Google data center.
More information
Artificial intelligence conquers the last frontier: designing its own chips
The machines already beat us, now they also convince us
The prophets of our time like to augur "the singularity," the critical point at which machines will take over and become the true engine of development, much to the anguish of Silicon Valley's enlightened flesh and delicacy organisms. In this context, the term singularity comes from cosmology and is not content to mean anything unusual. The two most famous singularities are the origin of the universe and the center of black holes, where Einstein's equations unravel in quantum uncertainty. The uniqueness announced by the wizards is not a mere change in trend, but the birth of a new and unpredictable world. This is the issue.
A central issue of singularity theology is the moment when machines begin to design themselves, as Google's chief engineer, Ray Kurzweil, argues in his 2017 book
The Singularity Is Near; when humans transcend biology
. Watch out for Kurzweil, who in addition to being a computer scientist and artificial intelligence specialist is a musician, atheist, inventor, writer and businessman. These types of polyhedral profiles attract my interest like a magnet pulls iron, and the truth is that Kurzweil's career could not have been more successful. That he founded a Singularity University in Silicon Valley raises questions about his academic sanity, but the guy is a machine.
In any case, his starting idea is very simple: that technology does not advance linearly (like 2 3 4 5…), but exponentially (like 2 4 8 16…), and that is a curve that rises more and more quickly and that, in the end, it has no choice but to overcome our poor brain, which was designed by biological evolution for other purposes. Kurzweil estimates that that critical point where machines overtake us, the singularity, will come in 2045. Predicting the future is the ideal way to screw up, but Kurzweil is 72 years old now, and 25-year-old predictions must matter very much to him. little. Many female readers and their children, however, will be here to witness the
big bang
of technology, or the lack of it.
There are some very curious news, however. Azalia Mirhoseini, Anna Goldie and their colleagues from Google's research team in Mountain View, California, have published in
Nature
the first artificial intelligence system that designs chips better than human engineers. Like so many others, the system is based on
deep learning
, or deep learning, which has revolutionized this discipline in recent years. In less than six hours, the machine designs a chip that surpasses that of humans in energy efficiency, size reduction and operational capacity. The machine learns from its past experience to improve the chips it designs. “We believe”, conclude the Google scientists, “that
hardware
(physical support, as opposed to software, or
software
) designed by artificial intelligence will stimulate advances in artificial intelligence itself, creating a symbiotic relationship between these two fields ”.
This is not the singularity yet, but take a good step towards it.
The same that gave life 4,000 million years ago.