increasing, while nuclear, though long a
reliable source of carbon-free electricity, is
not. Meanwhile, a number of startups are
promising cheap, safe, proliferation-resistant nuclear energy in the next decade
(see “Fail-Safe Nuclear Power,” page 38).
Can these startups fulfill their promises? Outside of China, nuclear power is
expanding nowhere. China has 21 new
reactors under construction; Russia has
nine, India six. The U.S. is bringing five
new plants online, but since 2012, five
other reactors have been retired, with seven
more to be shuttered by 2019. California’s
Diablo Canyon plant recently announced
it will close by 2025. With other plants
closing in Japan, Germany, and the U.K.,
more reactors may be decommissioned
than built in the near future.
So why is this happening? Because it’s
expensive and time-consuming to design
and build a new nuclear plant, and there
are cheaper, easier alternatives.
The U.S. Nuclear Regulatory Commission has been waiting since 2014 for applications for design certification licenses
for small modular reactors—smaller versions of the large and extra-large operating light-water reactors, with additional
safety features. Such plants, which could
be factory-built and snapped together on
site, hold the promise of providing cheaper
nuclear power in a more distributed fashion. Other designs are on the horizon,
including molten-salt reactors, which are
promising but won’t be ready for decades.
In 2015, the General Accountability
Office reported that it takes 20 to 25 years
to develop a new reactor in the United
States— 10 years for the design phase, 3. 5
years for a design certification license from
the NRC, four years for a combined operating license, and another four years for construction. And that’s only in an ideal world
where no unexpected problems occur.
The GAO also found that it’s not cheap
to bring a design to fruition: just to reach
the design certification point costs some-
where between $1 billion and $2 billion,
and only about $75 million of that is NRC
fees. There’s a reason it takes so long and
costs so much: manufacturers need to con-
firm that the design is safe and secure.
Some people blame the regulators for
holding up the plants. Yet the NRC hasn’t
been presented with any applications for
new reactors and probably won’t be for
years. Data from prototype plants would be
helpful, but then many of the “new” designs
are not so new at all. Sodium-cooled fast
reactors have been built by countries
including the U.S., Japan, Russia, Germany, France, and India since the 1950s,
but no country has been able to make a
plant cheap and reliable enough to even
come close to being a viable energy source.
Yes, new nuclear technology can provide carbon-free electricity. But it has to
do more than that. It has to be safe, secure,
and resistant to proliferation. It has to
compete in the marketplace. New nuclear
designs are promising, but they’re no short-term solution to the climate problem.
Allison Macfarlane was the chairman of
the Nuclear Regulatory Commission from
2012 to 2014.
AI’s Research Rut
When we think of AI as one particular thing,
we drag the whole field down.
When you picture AI, what do you see? A
humanoid robot? When you think about
a real-world application of AI, what comes
to mind? Probably autonomous driving.
When you think about the technical details
of AI, what approach do you name? I’m
willing to bet it’s deep learning.
In reality AI comes in many shapes
and forms. AI machines go far beyond
humanoid robots; they range from soft-
ware detecting bullying on social media
to wearable devices monitoring personal
health risk factors to robotic arms learn-
ing to feed paralyzed people to autono-
mous robots exploring other planets. The
potential applications of AI are limitless:
personalized education, elderly assistance,
wildlife behavior analysis, medical-record
mining, and much more.
Our failure to appreciate this spectrum
threatens to hold back the field. When
we collectively picture AI as one type of
thing—whether it’s humanoid robots
or self-driving cars or deep learning—
we’re encouraging the next generation
of researchers to be excited exclusively
about those narrow things. If students
are presented with a homogeneous pool
of AI research role models, then it’s a self-fulfilling prophecy that only students who
“fit in” will remain in the field.
Since AI has enticingly broad possible applications, we need people with
a comparably broad set of experiences
and worldviews working on AI problems. Wouldn’t research teams working
on AI medical applications benefit from
researchers trained in biology? Wouldn’t
teams working on AI hunger relief benefit
from researchers with firsthand experience
in poor countries? Wouldn’t teams working on AI assistive devices benefit from
researchers with physical disabilities?
Today there’s a lot of fascinating work
going on in AI (see “AI’s Unspoken Problem,” page 28), but we’re also kind of in a
rut. We’ve tended to breed the same style
of researchers over and over again—people
who come from similar backgrounds, have
similar interests, read the same books as
kids, learn from the same thought leaders, and ultimately do the same kinds of
research. Given that AI is such an all-encompassing field, and a giant part of our
future, we can’t afford to do that anymore.
Olga Russakovsky is a postdoctoral
research fellow at the Robotics Institute of
Carnegie Mellon University.