David Gilbert on AI, Protest Disinformation and Checking Before Sharing

David Gilbert says his children will grow up with AI built into much of their world. In his view, they will need to understand both its strengths and weaknesses if society is to reduce the impact of misleading information.

Shira Lazar introduces the discussion against the backdrop of Los Angeles protests and immigration enforcement. She says social feeds contain a mixture of real footage, old material presented in a new context and AI-generated images. Her question is how journalists and ordinary users can tell what is real before sharing.

Lazar welcomes Gilbert, a senior writer at Wired covering disinformation and online extremism. She refers to his reporting on people asking chatbots such as Grok and ChatGPT to check protest content and receiving inaccurate answers.

Gilbert says definitions vary, but he broadly describes misinformation as false material shared without a deliberate intent to deceive. Someone might not know it is false, or might think a joke is obviously fictional. He uses disinformation for material shared with a specific purpose, such as misleading or dividing an audience.

As examples of deliberate campaigns, Gilbert points to state actors and political groups. He argues that disinformation appears across the political spectrum, while saying that in his assessment the scale is much greater in the US pro-Trump camp. The distinction he emphasizes is intent, not merely whether a claim is false.

Lazar asks what makes this breaking-news moment different from earlier waves of misleading content. Gilbert says the protests were already tense, but what stood out to him was AI's role both in creating material and in attempts to fact-check it. In his view, the tools were amplifying rather than reducing confusion.

Gilbert says widely available AI tools can reduce the time needed to create videos, images or websites, including for organized actors trying to inflame opposing sides. He also points to large volumes of low-quality AI-generated images and videos in ordinary feeds. Together, he argues, they make it harder for a user to identify reliable information.

Lazar describes a viral video apparently showing an older woman confronting uniformed officers and says she found it difficult to know whether it was real. Gilbert says he recognized the video she means as fake, but notes that he looks at such material professionally; most users do not spend their days doing that.

During breaking news, Gilbert says, people turn to social media for immediate information. He remembers the earlier Twitter as imperfect but more useful for surfacing verified news organizations and people on the ground. He contrasts that with X, where he says engagement-driven accounts and false narratives can dominate the feed.

Lazar asks what changed and how Grok fits into the picture. Gilbert says Twitter previously had human moderation teams and systems that helped surface verified accounts, although he had criticized the company at the time. He argues that changes after Elon Musk's takeover reduced moderation and rewarded engagement-oriented content.

Gilbert says Grok's integration into X lets users tag the chatbot and ask it to verify posts. He gives an example involving photos shared by California Governor Gavin Newsom of National Guard members lying on the floor. Gilbert says the photos had been published by the San Francisco Chronicle, but Grok incorrectly said they came from Afghanistan in 2021.

According to Gilbert, a user supplied the newspaper link and asked for a correction, but the chatbot repeated its claim. He says screenshots of such answers can travel beyond X to other platforms, giving a false narrative a wider audience. Lazar adds that confusion online can feed tension in the real world.

Lazar asks whether platforms, journalists, lawmakers or others should take responsibility. Gilbert places primary responsibility on platforms, but says efforts to compel them to act have largely failed in the United States. He argues that some companies have rolled back safeguards and that previously removed groups have returned to large platforms.

Gilbert says European regulators have made more effort and imposed fines, but he questions whether penalties change the behavior of very large companies. Lazar favors a broader response involving platforms, policy and public education about how to distinguish reliable information.

Gilbert agrees that education matters but is pessimistic about changing platform and political behavior. He says people can work on their own habits and teach children to assess material critically. He wants schools to teach the strengths and weaknesses of AI from an early age.

Lazar mentions a US tax and spending proposal that, at the time of recording, could make state AI regulation harder. She asks whether an international approach is possible. Gilbert says nationalism and political divisions make a global agreement on disinformation seem distant to him.

Lazar speaks about her own anxiety as a Canadian citizen living in California. Gilbert says repeated exposure to frightening content can affect people even when they try to assess it critically. They also discuss fears facing journalists and what Gilbert sees as a chilling effect on media coverage.

Asked how he checks viral content, Gilbert says his methods change as AI improves. A once-useful clue such as a person with six fingers is less dependable as generators get better. His first advice is to slow down, especially when a dramatic post is spreading very quickly.

He asks whether an event depicted in a post is plausible and uses reverse image search to look for earlier appearances of a picture. Similar tools exist for video. He also examines the account sharing the material to see whether it has a record that warrants trust.

Gilbert says a blue check on X no longer provides the same signal of identity or reliability he once associated with it. He warns that some accounts seek engagement and revenue by posting without checking, while acknowledging that this does not apply to every checked account. Lazar adds that people can pay for checks and buy established accounts.

Even as a professional, Gilbert says, he has occasionally been fooled and shared questionable material with colleagues. He describes this as an embarrassing but realistic consequence of how difficult verification has become.

Lazar asks whether people should avoid AI tools when verifying content. Gilbert says they can be useful as one part of gathering information, but should not be treated as final authorities. His goal is to locate the original source of an image or video and independently verify what a chatbot reports.

That work is laborious and takes time, Gilbert acknowledges. He recommends checking every claim from a chatbot because a confident answer can still be fabricated or wrong. Reverse search may help reveal an earlier version of an image and point toward its source.

The conversation turns to moderation. Gilbert says Reddit relies on people as well as automated systems: moderators in individual communities can remove false or hateful material and communicate with members. In his view, that human involvement is an important part of the system.

Gilbert is skeptical of community notes systems on X and their adoption elsewhere. He argues that coordinated supporters can influence whether a note appears, and that even a visible correction may leave the original post circulating. He says it remains to be seen whether Facebook's version will work better.

Lazar asks about resources such as Lead Stories. Gilbert also names PolitiFact, AP and Reuters as organizations doing fact-checking work, while arguing that their capacity is small compared with the amount of misleading content. He describes an example in which a photograph of bricks was presented as evidence of organized protest activity in Los Angeles, but a fact-checker located the pile in New Jersey.

Gilbert says some fact-checking organizations had received funding from Facebook and would face constraints after that support ended. He worries that fewer resources will be available to check an already overwhelming volume of claims.

Asked for a hopeful sign, Gilbert says some teenagers he speaks with are aware that the internet can be used against them and are more cautious about sharing every part of their lives. He points to older posts affecting job prospects as one reason people may reconsider what they publish.

His optimism is qualified: online communities can provide valuable connection, but he also worries about people, particularly young men, being drawn into extremist or nihilistic groups. He says those communities can become dangerous when they are a person's primary connection to others.

Lazar asks whether AI can help journalism. Gilbert says he does not currently use AI tools in his own reporting because he does not find them useful for his workflow, though other newsrooms do. He expects the tools to develop and is interested in seeing how they are used.

Lazar recalls hearing about a specialized tool used to help journalists identify bombed areas in a war zone. Gilbert agrees that a tool designed for a specific task can help people do that work better. The discussion then returns to the difficulty of establishing trust in both technology and eyewitness reporting.

Gilbert says people can be quick to reject material that challenges their existing views and accept material that confirms them. Lazar asks for one rule before sharing. His answer: ask why you want to share, examine the account and look for the original source. Being right, he says, matters more than being first.