Critical thinking in the age of ai

I do this. A lot. Claude is an excellent tool for sharpening my writing and challenging my thinking, especially as I have taken the time to train it on how I think. Now I can argue with myself, which saves others the effort.

But that effort and that friction are valuable, in a way that I’m not convinced is as well-recognised as it should be. Zoe Scaman recently wrote about the importance of friction in generating good ideas in the creative space (Dull Knives), and the difficulty that AI has with this. She was looking through the lens of training and refining AI models, I’m talking about using AI models to train and refine my own processes. Same lack of friction. Same problem.

Critique and the challenge of our work matter, and we lose something when we abdicate this, especially when the reviewer will never ask us to provide that challenge to them in return. Not just the ability to critique and to think critically, but also the ability to formulate and articulate an argument. The willingness to sharpen the ideas of ourselves and others on the roughest of stones, in a way that means people still interact with us at the end of the process. Most importantly of all, the ability to adapt our thinking and our argument as the available evidence changes.

This is a cornerstone of most fast-paced, high-pressure, decision-orientated roles – operating at night in a low-level helicopter formation might not seem the time for existential debate (it isn’t), but the interrogative muscles built by not taking things at face value are the same ones that allow you to spot something out of place and not just assume that someone else is right and that you are wrong but rather to know how much to question, for how long, and what to do with that intuition. Because intuition on its own is no use when the stakes are high and time is short – the ability to assess the situation and act decisively with incomplete information requires a deep understanding of not just the domain and the context but also of yourself and how you react and process things. This doesn’t come about through luck, nor even by practice – it comes about by challenging yourself and others, by understanding how you think and how you react when pushed.

You can’t teach experience, but you can teach the process that is used to build it – if you can surface the process. The surfacing of unconscious competence is fascinating and a topic for another time, but the key here is competence – if we remove an element of the skill, then the level of competence drops. If the measure of competence is the quality of output then maybe we don’t see this for years – my writing has definitely improved as a result of having a copywriting partner with all of human knowledge at its fingertips. But that doesn’t mean that the thing that underpins my writing – my thinking – has improved. And at some point, like all muscles, that mental ability will wither until when I do need to use it I won’t have it anymore.

Easy. You train Claude to push back. You teach Grok to argue. You tell Gemini to play devil’s advocate and to make you defend your position and ideas. That’s not the way that most people use these tools, but it’s certainly a good way to do it. That way you get to play both roles and you get to sharpen your skills on both sides of the stone. Even better if before you switch perspectives you get the model to explain why it has made the editing suggestions that it has to your ideas – that way you can learn from the critiques that it has made to allow you to improve your critiques of other people’s work as well as your own.

If it was that straightforward then this would be a slightly pointless piece and wouldn’t add anything to the conversation – but the very act of writing this gave me a worked example of the real value of keeping the challenger role honed. Drafting on a flight back from holiday, my long-suffering wife read the half-formed and unedited brain dump. A lengthy back-and-forth trying to explain that the point I was making was about the 2 sides of the argument being valuable, her unable to understand why I thought I couldn’t do this with AI because she gets CoPilot to explain why it has suggested the edits that it has. Does that all the time.

I hadn’t thought of doing that explicitly – sometimes it happens, sometimes not, but it turns out we were making a very similar point in very different ways due to the misunderstanding. An exchange that at times bordered on exasperation because “she wasn’t getting it” (she was, turns out she was 2 steps ahead of me which is why her unspoken criticism of “that’s obvious” was coming through loud and clear). Resistance with a slightly confused expression, delivered with kindness by someone who was under no obligation to improve my thinking but still did so without making me so defensive of my ideas that I couldn’t hear her. Because of how the conversation went, I was still engaged enough to not just change my idea but to use the whole exchange as a worked example of what I was trying (poorly) to explain. Didn’t try and justify it away or dismiss the critique but rather used it to sharpen both my explanation and my idea. The critique isn’t the point, landing it with the recipient is.

Even if you got Claude to not only explain the critique but also explain why it had gone down that route and chosen those approaches to analyse the work and then got it to play devil’s advocate against your (now improved) ideas you still wouldn’t have the same outcome as having the discussion with a real person. Because the way you interact with an algorithm is different from how you interact with Alan from accounts.

Delivering the constructive feedback to a person who is invested in the idea, who outranks you or pays you, who might be hurt, who might retaliate, who has to keep working with you afterwards — that carries real risk in it, and risk is what makes this a skill rather than a technique. Claude does not get hurt. Claude does not go quiet in the next meeting. If someone doesn’t listen because your delivery sucks, then it doesn’t matter how right you are because they won’t hear you and nothing will change.

But Claude did write the last paragraph, and I left it in a draft verbatim and over time absorbed the words rather than the ideas until I got called out on it. By Claude.

And the only sentence in the paragraph which is actually mine is about how people stop listening if your delivery sucks.

I chose to not only leave my failings in the post but to highlight them because I enjoy trying to work things out in public. Yes it’s embarrassing and uncomfortable but I believe that if you want to set an example then you need to show your working, not just the outcome, otherwise people will reach for the result and not the process. Decisions made based on incomplete understanding are about the most dangerous ones around, because there isn’t even any uncertainty to cause you to ask questions about why.

The irony of this as a lesson in a piece about atrophying process is clear to me, and explaining the screw-ups as they happen could easily come across as being too cute. I’ll take that risk to make sure that people have the full story and can see the working. That’s worth marks on its own.

So what?

The use of AI as a thinking sharpener does not have to be a zero-sum game, but the additive version takes thought and deliberate effort. The value of the exercise has to be worth it to you – physical exercise is hard but we do it because the pay-off is worth it. If all you want is the performance of competence (looking good on the beach) then an operation or a drug is potentially a better way of achieving the aim, but it won’t help you be strong enough to lift your grandchild or climb that mountain.

Both are perfectly acceptable choices, but in the exercise example it is a conscious choice – that matters. Choosing not to keep up the practice of critical thinking and analysis because you don’t think it is valuable is one thing. Believing that you still have the skills because someone else is providing the heavy lifting is quite another.