Editorial

Cogito, Ergo. Schtumm

The annoying thing with AI is that it allows even idiots to sound intelligent - until they have to explain what they just said.

That isn’t a new phenomenon by any means, people have been borrowing impressive and fashionable buzzwords to disguise unimpressive thinking for centuries, corporate language has made a respectable career out of it. AI has just made the process faster, cheaper and available to anyone who can type “make this sound professional.”

What worries me is how easily producing an answer can be mistaken for understanding it. Someone can now generate a polished report, a technical proposal or a persuasive argument without being capable of judging whether it’s actually good or merely sounds good. The output looks like evidence of competence, even when the person presenting it can’t explain how its conclusions were reached or question the results.

I’m starting to see what I can only describe as brain rot, as people are increasingly using AI to produce results that they are not capable of telling if it is actually good or if it just sounds good. Not because using a tool makes someone stupid, but because a tool that produces the appearance of thinking can make it easier to avoid doing any of your own.

The worrying part is that judging the output requires the very skills people may be outsourcing. If you stop practising how to reason, write or solve a problem, you risk becoming less capable of spotting when the machine gets it wrong. You’ll still have an answer, but you may no longer have a reliable way to assess if it is correct.

If erring is human and AI is trained on data from human labour, we shouldn’t expect an AI to be immune to error. Human writing contains insight and ignorance, sound arguments and confident nonsense. AI doesn’t even need to copy an existing mistake to produce a new one either. Whatever the source of the error is, the person using the result needs enough understanding to recognise it.

Instead, “this sounds convincing” starts replacing “I understand why this is right.” Someone can produce a report without being able to defend its conclusions, identify its assumptions or recognise nonsense beneath the fluid language. The report may be excellent, the problem is that they wouldn’t know if it wasn’t.

Throwing more words into the mix doesn’t generate more meaning if the speaker doesn’t understand what they’re saying. A longer explanation may contain useful detail, but length alone doesn’t establish that. Neither does professional formatting, a confident tone or the inclusion of words nobody in the meeting wants to admit they don’t understand.

Ask what a sentence means, and you may get another sentence with more syllables - or an angry retort questioning why you’re asking questions .

Ask what supports the conclusion, and you may get a repeated restatement of the conclusion until eventually, someone describes the proposal as strategic, and the meeting moves on.

AI can make that escape route remarkably convenient. Someone who can’t judge the first answer can ask it to check the answer, explain the answer or produce a more convincing version of the answer. But if they assess every response by how plausible it sounds, the underlying problem hasn’t changed. Another explanation isn’t independent verification merely because it arrived in a separate message.

There are useful ways to check generated work, you can examine the sources, test the code, challenge the assumptions or compare the result against something you know. But those checks all require engagement with the substance from a critical mindset perspective, asking “are you sure?” and accepting “yes” doesn’t provide much of it.

This isn’t an argument against AI - used well, it can help someone explore an unfamiliar subject, work through a problem or find weaknesses in an argument. Asking it to explain something, then questioning that explanation until you understand it, is different from asking it to produce something you can pass on without reading.

The distinction is whether the tool helps you develop understanding or helps you get away without it.

In a workplace that rewards visible output, getting away without it can look remarkably productive. The document arrives sooner, the presentation looks better and the author appears to have mastered the subject overnight. The gap becomes visible when someone asks a question that wasn’t covered in the generated text, or when a decision based on it has to survive contact with reality.

That’s when the ability to explain, adapt and correct the work becomes more valuable than the ability to produce it.

AI can write the report, but it can’t make the person presenting it understand what it contains. We shouldn’t confuse access to an answer with possession of the knowledge behind it.

The danger isn’t about outsourcing the work, it’s outsourcing the judgment while retaining the confidence.

How we as a species proceed will determine whether AI provides cognitive benefits or a slow cognitive decline.

I use AI extensively at work, it is a priceless tool that comes with a heavy price tag. The genie will not be put back into the bottle.

This morning, I asked my AI sidekick to create a 1-pager from a massive wall of text we had produced in a brain storming session.

it reduced the font size to 7pt, which made everything fit.