The Return of Wisdom

Symphony3 recently hosted The Return of Wisdom, a webinar with Professor Ronan McDonald from the University of Melbourne, hosted by our co-founder Fergal Coleman. Professor Ronan McDonald came to talk about a question that sits underneath almost every conversation that leaders are having about AI right now: when machines get very good at sounding like they know what they are doing, what happens to human judgement?

The talk kept returning to a handful of ideas that are worth pulling out on their own. Here they are, along with what Professor Ronan McDonald actually said about each.

Practical wisdom

Ronan built the talk around phronesis, the Greek word for practical wisdom. He was careful to separate it from cleverness and from technical skill. Practical wisdom is judgement disciplined by experience, proportion and ethical insight: the capacity to work out what should be done here, now, with these people, under these constraints.

He framed this against a distinction borrowed from philosopher Brian Cantwell Smith: reckoning versus judgement. AI reckons well. It calculates, classifies, predicts and produces plausible answers. Judgement is different. It asks what matters, what is at stake, and who carries the consequences. Ronan's point was blunt: AI does not need real intelligence to do a lot of what knowledge workers are paid for. A manager does not need a tool with consciousness. They need a draft, a summary, a set of options. That is enough to change how work gets done.

Human friction against AI's smoothness

This was the phrase Fergal kept coming back to, and Ronan built much of the talk around it. AI is fast and smooth. It gives you a polished answer with very little visible effort. Ronan's argument was that leaders need to deliberately introduce friction against that smoothness: to slow down and interrogate an answer rather than accept it because it reads well.

The reasoning goes back to writing itself. Ronan argued that writing is not just where thought gets recorded, it is where thought happens. The awkward first draft is where contradictions surface and where you discover what you actually think. If AI supplies the polished version from the start, you skip the stage where your own judgement would normally form.

Ronan had an image for what this friction looks like in practice: brush against the grain. Rather than accepting an AI draft as the finished thing, he argued you need to learn to push against it. He linked this to how we have handled earlier tools we now rely on: we still teach children arithmetic decades after the calculator, because the underlying competency matters beyond the task itself. The same logic applies to AI-produced drafts. Using the tool does not mean accepting its output uncritically.

Strategic slowness

Closely tied to friction was what Ronan called strategic slowness: a deliberate counterweight to AI's speed. He was clear this is not a case against using AI. It is a case for building in moments where the pace slows down enough for a person to actually look at what has been produced, rather than letting fluency substitute for scrutiny. Fluency, he said more than once, is not evidence.

Owning responsibility

This was the idea Ronan returned to most directly, and the one the audience picked up on immediately. His example was the Air Canada case, where it was found that  the airline was responsible for information given by its own chatbot, not the chatbot itself. The lesson: judgement cannot be laundered through the tool. The organisation always answers for what is done in its name.

He also touched on disclosure. There is no settled norm yet for saying AI helped write something, and Ronan expects that to change. He expects it will become standard practice to note that AI helped draft a document, the same way disclosure conventions eventually catch up with new tools once the etiquette settles. It was this thread that led to his five practical questions for leaders: what to automate, what to still question, what needs explaining, when to escalate, and who remains accountable. None of these are complicated on their own. The discipline is in asking them every time, not just when something has already gone wrong.

What's worth protecting

Ronan named six areas where judgement can quietly erode if we are not paying attention:

• Verification - Fluency is not evidence. Keep asking what is true, what has been left out, and where something came from.
• First-draft thinking - Skipping the rough draft means skipping the point where thinking actually happens.
• Voice - AI imitates tone well, but leaders still need to ask whether that tone is genuinely theirs.
• Institutional memory - AI can summarise the record. It does not know the scars, or why a phrase inflamed a community before.
• Tolerance for ambiguity - AI produces clean answers quickly. Some situations call for hesitation or escalation instead.
• Responsibility - The most dangerous loss is people feeling less accountable because the answer came from a machine.

The closing line

Ronan ended with a line worth sitting with:

“The future belongs not to those who automate most, but to those who know what not to automate.”

For councils and organisations working through their own AI adoption, that is the real work. Not whether to use AI, but where to deliberately keep the friction, the slowness and the judgement that only people can provide.