The trouble with age-gating the internet

How the next wave of technology is upending the global economy and its power structures

Jul 10, 2024 View in browser   By Christine Mui Meta CEO Mark Zuckerberg prepares to testify on child safety before a Senate Judiciary Committee hearing in Washington on Jan. 31, 2024. | Manuel Balce Ceneta/AP With help from Ruth Reader and Mohar Chatterjee In the national arguments about how to make the internet safer for kids, “age gating” — a way to keep out underage users — is growing in popularity among would-be regulators. It sounds simple: having visitors confirm they’re old enough to access a website. Already, 19 states have passed laws requiring mainly two kinds of sites — social media and porn — to verify their users’ ages. And on Tuesday, federal regulators for the first time ordered an app to keep children away, with the Federal Trade Commission and Los Angeles District Attorney's Office accusing the anonymous messaging app NGL of using shady tactics to market to kids. But how do you build a wall to keep kids out? As a policy solution, it steers straight into a host of tech problems that companies (never mind regulators) haven’t really solved. In the NGL case, the FTC ordered the company to impose an age-verification system that locks out any minors. Specifically, the system has to be a “neutral” age gate, meaning it needs to pose the question without hinting at why it was asking: for example, “How old are you?” rather than, “Are you older than 18?” Regulators also discouraged NGL from mentioning certain features would be unavailable to anyone under a certain age in an attempt to stop kids from hitting the back button and lying about their ages. Is asking in a different way really the best enforcement method? Haven’t more technologically sophisticated methods become available? Maybe not — and the problem illuminates bigger issues around privacy and free speech rights on social media. Meta in 2022 began testing facial age estimation, which uses AI to guess a person’s age by analyzing the geometry and features of their face. A recent evaluation by the National Institute of Standards and Technology showed the technology’s accuracy varies, and it struggles with poor image quality, although the software has shown some improvement over the past decade. Other proposed age verification methods don’t use biometric data, but may run into their own legal or ethical issues. One online store selling tobacco products asks customers to provide their full legal name, permanent address, date of birth and the last four digits of their social security number to place an order. The company then verifies this information automatically with third-party software, which checks it against public records to confirm the purchaser is at least 21 years old. If the records don't match or customers prefer not to provide their social security number, they must upload a valid government-issued ID for manual verification. Proponents of facial recognition and third-party verification claim that users prefer them over providing formal identification — but privacy advocates have warned those methods are too intrusive. Critics also say that extreme age verification tools could exclude adults who are legally entitled to visit the sites they want. In its next term, the Supreme Court has agreed to consider whether a Texas law requiring porn websites to verify users’ ages to limit access for minors violates free speech protections under the First Amendment. The question of how to age-gate has flummoxed regulators overseas as well. In 2022, the French Commission on Information Technology and Liberties scrutinized methods for online age verification, including payment card checks, facial recognition, offline validation, ID submissions, government tools and inferential checks. Their report concluded none met all of three standards: “sufficiently reliable verification, complete coverage of the population, and respect for the protection of individuals’ data and privacy and their security.” Australia’s eSafety Commission concluded that current age verification technologies aren’t mature enough to ensure both online safety and user privacy. Instead, it recommended improving child safety through industry codes, transparent platforms, and better parental support. NGL co-founder Joao Figueiredo disputed claims that his site improperly targeted a young user base as “factually incorrect.” Still, he told DFD in a statement that he is up for making changes the FTC suggested (the 18-plus age gate is already in effect) and thinks the rest of the industry should follow suit. “We think the agreement is in our best interest,” Figueiredo said. “The agreed upon age-gating and other procedures will now provide direction for others in our space, and hopefully improve policies generally."   Understand 2024’s big impacts with Pro’s extensive Campaign Races Dashboard, exclusive insights, and key coverage of federal- and state-level debates. Focus on policy. Learn more.     