AI Knowledge: The Expanding Universe Theory

In 1929, Edwin Hubble made one of the more unexpected discoveries in the history of science. Not only did he confirm that the universe was not static, but that it was expanding - galaxies moving away from each other in every direction. Unsettling enough. But the deeper finding, confirmed decades later, was stranger still: the expansion is accelerating and the further away a galaxy is, the faster it recedes. At some point, galaxies currently visible to us will pass beyond what physicists call the cosmological event horizon - and they will be gone from our observable universe forever, not because they no longer exist, but simply because the distance between us grew too fast to close.

The same physics is at work in organisations and governments right now - and many leaders have not noticed yet.

Aside from the AI specialists, there is a growing cohort of leaders, founders, and directors who are genuinely working with AI tools - not at arm's length, not delegating it to a digital team, but embracing it personally, daily, experimentally. They are using large language models to think through strategy, draft and refine communications, build apps and agents, interrogate data, accelerate research, and stress-test their own reasoning and scenarios. And there is another cohort, much bigger, who aren’t. They may have used Copilot or ChatGPT once or twice, and they may be planning to invest more time, but to date, haven’t.

The Expanding Universe Theory of AI Knowledge

And here is the part that should concern anyone who has been waiting for the technology to "settle down" before engaging: the first cohort is not just ahead of those who have not started, they are accelerating away. Every week of practice compounds. Every prompt refined adds to a growing intuition about what these tools can and cannot do. The gap between them and the non-adopters is not static. It is expanding, and the further along on the learning journey they are the faster they are developing. It won’t be long before groups of these early adopters are so far ahead, the non-adopters will, figuratively at least, lose sight of them forever.

Unlike most professional development gaps - where a determined person can read the right books, attend the right course, and close the distance in a reasonable timeframe - this one has a structural acceleration built into it. Even allowing for the hype cycle, the tools themselves are improving more rapidly than almost anyone predicted. The people already using them are improving faster still. Those on the sidelines are not standing still; they are falling behind at an increasing rate.

I want to be careful not to be melodramatic about this. Learning AI tools is not the only thing on a leader's development list, and it would be glib to suggest otherwise. The craft of leadership – balancing trade-offs, reading people, building trust, making decisions under uncertainty, holding an organisation together through difficulty - is not going to be replaced by a prompt. Neither is deep industry knowledge, regulatory literacy, financial acumen, or the accumulated judgment that comes from years of hard experience. These things matter enormously and always will.

But AI is different from most other contemporary skills on that list, and the difference matters. Sharpening your executive presence, running better board meetings, understanding a new accounting standard - these are valuable, but they are largely incremental improvements to capabilities you already have. AI represents a genuinely new category of tool, with a genuinely new set of use cases, risks, and strategic implications. It is not an upgrade. It is a new instrument entirely. And the longer you wait to pick it up, the more the people around you - including your own teams, your competitors, and the advisers you rely on - will have developed a fluency you lack.

So what is a time-poor senior leader actually supposed to do?

What to do

Simple, start using it.

Not at arm's length, not through a briefing paper, not by asking someone to tell you about it.

Pay for access to at least one leading large language model - Claude, ChatGPT, and Gemini are the obvious candidates - and then actually use it, for real work. In fact, pay for two. Using more than one gives you a useful sense of their different characters, and the combined annual cost is roughly equivalent to a long lunch, if anyone still does those these days. There is no meaningful financial barrier here. The barrier is mostly psychological: a vague sense that you need to understand it properly before you start. You do not. You learn by doing, and the doing is genuinely accessible.

Before you load anything sensitive into these tools, spend twenty minutes understanding how they handle your data. The major platforms have business and enterprise tiers with meaningfully stronger data protections than their consumer versions. Your organisation's information, your clients' data, your own strategic thinking - none of it should be running through a free consumer product if you have not checked how it is stored or used. This is not paranoia; it is basic professional hygiene, and most platforms publish this information clearly.

Then learn how they actually work. A reasonable place to start is asking the model itself to explain the basics relevant to your role - which is, in itself, a useful first experiment. I would also recommend a book published recently by a friend of mine, Scott Cowans: The AI Factory Tour. It gets technical in parts, but it is designed for leaders, not engineers, and it does something most AI writing does not - it takes seriously both the mechanics and the limitations of these models, including some pointed observations about the concentration of AI power in very few hands.

Once you are using these tools, pay close attention to what they produce. They are extraordinarily capable. If nothing else they can, and should, save you hours of work, which can be redirected into other more human activities.

Don’t leave yourself at the door

But you will also notice other things – the way they sound, the way they structure responses, the way they might subtly (or disastrously) get things wrong. And, unless you instruct them otherwise, the way they default to American English, e.g. the em dash. For those outside the States the em dash—a widely used punctuation mark in North America—is a classic tell of AI produced content. It is practically invisible in Australian professional writing, yet it appears in AI-generated content with the frequency of a nervous tic. If your communications are liberally festooned with em dashes, you are essentially wearing a sign that says “I didn’t write this”.

More broadly, AI output left unedited tends to be fluent and well-structured, yet oddly characterless - polished in the way that a very confident person with nothing interesting to say can be polished.

Learning to edit AI output, push back on it, and make it actually sound like you, that is a skill in itself, and it only develops through regular, critical use.

Not everyone will agree with my little theory, but one thing I am sure of is that unless you are planning to retire as an analogue hermit, ignoring the AI tsunami is a grave error.

Do not go gentle into that good night

There is a version of this moment where someone reads this, nods, and then does nothing - because there are twelve other things that need attention and this feels like it can wait until next month. I understand that impulse completely.

But the expanding universe does not wait for a convenient time. The cosmological event horizon does not care about your calendar.

The practical question is not whether to develop AI literacy. That question has already been answered. The question is how far behind you are prepared to fall before you start.

While I have borrowed from science for these thoughts, I will leave you with the words of Dylan Thomas, published 75 years ago, which I think are apt:

Do not go gentle into that good night,

Old age should burn and rave at close of day;

Rage, rage against the dying of the light.