How Cam Stevens Thinks About Prediction Hype Before Safety Technology Scales

Episode 15 of the Headline Podcast, published on March 18, 2026, features Cam Stevens explaining why prediction hype can distract leaders from the operational problem they actually need to solve. His argument is demanding, because a safety technology earns its place only when it improves a real decision before exposure becomes harm.

Cam framed the risk in direct terms: “If we're not clear on the problem to solve, a technology catalog is very dangerous.” That position changes the executive conversation from buying a capability to proving that a specific control, decision, or response will become better.

Why prediction hype starts in the wrong place

Prediction hype starts when leaders ask what a system can detect before defining the safety decision that detection must improve. A useful technology proposal names the exposure, decision owner, response window, and evidence required to show that the intervention worked.

Technology proposals often arrive with polished dashboards, artificial intelligence claims, and a catalogue of possible signals. The catalogue creates a feeling of progress before anyone has agreed on the work problem. A board may approve a pilot because the product sounds advanced, although the operating team still cannot say which decision should change.

Cam's problem-first position challenges that sequence. Start with one decision that already matters, such as whether maintenance can proceed, whether a control is degrading, or whether a workload change has created a new exposure. The tool then has to earn its place by improving that decision.

The distinction aligns with ISO 45001:2018's management-system emphasis on operational planning and control.

How a defined problem changes the investment decision

A defined problem converts a technology purchase into a testable investment. The leader can state the current condition, the decision that is too slow or weak, the person who owns the response, and the result expected after 30 days.

“We need better visibility” is too vague. “The area authority learns about failed isolation verification after the maintenance window has started” identifies a decision, a time window, and a control that can be tested. It also exposes what prediction may not solve.

If the underlying problem is unclear authority, a more accurate model will not transfer accountability. If the problem is a physically unavailable control, an alert may only describe the failure sooner. If response takes 2 hours, a signal that arrives in 2 minutes still has little value unless the organization can act.

Use the difference between a dashboard, escalation matrix, and decision memo to make the investment case concrete.

Why human judgment remains part of the control

Human judgment remains a safety control when conditions are changing, evidence is incomplete, or the model cannot see the work context. A predictive system can prioritize attention, but it cannot remove verification and decision ownership.

Cam did not argue against technology. He argued against using technology as a substitute for thinking. A model may recognize a familiar pattern while missing a temporary change, an unusual combination of tasks, or a constraint only the crew can see.

Define the human action that follows each signal. Does the supervisor pause the job, inspect an isolation, change the sequence, or ask the crew to restate the work boundary? A signal without a response rule creates ambiguity.

The NIOSH hierarchy of controls keeps the discussion grounded. Prediction can support a control strategy, but it does not automatically replace earlier barriers.

When a data stream becomes a false safety signal

A data stream becomes a false safety signal when its availability is mistaken for evidence that exposure is controlled. Leaders should check whether the data represents the work, arrives in time, has stable meaning, and leads to verified action.

One quiet dashboard can mean at least 3 things. Exposure may be low, the system may be working, or the organization may not be collecting evidence that would reveal failure. The number needs context, a denominator, and a decision rule.

A polished score can look more objective than a supervisor's concern, yet it can hide missing data, changing work, or a model trained on conditions that no longer exist. In a 5-test review, ask whether the signal is complete, timely, understandable, actionable, and independently verifiable.

The four states between an observation and an operating decision help teams separate raw activity from evidence that deserves escalation.

How technology can increase psychosocial exposure

Safety technology can increase psychosocial exposure when it changes monitoring, evaluation, or response expectations without clarifying the work design. Cam Stevens warned that technology-driven risk will become increasingly psychosocial.

Cam described that risk as “overwhelmingly psychosocial, driven by technology usage in our organizations.” A system that ranks teams or sends alerts can improve detection while worsening workload, autonomy, role clarity, or trust.

Workers may stop raising concerns because they assume the system already knows. Supervisors may chase every alert and lose time for field conversations. Managers may treat a low score as proof of good performance even when the score reflects incomplete reporting.

Apply the five psychosocial blind spots in technology adoption before the pilot begins.

Prediction versus problem-first technology

Prediction-first technology begins with a capability and searches for a use case. Problem-first technology begins with an exposure and tests whether a tool improves the decision that controls it.

The comparison does not reject predictive tools. It places them inside governance. ISO 31000:2018's risk-management principles support structured, integrated, monitored, and improved decisions.

Recommendation

Choose one high-consequence decision and run a bounded 30-day test before scaling the technology. Define the problem, signal, response owner, verification method, and psychosocial checks, then conduct a 90-day review.

Document the current response time, the evidence available to the decision owner, and the point at which the control can fail. Then ask Cam's question in practical form: what problem are we solving, and what will be different if the tool succeeds?

During the first 30 days, track 5 things: whether the signal arrived, whether the owner understood it, whether the response occurred, whether the control changed, and whether the team experienced new workload or monitoring pressure. At day 90, compare the result with the original problem statement.

Listen to the full Episode 15 conversation with Cam Stevens. The Headline Podcast is the space where leadership and safety come together to shape better workplaces and better lives.