Gergely Máté (with help from Claude) - May, 2025
Our premise is that the act of thinking will be dominated by AIs. Therefore many activities that are dominantly utilising thinking will become economically obsolete - AIs will do them for significantly less cost than humans. This is likely going to transform labor markets, with high uncertainty and potential large losses of income to millions of people. Analysing and categorizing jobs with this in mind can help mitigate the risks of this transition, and can help in a better career orientation.
The goal of this project is to provide anyone with a rational, easy to use evaluation framework to assess any job from the job safety point of view in the AI era. It provides a schema to dissect jobs based on required human skills, estimate how AI and robot development fares in those skills, add in job mastery, social and economic adoption factors, and get an easily interpretable “job safety score” as the result. Using this framework should be easy, and anyone interested should be able to get feedback about the job security level of a particular job with a pen and paper and a bit of thinking in about 5-10 minutes.
Any job can be decomposed based on these fundamental dimensions of required human capabilities:
Development of AI primarily touches cognitive and sensing capabilities: AI models excel at applying knowledge and pattern recognition (including visual and auditory patterns). Thinking models can reason efficiently, and AIs tend to get better and better at solving academic problems. Models are also capable of text, audio, image and video generation, and thus may be capable of solving communication tasks. However, AI driven robots acting in the physical world are currently at a very early stage. Although there are some robots capable of moving and manipulating objects with varying precision, these are still rare, mostly in prototype phase, and costly. AI is also struggling at dealing with unseen or rapidly changing environments, and also has hardware limitations: systems showing higher intelligence are run on energy demanding GPUs in large clusters of servers. Therefore jobs involving field work are way less vulnerable to AI compared to office work.
After we’ve decomposed the job under scrutiny, we score each dimension based on how resilient it is to current progress in AI capabilities. This scoring is by nature subjective, we have to make educated guesses. The scores should be related to the specific activities the job requires.
Then we can apply the weighted job vulnerability formula, where higher scores indicate greater resilience against AI disruption:
Job AI Resilience Score = Σ(Component Weight × Human Advantage Score)
Humans advance their art during their years in a profession. We distinguish among beginner, advanced and master level based on experience by applying -1, 0 and +1 modifiers. With this we try to capture the fact that jobs of more skilled, experienced professionals are less vulnerable to AI than jobs at the beginner level.
We use this score to classify jobs into risk timeline categories of immediate risk, medium term risk, long term safety and very low risk (probably indefinite safety).
Finally, we modify the result based on four important adoption factors: first, for companies there is a higher incentive to replace workers with higher wages to cut costs, second, for some activities replacing humans with machines can have non-economical benefits - for example reducing risk of harm -, third, some development (particularly robots) can be prohibitively costly, and fourth, there are jobs (especially where something is “performed” in a broad sense) where most humans inherently prefer to see humans.
Quickly estimate the percentage breakdown of:
For each component, take into account the job specific activities, and score how much advantage humans have:
Ask these questions:
If any of the answer to questions 1 or 2 is "yes," move the job up one risk level.
If any of the answer to questions 3 or 4 is "yes," move the job down one risk level.
1. Job Composition
2. Human Advantage Score
3. Job Resilience Score
4. Result: Low to Very Low Risk (10-15+ years)
5. Adoption Factors Check: Safety benefits beyond cost savings might increase risk level slightly, but construction work still remains in the low risk category.
1. Job Composition
2. Human Advantage Score
3. Job Resilience Score
4. Result:
5. Adoption Factors Check: High salaries justify AI replacement, potentially moving advanced developers toward High Risk, while master developers with stronger stakeholder relationships remain in Medium Risk.
1. Job Composition
2. Human Advantage Score
3. Job Resilience Score
4. Result:
5. Adoption Factors Check: High salaries might justify automation investment (risk up), but implementation costs for robotic dentistry would be extremely high, and many patients are likely to prefer humans (risk down). These factors cancel each other out, keeping dentistry in the Low to Very Low Risk category.
AI is likely to be an extremely transformative technology, and this transformative power can bring uncertainty and fear regarding job security. By evaluating jobs across five key dimensions and accounting for economic and social adoption factors we provided a framework to evaluate these risks in a structured, rational, systematic way. I hope this framework can help people overcome fears stemming from the unknown, and will help in career orientation during this transition. I encourage readers to apply this framework to their own careers or when advising others about education and skill development choices.
[1] Actually we can feed our framework into AI systems to help us with analysis.