Why AI strategy needs more pessimists
Look around the table in many AI discussions and you will find a familiar mix: an innovation lead, a technology lead, a data lead and, often, a supplier. All capable people. All, by the nature of their roles, optimists.
That is not a criticism. Optimism is what gets pilots started. But a room full of it tends to produce the same result: enthusiasm mistaken for evidence, and plans with no answer to the question "what if this doesn't work?"
Boards need someone asking that question. Not to block progress, but to make it survivable.
Strategic pessimism is a skill
Readers of The Hitchhiker's Guide to the Galaxy will remember Marvin, the gloomy robot who was usually right about what would go wrong while everyone around him cheerfully walked into it. Marvin was poor company, but he would have made a useful committee member.
Strategic pessimism means imagining failure clearly enough to design around it. For AI, that means asking:
- What if it doesn't work? What is the fallback, and what have we switched off in the meantime?
- What if our competitors do the same? If everyone buys the same tool, where is the advantage?
- What if we are solving a problem that will not exist in three years?
- What if we are adding complexity to a process that should have been simplified first?
The risks many frameworks miss
Many AI risk frameworks focus, rightly, on bias, fairness and transparency. Those make AI safer to use. Fewer ask whether we should be relying on it in the first place:
- Dependence. If a critical process relies on a model or service you do not control, what happens when the supplier changes the price, the terms or the model itself?
- Automating the wrong thing. A chatbot that handles 10,000 queries sounds impressive. The better question is how many of those queries were caused by a confusing website or a broken process. AI applied to waste produces faster waste.
- Losing the ability to check. If people stop doing the work, does the organisation keep the knowledge needed to tell whether the output is right?
What a board can ask for
- A named owner for each significant use of AI, accountable for its risks as well as its benefits.
- A plain statement of the problem being solved, agreed before the technology is chosen.
- An exit plan: how we would stop using it, and what that would cost.
- Evidence of benefit after the pilot, not just enthusiasm before it.
None of this is anti-AI. The organisations that gain most from it will be those that asked the awkward questions early. Optimists start the journey. Pessimists make sure there is a way home.
