Users Like It or Not, A.I. Is Part of Health Care Now There's a key thing to keep in mind if you ask a chatbot for medical advice. Asking a general-use chatbot for health help used to seem like a shot in the dark--just two years ago, a study found that ChatGPT could diagnose only 2 in 10 pediatric cases correctly. Among Google Gemini's early recommendations were eating one small rock a day and using glue to help cheese stick to pizza . Last year, a nutritionist ended up hospitalized after taking ChatGPT's advice to replace salt in his diet with sodium bromide. Now A.I. companies have begun releasing health-specific chatbots for both consumers and health care professionals.

Unfortunately, AI systems are inherently unreliable. Within days of AI Overviews' release in the US, users were sharing examples of responses that were strange at best. It suggested that users add glue to pizza or eat at least one small rock a day, and that former US president Andrew Johnson earned university degrees between 1947 and 2012, despite dying in 1875. On Thursday, Liz Reid, head of Google Search, announced that the company has been making technical improvements to the system to make it less likely to generate incorrect answers, including better detection mechanisms for nonsensical queries. It is also limiting the inclusion of satirical, humorous, and user-generated content in responses, since such material could result in misleading advice.

We don't yet know exactly what happened with the Boeing 737 Max crashes. But it's clear from other non-fatal 737 Max incidents that there are concerns about how its automated systems function. And that's something to keep in mind when we think about self-driving cars. The problem lies in how AI-driven machines learn -- and share. Tesla, for instance, reported last year that its cars had driven over 1.2 billion miles in Autopilot and over 2.4 billion miles more in "shadow mode," when the Autopilot systems are learning, but not in control.

In April, Facebook's CEO Mark Zuckerberg testified in front of Congress that the company would eliminate hate speech on the platform within 5 to 10 years using AI technology the company is currently developing. But Zuckerberg says that he thinks both Holocaust Deniers and Sandy Hook truthers, two fringe conspiracy groups that even GOP leaders have condemned, would not be subject to the crackdown. In a 90-minute interview with Recode's Kara Swisher, Zuckerberg discussed issues of censorship on the platform, including the company's alleged conservative bias and its interference in Myanmar. When asked why he couldn't take down content that calls Sandy Hook a fraud, Zuckerberg conceded that he believes the claim is false, but that only direct harassment against victims of the tragedy would be removed. In his justification, he likened the group to Holocaust deniers.

In just a few years, deep learning-powered AI and other forms of machine learning have exploded from niche tools into the underlying basis of nearly every major modern online platform. Yet, even as these algorithms increasingly wield near-absolute power over what we see and say online, we have precious little visibility into how they function. If AI tools can robustly prevent hate speech and terrorist recruiting, flag false news and delete financial scams, they will be a welcome addition to our online lives. On the other hand, without any visibility into how often they get things wrong, we have little reason to trust their successes. AI and machine learning have become ubiquitous on the modern web, powering everything from security scanning to content moderation.