■ SAFE RISK ■ Science & Research
No — and the forecasting tools are making the role more important, not less, because someone still has to convert an ambiguous signal into an order that moves thousands of people.
“Predicting eruptions needs AI. Evacuating villages needs human authority and urgency.”
Our AI replacement risk score — how we score jobs
During quiet periods the work is monitoring network maintenance, model development and community preparedness: keeping seismometers, tiltmeters, GPS stations and gas sensors alive on hostile terrain, processing satellite InSAR and thermal data, mapping past deposits to model likely flow paths, and running drills with local authorities. During unrest it becomes a duty rota of round-the-clock signal interpretation, alert-level decisions, briefings to civil protection, and press conferences where every word is parsed by people deciding whether to leave their homes.
Automation has transformed detection. Machine learning classifiers now pick and label volcanic earthquakes far faster than analysts, satellite systems flag ground deformation and thermal anomalies globally without anyone requesting a scene, gas-flux instruments stream continuously, and probabilistic hazard models simulate lahar and pyroclastic flow inundation in minutes. Much of the routine analytical labour of the last generation has been absorbed.
The hard part was never the data. Volcanic unrest frequently fails to become an eruption, so a scientist has to decide, under uncertainty, whether to recommend evacuating a valley — knowing that a false alarm burns credibility needed for the next one, and a missed call kills people. That decision is legal, political and communicative as much as scientific: negotiating with mayors, handling communities who refuse to leave livestock or ancestral land, and holding public trust in a language and culture the model knows nothing about. Field sampling on an active crater rim also stubbornly requires a person willing to go there. This is a small profession, but a durable one.
Automatability: our editorial assessment of current and near-term AI capability
Automated detection and satellite monitoring are already standard and will keep improving through this decade, removing analyst hours rather than posts. The decision, communication and civil-protection role remains firmly human beyond 2040 — an alert level is an act of authority with legal consequences. If anything, expanding global monitoring coverage increases demand for people who can interpret and act on it.
It can detect and classify precursors far better and faster than manual analysis — earthquake swarms, ground deformation, gas and thermal anomalies — and it improves probabilistic forecasts. It cannot deliver certainty, because many unrest episodes never erupt and each volcano behaves differently with limited historical data. The forecast is a probability; deciding what to do about it is a human judgement with lives attached.
Secure in nature, small in size. Posts are concentrated in national observatories, geological surveys and universities, and funding rather than automation limits the numbers. The skills in demand are shifting toward computational monitoring, probabilistic hazard assessment and crisis communication. Our risk score of 16 reflects genuine resilience: the decision-making core of the role is not delegable to software.
Runs continuous interpretation of monitoring streams on shift, convenes with colleagues to recommend alert-level changes, briefs civil protection and government officials, and faces the public and press. Alongside that come field measurements at the volcano, hazard-map updates as conditions change, and negotiation with communities reluctant to evacuate. The science feeds the decisions; the decisions are the job.
They redistribute the work. Automated picking, satellite tasking and streaming instruments cut the routine analysis that once consumed most of an observatory's staff time, but they also extend monitoring to volcanoes that previously had none. That produces more alerts needing interpretation and more communities needing preparedness work — both of which require people, not more sensors.