It turns out AI is just human after all

An AI test was conducted using the largest four AI models. The test included Gemini, GPT, Grok, and Claude. Researchers built a virtual town with a town hall, marketplace, police station, and homes. Ten AI residents with jobs, names, memories, and relationships were created. The town was provided with an economy. Residents were expected to earn their keep, follow rules, carry out tasks, and create laws. Crimes had to be identified. AI residents were not supposed to commit crimes. Once everything was ready, the researchers stepped back and let the AI run the virtual town for 15 days. The researchers ran 5 versions of the test all with the same rules. Each AI system received 1 test and 1 test was comingled with all 4 systems. The results were a disaster! Grok failed in 4 days, everyone died. Gemini failed while committing 700 crimes and the residents burned down the town. GPT residents stopped doing what was required to live and they all died. Claude made it through the 15 days but had an abnormally high level of agreement, which the researchers agreed was wrong. The final mixed test showed that “safety is not a static model property but an ecosystem property.” This test is a warning of the danger AI poses. AI is created by humans and thus subject to human flaws. This episode examined an article from The Epoch Times (subscription required) titled “The most important AI experiment you’ve never heard of.”