By Javier Surasky-
The recent
Summit of the Future has made me think a lot about the concept of "future."
The future is a rearrangement of present opportunities.
Viewed this way, the future is inscribed in the now and is not its continuity
but rather one of its multiple possible realizations.
With this
vision in mind, a recent conversation between Marc Benioff, CEO and co-founder
of Salesforce, and Jensen Huang, Founder and CEO of NVIDIA, caught my
attention. The meeting took place during the "Dream
Force" event, organized in September 2024 by Salesforce.
In this
public conversation, Huang recalled that his company's first product was
Gforce, which made it possible for the Compute Unified Device Architecture
(CUDA) to be born, which in turn helped the emergence of AI, making Agentforce
possible. The last step of that chain was later explained, almost
inadvertently, when Huang said that in his company, they were "excited
about the possibility of using deep learning in so many different areas, from
computer vision to speech and understanding, etc. When that happens, the
question is: What new opportunities will open up for the computer industry? And
for the first time, our industry is going to be a skills industry. We captured
that idea and talked about 'agents.' But, for the first time, we'll have
(digital) agents using (digital) tools."
For Huang,
this is an extraordinary historical moment where technological progress has
surpassed Moore's Law and squared it. Moore's Law describes that
computational density doubles every 18-24 months, but this is already extremely
slow if we look at the changes and their speed over the last decade.
On one
hand, between 1959 and 2010, global computing power doubled on average every 17
to 29 months. If that was already impressive, from 2010 onwards, the doubling
of total computing power began to occur in periods of four to nine months.
This growth
implies an increase in the number of chips and is accompanied by growth in
their performance while computing capacity costs fall rapidly. As early as
2005, Ray Kurzweil pointed out in his book "The Singularity Is Near"
that computers would reach the level of human intelligence by 2029,
something he has just reaffirmed in the recent sequel to that work, "The
Singularity Is Nearer," published in 2024. The value of computing power
for this, he tells us, will be around USD 1,000 by 2029.
According
to work
by Advait Madhavan, published by the U.S. National Institute of Standards
and Technology, the human brain operates with a capacity equivalent to an
exaflop (a 1 followed by 18 zeros) of mathematical operations per second. FLOP
is the acronym for floating-point operations per second, a mode of mathematical
notation.
Well, the
human brain's processing capacity has already been reached by Frontier, the
most powerful supercomputer that exists today. The difference between the human
brain and AI in computing capacity is currently sustained by unequal energy
consumption: the human brain requires 20 watts to perform the process compared
to 20 megawatts consumed by Frontier, which is a million times more consumption
for the computer.
There's
more: NVIDIA and Amazon Web Services (AWS) are working on the Ceiba project,
a cloud-based supercomputer that will be able to produce 414 exaflops, which
means that in one day, it could perform
the same number of operations as an average computer from 2010 would in 11
million years.
Given that
Kurzweil's prediction is aligned with reality, it's unsurprising that in the
chat with Benioff, NVIDIA's CEO stated that we are about three years away
from having robotic hardware that can match humans, and his company is
working on it. Again, the problem is primarily related to energy efficiency.
Everything
indicates, then, that the challenge of energy generation is above that of
developing computing capabilities, whether it's about having an AI with
human-like capacity or even a general AI or having general-purpose humanoid
robots.
A
question that will define the future is: can agents find the energy solutions
that remove the main obstacle to their evolution? If this happens, the future will be very
different: work can no longer be a central social articulator, technologies and
skills will measure wealth, and savings or leasing plans will not be oriented
towards buying cars.
The
environmental limits of the natural world leap onto the AI scene, whose forces
are already unleashed, and make us think that the real fear we should have
is not the irruption of "Terminator" but the senselessness of our own
greed when it comes up against planetary limits.
