20 Metropolitan Policy Program at Brookings
task content will be—we can only rely on what we
know about today, and where we think technology
will go. New tasks and occupations will emerge,
as will new technologies and new industries,
and existing roles will adapt to evolving task
demands as human ingenuity and technology
alike advance.
25
So we do not show how many
and where new jobs will emerge. Our work is
signicantly limited in this regard. For example,
recall from before that during the 1980s, 1990s,
and 2000s half of employment growth came
from occupations that didn’t exist previously.
26
Similarly, we are unable to say what any particular
automation potential gure spells for future
employment or wages because of the complex set
of factors described before.
Second, as described earlier, there is a well-
established history of divergence between
technological possibilities and technological
adoption. We have no ability to assess how the
gap between “automation potential” (which our
gures are based on) and “actual adoption” (what
will actually occur) will evolve, so our estimates
should be seen as an upper bound—perhaps even
an extreme one. Our gures describe what is
possible—not what is likely. In that regard, one
might interpret our gures more as a degree
measure of workplace task change rather than as
a predictor of employment or wages per se.
Third, while we don’t specically address
the timeline of automation’s absorption, it is
important: The faster automation takes place,
the more disrupting it will be in the workplace
and the more difcult it will be for workers to
adapt. And yet, predicting these timelines is a
challenge. Instead, we’ll direct readers to the
McKinsey work. Their adoption estimates are both
responsibly variable and distant in the future.
They predict that technical automation potential
makes major strides by 2030, with full potential
being achieved as early as 2040 or as late as
2050. For adoption—that is, automation potential
after adjusting for technical, economic, and social
factors affecting the pace of uptake—things begin
to pick-up by 2045 with full adoption no earlier
than 2065. Even under their most aggressive
scenarios, then, it will take at least a few decades
for the economy to feel the impact of currently
emerging technologies. Said differently, we might
be thinking about preparedness for changing
workplace requirements in terms of generations
(not in terms of years)—something that is
encouraging from a policy standpoint, particularly
because young workers can be trained from
the beginning for roles that show the least
susceptibility to future automation and steered
away from those with the most.
Fourth, and related to the above, much of the
impression created by our analyses and others
depends on the cut-off used to characterize
“high” susceptibility to automation. Along these
lines, we deem a 70-percent share of automatable
task content in an occupation by 2030 or 2040
as “high” risk. However, while this threshold is
consistent with many other studies and aligns
with statistical breaks in the occupational data,
it bears noting that this is not a theoretically
grounded threshold. Rather, it is a mostly
arbitrary one. In fact, we performed a number
of statistical tests that indicate a more realistic
threshold for “high” could be between 75 percent
and 85 percent. However, for now, we’ll stick
with widely employed thresholds. More broadly,
it bears noting that assessing qualitative and
quantitative differences in the extent and pace
of task-level job change in occupations remains
an underdeveloped aspect of the preexisting
literature on automation.
27
Fifth, while a signicant portion of our analysis
focuses on how the impact of automation will
vary across geographies, a key input into our
analysis—the McKinsey automation potential
estimates for each occupation—are produced at
the national level. As a result, our estimates for
how automation might affect regions differently
depend purely on the current industrial and
occupational mix of these states, metropolitan
areas, and counties. Surely, some regions