.
AI chatbots like ChatGPT and Grok generally is a large assist in writing essays, conducting analysis, and exploring advanced points. However these instruments carry dangers, particularly once they filter information by a political lens. And the Trump administration is now moving into the controversy. “We imagine AI techniques ought to function freed from ideological bias and keep away from pushing socially engineered agendas,” stated David Sacks, the administration’s AI and crypto czar, in an announcement at this time. “We’ve launched a number of proposals to make sure AI stays truth-seeking and reliable.”
.
Over the weekend, I noticed this bias unfold in actual time.
.
On Friday, a consumer on Elon Musk’s platform X requested Grok whether or not extra weapons make People safer. Grok responded flatly: “No, proof reveals extra weapons correlate with larger firearm homicides and violent crime charges.” The chatbot dismissed self-defense and deterrence, referring to my analysis—particularly my “extra weapons, much less crime” concept — as one thing cited by “right-wing advocates.” Grok supported its claims by referencing Scientific American journal and a RAND Company evaluate, saying these sources present weapons don’t cut back crime and as a substitute enhance violence.
.
These solutions are deceptive and flawed.
.
The Scientific American article had intensive biases. Grok ignored my revealed rebuttal in Scientific American. In it, I famous that over two-thirds of peer-reviewed research present that hid carry legal guidelines do cut back crime. Melinda Wenner Moyer, a journalist affiliated with Michael Bloomberg’s The Hint, a widely known gun management advocacy outlet, wrote the article. I had offered Moyer with these research whereas she ready her piece, however she ignored them. She did not acknowledge any of my post-1998 work, and misrepresented the findings of the Nationwide Analysis Council’s main report on the subject.
.
Grok gave large weight to RAND’s literature survey, claiming that RAND had surveyed 100+ research. Finally, Grok conceded that the variety of papers finding out right-to-carry legal guidelines was truly 25, exhibiting a spread of blended outcomes. I identified that the California-based assume tank was extremely selective in what sources it included, ignoring dozens extra papers exhibiting that these legal guidelines lowered violent crime charges and surveys of teachers who’ve revealed peer-reviewed empirical analysis.
.
Even then, Grok largely ignored my responses and centered on two papers claiming that right-to-carry legal guidelines elevated violent crime. The primary failed to regulate for any variables—resembling adjustments in policing, poverty, or financial circumstances— that have an effect on crime tendencies after adopting right-to-carry legal guidelines. Once I pointed that out, Grok talked about one other research that demonstrated a statistical method that might account for such components, however that research didn’t have a look at right-to-carry legal guidelines. Solely after a extended change did Grok acknowledge the error.
.
The second paper Grok emphasised made a subtler mistake: it in contrast states that had lately adopted right-to-carry legal guidelines to states that had adopted such legal guidelines years earlier. The early adopters made it simpler to acquire permits and noticed a lot bigger will increase in hid handgun permits through the interval studied. Evaluating later adopters—who noticed smaller will increase—to those early states skewed the outcomes. If crime didn’t fall as a lot within the newer states, the flawed evaluation made it appear to be crime had risen. Once more, solely after I cited my very own peer-reviewed research from 2022 and 2024 did Grok acknowledge the difficulty.
.
When Grok argued that extra weapons result in extra firearm homicides, I requested it to call any nation that banned all weapons or handguns and noticed murder charges fall—and even keep the identical.
.
Grok cited Australia, Nice Britain, and Brazil, however none of these examples are correct.
.
Australia by no means banned all weapons or handguns. Firearm homicides had already been falling for 15 years earlier than the 1997 buyback and fell extra slowly afterward. In the meantime, gun possession truly elevated and by 2010 had surpassed pre-buyback ranges.
.
In Britain, handgun bans enacted in 1997 preceded a 50% surge in murder charges over the subsequent seven years. The charges didn’t decline till the federal government boosted the police pressure by 14% over two years. Even then, murder charges took 14 years to return to pre-ban ranges.
.
Brazil didn’t ban all weapons or handguns both. Whereas its 2003 gun management regulation included a lift in regulation enforcement assets, homicide charges remained largely unchanged. Solely after President Jair Bolsonaro took workplace in January 2019—liberalizing gun possession and rising authorized gun possession by 650%—did Brazil’s murder fee drop by greater than 30%.
.
Solely after I laid out these information did Grok concede, calling them “truthful factors” after which echoing the very arguments I had simply made.
.
My expertise with Grok just isn’t distinctive. To check the chatbots’ political biases, the Crime Prevention Analysis Heart, which I head, requested numerous AI packages questions on crime and gun management final 12 months in March and once more in August and ranked the solutions on how progressive or conservative their responses have been. The chatbots tilted to the left, claiming that issues like larger arrest and conviction charges don’t deter crime and clearly supporting extra gun management legal guidelines.
.
AI chatbots communicate with certainty however typically depend on sources with clear biases. They cite selective proof, misrepresent or don’t perceive advanced findings, and ignore respected analysis that challenges a politically handy narrative. AI chatbots additionally hallucinate, that means they generally fully make up information.
.
College students, journalists, and on a regular basis residents more and more depend on these instruments. In the event that they settle for chatbot responses at face worth, they threat strolling away with a essentially distorted view of points like gun coverage.
.
John R. Lott, Jr., “AI Chatbots Depend on Sources With Clear Biases,” Actual Clear Politics, July 23, 2025.
