The threshold of AI Labour Replacement

The world is constantly in fear and excited anticipation for when AGI is declared and supposedly virtually all human labour could feasibly be replaced. What would be more useful for governments, forecasters, policymakers is to have meaningful metrics to track labour. I propose 3 core metrics:

LC-P and LC-F would provide useful warning at different levels of urgency as to what to anticipate in job loss/job transformation. Providing invaluable insight for governments, policymakers, educational institutions etc that are concerned with how we incentivize and build our labour force.

LC-D would indicate the actual replacement potential of AI over human labour, in the current cycle. A sustained high level of labour coverage in real deployed systems, would be a strong indicator that AI has a durable trajectory in replacing human work.

What these metric differs from many other job-replacement metrics is that the numerator and denominator are updated at every measurement period (this can be monthly, quarter, 6 months, 1 year etc). Tasks that have been automated away, i.e. humans don't do them anymore, no longer count as tasks that humans do, thus removing it from the denominator, likewise, tasks that have been replaced in the previous cycle don't count towards tasks being replaced this cycle. Thus "resetting" the metric if no new tasks are being automated.

The real question of labour replacement isn't how much can AI do in totality, but how much of what a human is currently doing, is being replaced by growing AI capability.

As an example, say if an accountant's job of tallying up transactions takes up 4 hours of an 8 hour day, the other 4 hours are spent on report-writing, meetings, calls. AI comes along and drives the tallying task down to zero, the likely outcome is the accountant now spends 8 hours doing the other tasks. However if the next cycle comes and AI absorbs another 4 hours of the accountant's tasks, then we have a recurring trend of AI catching up to replacing the role. Aggregate this to a representative sample of the labour market, and we have more insightful metric about labour replacement trends than a vague notion of AGI/ASI.

On the flip side, it may very well be possible that as technology develops, our economy grows, that humans find new job and task categories that AI isn't capable yet. This suddenly adds a huge number to the denominator of this equation.

When one of these metrics are high for a sustained period (e.g. 6-12 months) there can be declared a threshold state, similar in structure to recession declarations. A high sustained LC-P/LC-F may give a state of warning incoming mass labour replacement, sustained high LC-D would suggest substantial real labour replacement that isn't replaced by new human work directions.

These metrics can be captured by an independent or governmental organisation, likely an office within something like the US Bureau of Labor Statistics or the UK's Office of National Statistics. The approach to measuring Labour Coverage can be informed by methodologies underpinning CPI and GDP, involving audits, surveys, expert opinion. It is important that this captures work tasks/hours being replaced NOT whole jobs, this is a moving metric, if a task category can be automated away consistently, it no longer counts as new labour replacement.

These are periodically updated for every industry, a central idea is that whilst AI/automation can replace work, it's not a given that future work that humanity moves towards may just as easily or quickly get replaced. The flaw with the concept of AGI is that it assumes once it crosses that threshold, it will always stay in step or ahead of human labour innovations. It may be the case that's true, but it's far better to have the metrics to show for it.