Top Ten AI Concerns

Enthusiasm and unease can exist together. Steve Bolton’s account of finding AI useful during a difficult period does not lead him to dismiss the questions it raises.

In Think With Me, he groups ten concerns into two sets. The first involves systems and decisions larger than any individual user. The second focuses on how a person works with AI day to day. This is a way of organising attention, rather than a ranking of risks or a claim that the boundary between the groups is absolute.

Five wider challenges

1. Job loss and economic displacement

The book asks readers to take seriously the effects of changing tasks and roles. An individual cannot determine the outcome for an entire economy. A business leader can examine the work changing inside the organisation, involve the people affected and consider which capabilities they will need to develop. The wider labour-market question remains open.

2. Misinformation and the loss of shared truth

A convincing answer may lack reliable evidence. Steve treats verification as part of the work: finding the underlying source, comparing the claim with it and recognising when a mistake would matter. A polished summary should make information easier to examine, without becoming the only basis for trusting it.

3. Long-term risk

The manuscript acknowledges disagreement about the likelihood and scale of the most extreme outcomes from advanced AI. It does not offer a prediction. Its position is that consequential uncertainty deserves attention, even when it cannot be resolved by the person using a tool to complete today’s task.

4. Privacy and surveillance

A prompt may contain more than the writer initially notices: a customer’s identity, an employee’s circumstances, financial information or a confidential plan. The book places judgement before disclosure. Which information is necessary, whose information is it and is this an appropriate place to share it? Those questions sit alongside wider decisions about products and governance.

5. Environmental cost

AI depends on physical infrastructure, including electricity, cooling, processors and data centres. The book gives that reality a place alongside the apparent ease of a conversation on a screen. Procurement, infrastructure and investment decisions bear on the wider issue; it should not disappear simply because the resource use is out of sight.

Five areas for everyday judgement

6. Thinking and judgement

Consider how much of a decision has been delegated. If AI frames the problem, chooses the evidence, assesses its own answer and recommends the action, the person may have handed over more than intended. Retaining the work of framing, questioning and deciding keeps human judgement active.

7. Bias

The book asks readers to look for omissions and assumptions as well as errors. Whose perspective is missing? Which circumstances have been treated as normal? What would make the answer different? Asking for another view can reveal something worth examining, but the resulting answer still needs assessment.

8. The black box and loss of control

An explanation of an answer does not establish that the answer is correct. Useful checks include examining sources, testing assumptions and considering alternatives. Steve’s principle is to make the strength of the check proportionate to the consequence of acting on the result.

9. Dependency

AI can be useful without becoming necessary to every task. The practical question is whether the person can still begin, assess and finish appropriate work independently. Deliberately keeping some tasks human provides room to exercise those capabilities and to notice when reliance is growing.

10. The pace of change

A succession of new models and features can draw attention away from the reason for using any of them. The book returns to the actual task: what needs to be understood, decided, made or changed? That question provides a stable starting point while the available technology continues to change.

Let concern inform the practice

These concerns need not be settled once and for all before useful work can begin. They can shape choices about the information shared, the authority delegated, the evidence checked and the work retained by people.

Steve’s own experience gives the question particular weight. AI became useful when he was exhausted and carrying a great deal. In the book, he recognises that those were also circumstances in which he could have handed over more of his judgement than he intended.