Meta-Invention Test
Ask whether an invention automates a task or the inventor
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 5
- Confidence
- 91%
The Meta-Invention Test separates inventions that improve a task from inventions that may reproduce the general capacity to perform tasks. In earlier automation waves, a tool displaced some workers but left people to supervise it, move into adjacent roles, or invent new work. Yampolskiy argues that general AI changes the analogy because the new agent could also supervise, retrain, and perform jobs created after deployment. Apply the test by identifying whether the technology is merely a bounded instrument or a transferable worker, then ask whether proposed replacement careers are themselves within its reach. The result is a scenario classification, not a dated unemployment forecast: tool-like change supports conventional retraining plans, while worker-like change requires broader economic and meaning-of-work planning.
Origin
Extracted from The Diary of a CEO
Core principles
- 01Ordinary tools automate bounded tasks
- 02An agent can be applied to newly created tasks
- 03Historical automation analogies fail when the worker is automated
- 04New-job creation is not protection if the same agent can do the new job
How to run it
- 1
Identify the automated unit
State whether the technology replaces one action, one occupation, or the general ability to learn occupations.
Pro tip Describe capabilities rather than relying on product categories.
Watch out A computer interface does not automatically make a system a mere tool.
- 2
Classify tool or worker
Determine whether the system stops at its assigned function or can pursue goals and transfer across tasks.
Pro tip Test unfamiliar tasks, not only benchmark tasks.
Watch out Today's limited deployment does not establish tomorrow's capability boundary.
- 3
Generate replacement roles
List the supervision, maintenance, coordination, and creative roles people expect the technology to create.
Pro tip Include the popular retraining path, not only current jobs.
Watch out Do not assume a new job is human-only because it does not exist yet.
- 4
Reapply the automation test
Ask whether the same transferable agent could perform each proposed replacement role.
Pro tip Repeat the test after each new role is proposed.
Watch out One surviving role does not imply a labor market large enough for everyone.
- 5
Choose the planning horizon
Use retraining for tool-like displacement, but add income distribution and purpose planning for worker-like displacement.
Pro tip Separate technical capability from the slower pace of real deployment.
Watch out The test does not establish when widespread automation will occur.
In the wild
A logistics firm expects autonomous driving to displace drivers and proposes retraining them as fleet coordinators. The planning team then tests whether the same AI can schedule routes, monitor vehicles, and resolve routine exceptions. Because most of the proposed role is also automatable, it adds income-transition planning instead of presenting coordination as a durable universal fallback.
→ The plan distinguishes short-term redeployment from long-term job security.
Common mistakes
Reusing Industrial Revolution logic
The analogy breaks if the new technology can also perform the replacement jobs that automation creates.
Equating capability with deployment
A technically automatable job may persist because adoption, regulation, or preference slows deployment.
Is it for you?
Best for
It is best for workforce, policy, and strategy discussions about technologies that can learn and act across domains.
Not ideal for
It is not ideal for a narrow tool that performs one stable function and cannot transfer to new work.
From the transcript
“If you creating a meta invention, you inventing intelligence, you inventing a worker, an agent, then you can apply that agent to the new job.”
“All the inventions we previously had were kind of a tool for doing something.”
From the episode
Roman Yampolskiy: These Are The Only 5 Jobs That Will Remain In 2030 & Proof We're Living In a Simulation!