THE FTC WADES INTO OPEN SOURCE AI The Federal Trade Commission just waded into an obscure-seeming arena of AI policy that could nevertheless define the future of research and competition in the field. In a blog post on Wednesday, the agency’s Office of Technology wrote that open-weights AI models can “drive innovation, reduce costs, increase consumer choice, and generally benefit the public.” For AI companies worried about whether the agency might crack down on anticompetitive practices, that was a signal that one approach to AI might be more acceptable than a fully closed, proprietary model. For those getting really technical, it’s notable the agency emphasized “open-weights” rather than “open-source,” a term more commonly thrown around. (“Weights” here refers to the data underpinning a trained AI model, which some companies keep secret even if they disclose other information about the model.) The agency drew a sharp distinction between models whose weights are publicly available and those — like Meta’s LLaMa 2 model — that claim to be open-source but have strict licensing restrictions. The FTC also warned against companies adopting a “open first, closed later” strategy to lock out rivals, referring to the idea that companies might provide wide access to their AI models initially to gain market share, then pivot to a more restrictive approach. Overall it was a sharp warning from an agency that, together with the Justice Department, has brought landmark challenges to Amazon, Meta, Google and Apple. And although the FTC praised open-weights models for their potential to “improve privacy, security, and auditability,” the agency warned those models were also at greater risk of being used for malicious ends, including nonconsensual intimate deepfakes and child sexual abuse material. — Mohar Chatterjee The health benefits of missing data Artificial intelligence needs data — vast amounts of data — to make accurate predictions. And health care, one of the most important fields now adopting AI, has a complicated relationship with data. Health systems collect a lot of data on patients, but they collect it in different ways, keep it in different places — and sometimes it’s just missing.   A new AI product just approved by the FDA tries to make a virtue of that missing data, exploiting the gaps as a form of information itself. It’s an early warning tool called eCART that uses routine patient data, like basic vital signs and routine lab work, to predict when a patient might take a downturn in the next day or so. The FDA just gave its owner AgileMD the go-ahead to market its product to hospitals. In addition to tracking patient data, the platform is also trained to see missing information as a way to understand how the health care provider is thinking. For example, if a patient hasn’t received a test for a certain digestive enzyme called lipase, it’s probably because the doctor doesn’t think that the patient has pancreatitis, said Dana Edelson, a physician and founder of AgileMD. Another example is respiratory rate, which Edelson said can reflect how much a nurse worries about a patient. If someone is breathing normally, doctors and nurses will approximate their respiratory rate as either 16, 18, or 20 beats per minute in their health record. If the number in the record is far from that, or a prime number, it means the clinician actually watched someone breathe for a full minute, counting every breath. “I tell the residents when I teach them in the hospital, if you ever see a prime number written for respiratory rate in the chart, you should be very worried, because some nurse was worried enough to count for a full 60 seconds,” she said. The algorithm has been shown to reduce deaths in hospitals, largely by helping to move patients to the intensive care unit sooner. It also has the backing of the Biomedical Advanced Research and Development Authority, the military’s biomedical innovation arm, which worked with eCART to develop and deploy its technology in 2022 to identify Covid-19 patients at risk of death. The algorithm is already on the market and being used in five health systems.— Ruth Reader   SUBSCRIBE TO GLOBAL PLAYBOOK: Don’t miss out on POLITICO’s Global Playbook, our newsletter taking you inside pivotal discussions at the most influential gatherings in the world. Suzanne Lynch delivers the world's elite and influential moments directly to you. Stay in the global loop. SUBSCRIBE NOW.     Tweet of the Day The Future in 5 Links OpenAI does away with board observers after Microsoft’s departure. Why are governments placing identical limits on quantum computer exports? Trump’s 2024 playbook: Betting on crypto, AI, and Elon Musk.  Smart devices keep shrinking. Tesla’s grip on the U.S. EV market has slipped to below 50%.  Stay in touch with the whole team: Derek Robertson (drobertson@politico.com); Mohar Chatterjee (mchatterjee@politico.com); Steve Heuser (sheuser@politico.com); Nate Robson (nrobson@politico.com); Daniella Cheslow (dcheslow@politico.com); and Christine Mui (cmui@politico.com). If you’ve had this newsletter forwarded to you, you can sign up and read our mission statement at the links provided.   Follow us on Twitter Ben Schreckinger @SchreckReports Derek Robertson @afternoondelete Steve Heuser @sfheuser   Follow us

To change your alert settings, please log in at https://login.politico.com/?redirect=https%3A%2F%2Fwww.politico.com/settings This email was sent to rouf@idiot.cloudns.cc by: POLITICO, LLC 1000 Wilson Blvd. Arlington, VA, 22209, USA Unsubscribe | Privacy Policy | Terms of Service