On the cover of The New Yorker magazine on October 23, 2017, a young human sits on the streets of futuristic Manhattan and begs for money from a robot. It can be said to reflect human beings' fear of machines. Indeed, in almost every public occasion, I will be asked: Will machines replace human jobs in the future?
You can think about it first. Cars are so much faster than humans. Have you ever worried that your car will make you unemployed? If not, then there must be a deeper reason for human beings' fear other than "being more powerful than humans". In 1996, the IBM supercomputer Deep Blue defeated the world chess champion Garry Kasparov. Without mentioning this history again, everyone seems to have almost forgotten that humans had lost to computers so long ago. But why has the fear of AI continued to rise after AlphaGo defeated former world chess champion Lee Sedol in 2016?
Because no matter how fast a car is, we only regard it as a means of transportation. IBM's Deep Blue uses brute force to solve problems. At this time, computers are just tools that can handle large amounts of calculations. But when AlphaGo wins, it is associated with real "thinking ability," and "thinking ability" is often related to "self-awareness." "link. The fear is that after the tool becomes self-aware, it will become uncontrollable and start to imagine the development of various horrific plots.
The human brain is equivalent to 574 of the world's most powerful supercomputers
No matter how much we talk about the fact that "no machine can think" at present, it is really difficult to convince people in the modern era when everyone is immersed in talking to ChatGPT every day. Taking a step back, even if the machine has self-awareness, we have returned to a more A simple question, how much does the power of the human brain compare to the most powerful supercomputer in the world today?
There are public data on the computing power of supercomputers, but what about humans? According to "Whole Brain Emulation: A Roadmap" published in 2008 by the Future of Humanity Institute, Oxford University, this study has explored various assessment methods, among which electrophysiological model calculations are more reliable. The middle number, which tells us an astonishing number, is that the computing power of the human brain is 1022 FLOPS (see notes).
How big is this number? The author compared the computing power of the Apollo moon landing in 1960 with the world's number one supercomputer in the past 10 years. The world's number one supercomputer in November this year is El Capitan based on ARM chips. The force scale is 1.742 EFLOPS, which means that the human brain is equivalent to 574 of the world's most powerful supercomputers. Yes, you read that right, 574. Although the advantages of the human brain over machines are shrinking, this number should scare many people. If the object were replaced by NASA's supercomputer during the Apollo moon landing, this number would At that time, the "brain-computer computing power equivalent (the computing power of the human brain is equivalent to how many supercomputers)" will be even more amazing "2 million terabit" units
The fragility of the human body is also a source of power. AI consumes far more energy than the brain
From 1960 to the present, the equivalent of brain-computer computing power has shown an almost perfect logarithmic linear relationship with time. According to the trend, the time when machine computing power will exceed that of humans will be around 2036. But that's not all the strength of human beings. Although El Capitan uses the latest energy-saving technology, its energy consumption is still as high as 28,000 kilowatts. In comparison, our brain consumes only 0.02 kilowatts. That is to say, while the machine needs a dedicated power plant, we humans only need a few lunch boxes a day and moderate drinking water to operate. Another indicator is called "brain-computer computing efficiency equivalent", which is the energy consumption efficiency ratio when the brain generates the same amount of calculations as a supercomputer. This number is something to be proud of. The energy-saving efficiency of our human brain is El Capitan's 800 million times.
When talking about AI, we often hear "computers will soon surpass humans." Even in 2024, the computing power of El Capitan, the world's most powerful supercomputer, will only reach 1/574 of the human brain. The human brain, which took millions of years to be built by natural evolution, is far more sophisticated than current AI. AI is still a long way from truly reaching the capabilities of the human brain.
We always feel that the human body is too fragile and will get tired and irritated. It is far less stable and powerful than machines. However, because of the fragility of our human body, as long as energy consumption is not controlled well, we will often starve to death or die from fever. On the road of evolution to the present, we must work diligently to balance the development of intelligence and the energy consumption of the body at every moment, in order to create the miracle of ultra-low energy consumption of our brains. The fragility of our bodies creates the strength of our brains. That's why I never worry about whether humans will be replaced.
When the first photograph appeared in 1826, it also sparked panic in the painting community, fearing that photography, which could recreate the world better than painting could, would eliminate painting. But painting has not disappeared. Photography can only reproduce the world, but painting can create the world with various aspects. Only things that are truly worthless disappear from history. So instead of worrying, don't forget the key to humans becoming intelligent creatures: using tools. Using this powerful tool called AI is the right choice for us to continue to move more steadily on the road of evolution.
Note: FLOPS is the abbreviation of "Floating Point Operations Per Second", which is an indicator used to measure the computing power of a computer or processor. This unit is used to describe how many floating point operations a system can perform in one second. PFLOPS is 1015 FLOPS, and EFLOPS is 1018 FLOPS.
