AI and the problems that follow a breakthrough

November 13, 2026

In 1923, 42,000 people in Cincinnati signed a petition to force cars to have mechanical devices fitted that would stop them going above 25 mph.

People were being killed on the roads. There was real anger about it, and a serious argument about how much freedom drivers should have. The proposal went to a vote and failed. [1]
I started looking into this after a friend shared a post about a lawyer having to spend longer sorting out AI-generated contracts and arguments than he previously spent doing the work himself.

Before he can help, he has to unpick what the client has brought him, explain why it doesn’t work, and try to do that without offending them.
I can see why that’s frustrating. But it also feels like a familiar stage in how we deal with something new.

We create something powerful, discover the problems, some people misuse it, and people demand change. Then we gradually make it work better.

Featured Content

Download our latest Salary Guide 2026

The problems tend to arrive before the rules

Take railways. In 1889, the Armagh rail disaster killed 80 people. It helped bring about legislation giving the Board of Trade powers to require automatic continuous brakes on passenger trains, alongside other safety measures. [2]

The technology existed. Some companies were already using it. The law gave the authorities powers to make others act. [2]

With aviation, a mid-air collision over the Grand Canyon in 1956 killed all 128 people on the two aircraft. It became a major factor in the creation of the Federal Aviation Agency in 1958 and changes to air traffic control. [3]

Those improvements took pressure, engineering and decisions about what operators could be allowed to do. Getting used to flying wasn’t enough.
The internet brought a different set of problems. Email made it cheap and easy to reach people, which also made it cheap and easy to fill their inboxes with rubbish. In 2003, the US passed the CAN-SPAM Act specifically to tackle unsolicited commercial email. [4]

That didn’t make spam disappear. But the problem had become big enough to require rules of its own.

There’s a pattern here, although I wouldn’t pretend it’s a comforting one. People were harmed while we worked things out. Improvements sometimes took far too long. And the existence of rules doesn’t mean the problems have gone away.

Does AI fit that pattern

At the day-to-day level, I think it does.
We’ve given people a tool that can produce a convincing answer on almost anything. That’s useful. It also makes it easier to feel informed without understanding enough to judge whether the answer is any good.

The Alex Spiro clip that accompanied the post gets at that. His point was that clients thinking they know better than their lawyer is an old human problem. [5]

AI gives that confidence something polished to arrive with.
A contract. A legal argument. A technical document. Something that looks finished, even when the person bringing it can’t explain or defend what’s in it.

For me, that means the work ahead includes getting people to understand what they’re using and take responsibility for what they put out. The technology needs to improve too. I expect both will happen, although neither will happen just because we wait.

Where the nuclear comparison matters

The bigger questions about AI are harder to fit into that everyday picture.

If we’re talking about the possibility of losing control of powerful systems, I think nuclear weapons are worth discussing as a comparison. I’m not claiming the technologies or their risks are identical.

Public concern over radioactive fallout, including contamination entering the food chain, helped bring about the Partial Test Ban Treaty in 1963. It prohibited participating countries’ nuclear tests in the atmosphere, outer space and underwater. Underground testing continued. [6]

We put limits around a danger we had created. We didn’t remove it.
That’s why “we’ve been here before” only takes the argument so far. The history doesn’t promise that every invention becomes safe, or that every use deserves to continue.

Concorde is another useful reminder. It wasn’t banned outright: restrictions on supersonic flight over land limited its operation, and British Airways eventually retired it in 2003, citing falling revenue and rising maintenance costs. A remarkable invention didn’t become the normal way we travel. [7]

I’m hopeful about AI. I think a lot of the mess we’re seeing in everyday work is something we can improve as the tools get better and people learn where they help.
But the practical problems need attention now, and the bigger safety questions deserve serious scrutiny.

Maybe humans won’t need to make a comeback. Maybe we’ll get clearer about where we still need human judgement.