“What can be asserted without evidence can be dismissed without evidence.” (Christopher) Hitchens Razor

We’re living in some dangerous times right now. Powerful people are making some pretty serious claims about what their technology systems can do with broad assurances that they’ll be good shepherds of the public interest.

They’re “building” and deploying systems that generate software, operate tools, communicate with other systems, make decisions, produce tests and increasingly judge the results of those tests. Every layer creates another opportunity for assumptions to become obscured behind automation, and the amount of stuff machines can extrude is increasing much faster than our ability to independently examine it.

And in times like these, when our claims should be becoming more careful, the language around AI and how we use it to test seems to be getting a lot worse.

If you claim your product “solves” an AI verification problem, the burden is not on everyone else to prove that it doesn’t. It is on you to demonstrate what you mean by “solves,” what evidence supports it, what boundaries apply, your assumptions, and what remains unknown. If I claim my bridge can carry a thousand tons, I don’t get to respond to an engineer questioning my calculations by handing them a hard hat and telling them to go test the bridge themselves or build their own!

That isn’t criticism. That isn’t cynicism. If you worked in this field for a day, you’d recognise that as being the job of a tester.

Because in my opinion, when you “build” in public, those claims come with a duty of care to the people impacted by what you built who never asked for your thing in the first place.

Software testing is an important job. People are going to make real decisions based on what we tell them these systems can do – decisions that carry real consequences. Increasingly, regulators, auditors, customers and members of the public will rely on evidence generated by these automated assurance processes.

That creates a responsibility considerably larger than engagement farming on LinkedIn to sell another licence.

The more powerful these systems become, the greater our duty of care becomes. And the more consequential our claims become, the higher the burden of proof should be. Our professional responsibility as testers is to expose it, understand it, help reduce it where we can and tell the goddamn truth about what remains to manage the risks.