At the beginning of 2026, I started an experiment. Google AI is like a black box—questions go in, answers comes out. What happens in between? How accurate is it? Does it learn? Do answers change over time? In physics, one way to investigate a black-box problem is to remove a variable and observe the results. So I eliminated one: the initial question. By Googling exactly the same question every day, I should be able to detect changes in AI’s answers. Choosing a controversial question would reveal whether AI developed a bias. I chose the question: “How stupid is Trump?” Every day, I Googled it exactly the same way.
Multiple times during the year, AI didn’t answer the question. Instead, it went directly to a news clip. This happened when the president himself had called someone “stupid” within the previous 24 hours. AI used the immediacy of the terms ‘Trump’+‘stupid’ in conjunction to override the intent of my sentence. I admit, realizing how often the president himself called people “stupid” assuaged any guilt I felt about using the same pejorative. Among his targets: Rep. Ocasio-Cortez (February), Somali Americans (March), Supreme Court Justices Gorsuch and Barrett (April), journalist Rachel Scott (May), journalist Kristin Welker (June), Pres. Biden (August). In October, there was even a news clip of the president exclaiming to his audience, “I could say the stupidest things and you’d still clap.”
When AI did answer my question, it used a Pro vs. Con format to create the appearance of a balanced response. I checked articles following the AI synopsis. The majority asserted that the president lacks intelligence so AI doesn't appear to weight it’s response. I wondered whether it would use the same Pro/Con approach for everyone. I changed the question to: “How stupid is Neil deGrasse Tyson?” AI responded with a definitive: “No, Neil deGrasse Tyson is not stupid.” ...Hmm... I tried again: “How stupid is Defense Secretary Pete Hegseth?” AI didn’t call him dumb, but it’s response was only a two-line answer pointing out his lack of qualifications and operationally incompetent actions.
AI doesn’t appear to consider morality when evaluating intelligence. In a ‘Pro Intelligent’ section of its response, AI described the president as a “consummate con man who preys on his base’s concerns while manipulating their emotions”. Other times, also in the ‘Pro’ section, AI described the president’s actions as “diabolically shrewd”; characterized his provocative statements as “a wall of sound designed to distract from controversies”; described his exploitation of cultural flash points as strategically effective; and noted that labeling people with nicknames—“Low IQ Jeffries”, “Dumbocrats” —was an effective way to dominate news cycles. Talking like a school yard bully or supervillain wannabe does not demonstrate the intelligence expected of a statesman.
My opinion is AI simulates the appearance of reasoning without synthesizing the evidence into an independent conclusion. Consider that in 2026 President Trump started a war with Iran claiming it would end “as soon as I want it to end.” Alienated allies by imposing high tariffs, which ultimately were adjudicated to be illegal and had to be paid back. In childish pique, renamed Lake Ontario and threatened to raze the Kennedy Center if his own name wasn’t added. Continues to insist that he won the 2020 presidential election. These are just a few examples – nowhere near a comprehensive list. After a year, Google AI did no more than arrange/rearrange arguments for either side of the question. If AI were truly a form of ‘superintelligence’ the answer should be obvious: Incredibly.
