The economy at machine speed
The economy is delicate, and made of many economic agents making different
choices with different information at different times. What happens when those
choices fall onto very intelligent agents that see the same signals and act at
machine speed? The agents don't have to make the same stock trade at the same
millisecond, or recommend the exact same product to its respective humans for
there to be large changes in the economy. Even a small rise in correlation
could cause demand to synchronize very quickly, create large new crowd
investment strategies, and overwhelm physical supply chains. Consumers could
funnel to few incumbents, destroying competition.
Markets, companies, and agents might eventually adapt. But what changes will
occur, and how suddenly? Will we have to get used to loops where people
discover, crowd, and abandon opportunities before moving on? As we automate
decisions, what transformations will our beloved bell curves undergo? Which
outcomes will become more concentrated or more volatile as a result?
Deciding under accelerating uncertainty
People today, largely as a result of new innovations (and, more
importantly, the speed at which they propagate), don't know what to
do. That uncertainty is itself becoming widespread. What can we do about
it? How do we begin to grasp at unknown unknowns? Will people change
careers because of AI, and on what basis, if nobody can verify the
predictions they're changing careers in response to? What happens once an
incorrect prediction becomes accepted as the most probable outcome
anyway? How do we minimize the downside and maximize the upside of
decisions made under this kind of uncertainty, and where might new
problems appear that we aren't yet considering? And if a plan for any of
this is actually formed, how does it get carried out?
Culture after machine intelligence
Cultures change as people select, imitate, and adapt inherited values and
behaviors. Generative models are now ubiquitous and introduce a new interface
for cultural adoption: models are trained on the cultural geist, and people
exposed to them generate new cultural artifacts that inform the model's next
mint. Models already have different cultural tendencies based on factors like
language and user location.
With successive iterations of the human-model transmission loop, what comes
of both groups' cultures? Do they converge toward the highest-probability modes,
or will repeated inheritance of certain behaviors augment them? Which dialects
and aesthetic standards are the most vulnerable to disappearing? Should we make
an effort to preserve certain types of tacit knowledge? Would intentional
preservation of cultural differences hamper their natural evolution? Is that a
bad thing?