We propose the following likely outcomes.
Industrialization of data pro-
duction
via large-scale automated experiments and observations, already occurring
in astronomy [
241
], functional genomics, and materials science [
159
], will expand to
many more domains. The resulting data glut will in turn drive
industrialization
of data analysis
, by which we mean large-scal e computational platforms that
automate quality control, analysis, inference, and other step s. These developments
will greatly reduce the costs of hypothesis generation and testing. They will also
improve reproducibility because experimental configurations and data processing
steps will be captured precisel y.
Meanwhile, the digital encoding of large quantities of scientific knowledge
from such experiments and other sources (e.g., the scientific literature) will enable
the creation of a
universal knowledge base
supporting both rapid access a nd
automated inference. It will become routine to ask questions via a scientific search
engine, to be notified of potential inconsistencies across existing knowledge, and to
vote on the next set of experiments to be performed by industrial-scale facilities.
Other experiments will be performed by quasi-independent
robot scientists
[
171
,
262
] that apply inference and experiment design methods to guide their choice of
the next experiment.
These steps towards economies of scale may sound dehumanizing, but experience
suggests that, if implemented in the right way, they can unleash a flood of creativity.
If the universal knowledge base is treated as a globally accessible public good, then
the scientific playing field becomes more level. A high school student in Angola,
India, or New Zealand will be able to search for new drugs for rare diseases or
new materials sourced from local materials. They will use powerful tools accessible
from this
discovery cloud
[
124
] to collect new data, analyze extant and new data,
test hypotheses, and contribute to knowledge.
This discovery cloud will also empower the bench scientist. Ama zon’ s Echo
and Alexa services can keep our calendar, order a pizza, and summon a car service,
and Alexa can invoke an Amazon Lambda function to start an experimental
analysis in the cloud. All of these actions can be driven by voice commands.
As machine learni ng continues to progress, future s cientists will benefit from a
cloud-based research assistant
that not only monitors experiments but also
performs background research, such as scanning the literature for related work
and checking our mathematical d erivations. Such a system will respond to vocal
instructions while also reading (and writing) our computational notebooks.
Some of these developments may b e some way out, but the techno logy is
evolving fast. As Roy Amara observed, “[w]e tend to overestimate the effect of a
technology in the short run and underestimate the effect in the long run.”
346