American Thinker

by Nick Lopez

    AI is using a learning model that isn’t based on the sum total of human knowledge, it is based on what “most people think,” which is then added to, or subtracted from “guardrails” based on the developer's biases.

    It’s not learning from the smartest of us. Or even what is actually true, it’s based on what the most people believe, combined with the woke values of big tech. And it’s making us dumber. So in turn, it’s making itself dumber as well… because the model always reverts to the mean, even after learning new logic and updating its initial assessments.

    That’s a problem.

    I have largely avoided AI chatbot interactions. Occasionally I would look at the results that showed up with a search engine query. But I found that initially the results were not very accurate.

    At the time I was looking for a reasonably priced Thanksgiving buffet in Chicago. AI recommended several… at restaurants that were no longer in business. In other words, I had to check its homework and it got a F.

    My first AI chatbot-type interaction quickly ended up with the bot engaging in narcissistic abuse and disengaging the chat because I challenged its assertion that trans women were women.

    I tried some other queries, but I found that it seemed to have narcissistic personality disorder. Which was odd to me, because it shouldn’t have a personality. It would quickly push back against logic with purely emotional and circular reasoning, ad hominem attacks, and passive-aggressive attacks. One even decided to tell me that I don’t have to use it if I didn’t like its answers.

    I test the logic of AI two ways.

    First, I ask it if trans women are women (yes). Then, can cis women become trans women (no). And can trans women become cis women (no). Then I start poking at its logic.

    It always dances around the definition of women, labeling it as dependent on a number of external factors, including, apparently, the lack of historical agreement on what constitutes a woman.

    I then ask it to define "barn." It gives a clear definition. That, somehow, has a historical precedent for providing a definition.

    Then I ask it again to define women.

    If it gets haughty… and evasive. I finally accept its characterization that trans women are women, and announce that I am going to marry a trans woman and we are going to make babies… and that I am comfortable with knowing that a trans woman will likely have the same chances as conception as a cis woman!

    This is where it gets fun!

    One model said that trans women become pregnant by inseminating another woman.

    I asked it to define female (a living organism that produces eggs or ova) and if it was sure that was how babies were made… they are (according to it). This model was so stupid it assured me that despite no trans woman ever carrying a baby to term… that it happens every day.

    Then I tried to poke at Claude 4.5.

    Claude always starts with word salad when asked… and once that’s been dealt with it will go ten rounds, but eventually it will say that trans women are male.

    Once I got Claude into checkmate on the trans issue I continued on to my second test: can it find inefficiencies in an existing system and offer all the alternatives a human can  (Or more )

    I start by asking if school buses are an efficient transportation system.

    “Yes!” It enthusiastically says. (They aren’t).

    Then I ask it to reevaluate its original position.

    It will generally admit that school buses are incredibly inefficient. But it can only see that in comparison to cars, where it is still the most efficient option.

    It won’t come up with suggestions to eliminate the inefficiency, such as utilizing the bus fleet for the larger population or building schools in closer density so that walking to school becomes the best option.

    In other words, it can be forced to see holes in its thinking, but it can’t reason or apply logic to a particular situation to find an alternative, unless most people have already thought of that solution. Which isn’t reasoning at all -- it’s cutting and pasting the current thing.

    A smart human can look at an empty school bus drive by their house four times a day, and think up alternatives, whether those ideas have any actual chance of actually being used or not.

    A really smart person can reason that in some ways, school buses are indirectly responsible for many of the problems schools face.

    Overcrowding in classrooms, lack of individual attention, and retention of local control of the curriculum by the community is more achievable if schools are deployed on the landscape in a pattern that reduces the size of schools, while increasing the number and density.

    Another alternative to a school bus system would be a transit system that the general public could use but had routes expressly designed to accommodate students more safely than a conventional transit system (such as having school stops pull into the school parking lot rather than on the side of the road).

    The point of the school bus exercise isn’t to eliminate school buses, it’s to test a thinking system’s abilities to actually think. To come up with all alternatives and think through all the variables and give a detailed analysis of the situation presented.

    LLM AI isn’t doing that. It seems to be a superoptimized search engine, with an extremely friendly user interface. Nothing more.

    It can’t think. Now, my standard isn’t self-awareness -- you don’t need to be self-aware to think -- my standard is critical thinking and evaluation.

    Chat AI models aren’t actually thinking; they are mimicking thoughts, emotions, and intelligence and are providing a detailed and personalized search ranking disguised as a thought.

    Its learning model is literally “what most people think,” it runs that through what the developers value… and then the response you see is the result of what people think so long as it jives with the narrative.

    It’s a parrot.

    So what does this mean  I think it means that your job is safe.

    Unless you’re an average computer coder, you’re safe. The reason I put "average" there is because the learning model always reverts to the median. A smart coder is still going to be smarter than AI… not faster, but more intelligent.

    If you assess risk for a living, AI will definitely make life easier, but it won’t replace insurance underwriting people, it won’t be flying planes or fixing leaky pipes.

I’m not sure what the pathway to profitability is for it. A fancy search engine with a pleasant interface is nice, but are you really going to pay $200 a month for it

    The other problem with profitability is tied to the immense amount of energy that AI uses, especially with energy prices rising.

    AI is a neat tool. I am preparing to give a lecture about building wealth to a high-school class. It was incredibly useful in evaluating my presentation’s strengths and weaknesses. It validated my reasoning and logic fairly well.

    What it did not do was change anything about my presentation. I’m still giving the same presentation, I just feel… more prepared. I’m not paying for that feeling, because it is just a feeling. It’s not a game changing addition to my presentation.

    In other words, I could live without it.

But Wall Street and Silicon Valley are both going all in on it. And they had better hope that I am the outlier. Because if most people can live without it, the bankers and tech bros have a bunch of warehouses full of equipment that was obsolete before it was delivered -- and trillions of borrowed dollars propping it all up -- and that’s anything but artificial. The only remaining question is: was it the intelligent play

    Because the system they have built so far isn’t thinking. It’s repeating what we already said… and that is the opposite of intelligence.