could be set even higher for content promoted to people under 21. Selling cigarettes or alcohol to minors is a serious criminal offense. While it is not feasible to forbid young people from seeing content that damages their mental health, a high rate of taxation on advertising revenues derived from promoting such material is entirely appropriate. The proceeds could be devoted to strengthening mental health programmes, not least those for teen suicide prevention. If there is any doubt about which content is hurting young people, we can just ask the AI recommendation algorithm.
Daron Acemoglu is a professor of economics at MIT; and Simon Johnson is a professor at MIT’s Sloan School of Management, a co-chair of the Covid-19 Policy Alliance and a former chief economist at the International Monetary Fund.
Copyright: Project Syndicate