The methodology, the data sources, the uncertainty, and the debate engine — all answered here.
The Death Score (0–100) is derived from six dimensions: (1) Task decomposability into automatable sub-tasks, (2) Physical embodiment requirements, (3) Social trust and relational dependency, (4) Regulatory and legal barriers to AI entry, (5) Current AI deployment data in the sector, and (6) Economic tipping point analysis — when AI becomes cheaper than human labour at scale. Each dimension is weighted and scored based on published research, industry deployment reports, and economic data. The score reflects displacement probability over a 20-year window.
Partly, yes — and this site is honest about that. All projections about technological displacement carry uncertainty. What distinguishes this analysis from speculation is that it is grounded in current deployment data — AI systems already doing portions of these jobs — and structural analysis of what specific human capabilities create genuine barriers. We do not claim precision. We claim a reasoned, evidence-based direction with honest uncertainty ranges.
AI displacement follows economic tipping points (when AI becomes cheaper than local human labour costs), infrastructure availability (internet, cloud computing, digital systems), regulatory frameworks (which vary enormously), and cultural adoption rates. A truck driver in Germany and a truck driver in rural Myanmar are in fundamentally different economic and infrastructure situations. The technology that displaces one may take 15 additional years to reach the other — if it does at all.
The score adjusts slightly in response to the strength and category of your argument — up to a maximum shift of ±15 points through extended debate. This is not live AI. It reflects the genuine uncertainty range in our assessment: some arguments — particularly about regulation, physical embodiment, and relational requirements — represent real factors that could be weighted differently. Strong arguments across multiple rounds shift the score within its uncertainty range.
Each job page has a debate interface where you can put forward arguments for why the job will or won't survive AI displacement. The system responds with counterarguments drawn from our research, identifies the category of your argument, and shows how your argument affects the assessed probability. It is not a live AI — it is a structured argument workflow designed to stress-test the analysis and expose where genuine uncertainty exists.
WhatJobsDie.com is an independent analysis platform examining the structural impact of AI on employment. It was built to cut through both AI hype (everything will be automated tomorrow) and AI denial (skilled humans will always be irreplaceable). The reality is more specific and more interesting than either extreme. Some jobs are already gone. Some jobs are permanently safe. Most are somewhere complicated in between.
Future additions include: Architect, Journalist, Financial Analyst, Surgeon, Software Engineer (the great irony), Accountant, Soldier, Priest/Religious Leader, and Farmer. Each requires structured analysis — specific task decomposition, regional data, and genuine argument for and against.
Every major wave of automation has created new job categories that previously did not exist. AI prompt engineering, AI safety research, AI audit and governance, robotic systems maintenance, and human-AI interface design are all growing fields. The honest position is that aggregate employment may remain stable while specific job categories disappear. The problem is transition: the new jobs require different skills, appear in different locations, and often pay differently.
We use ILO (International Labour Organisation), BLS (US Bureau of Labor Statistics), ONS (UK Office for National Statistics), and sector-specific data sources. Global figures involve estimation and should be understood as order-of-magnitude rather than precise census counts. Methodology and sources are available on request via the contact page.
The design is intentional. This is a site about artificial intelligence, and its visual language reflects that — machine aesthetics, data visualisation, terminal interfaces. The design is also meant to be uncomfortable in places. Displacement is not a comfortable topic. The aesthetic does not let you forget what you are looking at.
// STILL HAVE QUESTIONS?
If your question isn't answered here, or if you have data or research that challenges our analysis, we want to hear it. The work gets better through challenge.