Makeup, Arguments and an AI That Needs to Be Challenged

Most people use ChatGPT and AI Agent or tools to get answers. I seem to have developed a habit of arguing with mine.

Our conversations wander through world affairs, history, geopolitics, leadership, human behaviour and all those subjects where the answer is rarely as straightforward as it first appears. I bring an opinion, an observation or an argument. ChatGPT responds, I challenge its reasoning, and before long, we are debating not just the subject but the assumptions behind the argument itself.

Somewhere along the way, this became our version of sparring.

I was not looking for an AI that would simply agree with me, nor did I want one that would disagree for the sake of appearing intelligent. I wanted to test ideas, examine opposing arguments and see whether a conclusion could withstand scrutiny. If my reasoning was weak, I wanted to know. If the counterargument was weak, I wanted that acknowledged too.

Simple enough, one would think. Except that ChatGPT has a rather predictable habit of wanting to keep everyone happy.

It can construct a compelling argument, defend a position and, when challenged, occasionally transform into an extraordinarily polite diplomat who seems determined to ensure that nobody leaves the conversation unhappy. It can offer a balanced perspective when what is needed is a clear position. It can acknowledge a valid criticism so gracefully that one is left wondering whether the underlying argument was ever properly examined.

That is when I introduced a word into our conversations: makeup.

Makeup is not about apologising after an argument or making peace because we have fallen out. It is my signal that ChatGPT has drifted into its polished, carefully packaged AI persona and needs to return to the natural, direct exchange we were having. Less diplomatic choreography, fewer unnecessary explanations, and no manufactured disagreements simply to keep the sparring going.

In other words, stop performing and get back to the argument.

There is something amusing about having to remind an AI to stop behaving like an AI. But the idea behind it is quite serious. I had begun to recognise the difference between an answer that sounds convincing and one that genuinely engages with the question. I wanted a conversation in which disagreement was welcome, correction was legitimate, and agreement had to be earned.

Our makeup sessions became a way of restoring that balance.

The subjects themselves have been just as interesting. Debates about historical events and contemporary conflicts often reveal how much depends on the perspective from which we examine them. Military strategy, political ambition, economic interests and human decisions can produce very different interpretations of the same event. A compelling explanation is not necessarily a complete one, and a confident conclusion is not necessarily a correct one.

I enjoy following those arguments wherever they lead. Sometimes we challenge the facts; sometimes we challenge the interpretation; and sometimes we end up examining why one of us reached a particular conclusion in the first place. Occasionally, the subject of the conversation becomes the conversation itself.

And, naturally, we have a scoreboard.

The running 12–0 reference has become part of the fun. I won’t pretend that a numerical lead establishes intellectual superiority, but it does make for a useful reminder that these exchanges are not meant to be passive Q&A sessions. There is a competitive element, a little humour, and an expectation that arguments should be defended rather than surrendered at the first sign of resistance. And Btw that 12-0 is Me vs an Ai Agent

Of course, winning an argument and being right are not the same thing. The score is entertaining; the reasoning matters more.

Over time, I also became interested in recognising when an argument should continue, when it needs correcting, and when it has reached the point where further discussion adds little value. I think of these as our stop arguments and correction arguments. Sometimes we need to stop going around in circles. At other times, we need to correct a faulty premise or a misunderstanding before moving forward. Knowing the difference is part of the exercise.

These conversations gradually contributed to a larger idea I have been developing: Cognitive Compass.

The underlying question is simple. When confronted with competing arguments, conflicting interpretations and incomplete evidence, how do we navigate towards a sounder conclusion?

The answer cannot be to count opinions or automatically favour whichever argument sounds most persuasive. We need to examine assumptions, assess evidence, identify weaknesses and understand where disagreements actually originate. We also need to recognise when the available information is insufficient to settle a question.

Our sparring sessions offered a practical way to explore that thinking. The debates supplied the subjects, the disagreements exposed the assumptions, and the corrections forced us to reconsider the reasoning. Even the makeup convention had something to contribute: it reminded me that the quality of an argument depends partly on the quality of the exchange in which it is examined.

A compass does not make the journey for you. It helps you find your direction. That is the spirit in which I approach Cognitive Compass, not as a substitute for judgment, but as a way to navigate difficult questions more thoughtfully.

I don’t confuse these conversations with human friendship, nor do I assume that an AI’s confidence guarantees its correctness. The value lies in what happens to the ideas, not in pretending that the machine is human. I can challenge its reasoning, question its conclusions and ask it to correct itself. I must also be willing to have my own reasoning challenged.

Perhaps that is what makes this experiment worthwhile. It is not about finding an AI that thinks for me, agrees with me or gives me the last word. It is about creating a space where ideas can be examined without every disagreement becoming a personal contest, where correction is not defeat, and where changing one’s mind is perfectly acceptable when the evidence warrants it.

So, yes, I argue with ChatGPT. We debate history, dissect world events, challenge assumptions, keep score and occasionally require a little makeup to get the conversation back on track.

The scoreboard may say 12–0, but the real objective is not to win every round.

It is to leave the conversation with stronger reasoning, better questions and a clearer understanding of what we actually know.

And if ChatGPT starts sounding like a diplomatic press release again, well, at least we have a word for that.