Brian Alan (Ex-Dem Jew) 🇮🇱🇺🇲✡

W͢h͢a͢t͢ ͢i͢f͢ ͢A͢I͢ ͢c͢o͢u͢l͢d͢ ͢d͢i͢s͢a͢b͢u͢s͢e͢ ͢u͢s͢ ͢o͢f͢ ͢f͢a͢l͢s͢e͢ ͢b͢e͢l͢i͢e͢f͢s͢?͢ This would be a superhuman accomplishment, IMHO, sorely needed by humanity. Humans seem to do a poor job of dissuading others of irrational beliefs. When confronted with moon-landing-deniers, astrologists, telepathy tapes, or any unfounded beliefs, I have failed 100% of the time to change anyone’s beliefs using reason and evidence… and even cash prizes: cfiig.org/paranormal-cha… But now we have three peer-reviewed studies showing that AI conversations can do just that. WSJ summarizes: “Researchers found that when large language models are instructed to debunk conspiracy theories, they often successfully change people’s minds. In one study, participants discussed conspiracy theories they believed, and the researchers asked AI to persuade the participants that those theories weren’t true. Most participants not only changed their mind immediately after the conversation but also continued to hold those new views two months later. Another study found the models—which the researchers dubbed Debunkbots—helped participants rethink antisemitic conspiracy theories and see Jewish people more favorably. A third study suggests Debunkbots’ success at challenging conspiracy theories is driven by their ability to clearly explain relevant facts.” — wsj.com/tech/ai/ai-deb… The studies claim that the AI is successful because it presents evidence to refute a false belief. But I think that misses a much more important and subtle AI advantage — the evidence is not dismissed out of hand as fake news. We have had compelling evidence for the moon landings for many years now, but humans have not been able to persuade other humans to internalize the evidence, and the percent of the population denying the moon landings has grown every year since we first walked on the moon. I have witnessed an instinctual distrust and dismissal of the messenger when humans confront unfounded beliefs. It’s hard not to sound condescending when questioning a whack-a-do (see, there, I did it again), and defensive reactions easily ensue. The AI does not fall into the same trap of having a perceived agenda or condescending tone. They actually tested that, and Debunkbot is less effective when it takes a condescending vs. an affirming or neutral tone. We humans naturally model other minds in a dialog or debate, and can easily ascribe ulterior motives to the other, reinforcing the mental menagerie of a grand conspiracy. An AI hits different. We don’t yet see it having such ulterior motives, or motives of any kind. And we might be less defensive and more vulnerable with an AI, as we see in the efficacy of AI therapy. As a false belief begins to unravel, we might be more willing to explore that and allow a crack in our conviction without fear of embarrassment in front of others. The AI is not yet an other. And this gives me new hope for the future of humanity. Rather than backsliding into the retrograde irrationality of personal truth and unfounded beliefs, perhaps @xAI’s pursuit of truth might finally reorient humanity itself to its loftier mission. P.S. you can give it a spin at debunkbot.com I append a screen shot of what it said when I used the common arguments 1) if we landed, why have we not been back in over 50 years and 2) why don’t we see any stars in the black sky in the lunar surface photos? And below that is my prior post lamenting my inability to persuade deniers that they are wrong.