Nvidia’s Chips Have Powered Nearly Every Major AI Breakthrough

 “Within 20 years, machines will be capable of doing anything man can do.”

Take a stab at when this quote is from. It wasn’t this year,  2010, or even during the ‘90s tech boom. It’s from one of America’s top computer scientists: in 1960.

You’ve surely heard about Artificial Intelligence (AI) before. “AI” often conjures up images of intelligent robots taking over the world. You’ll often read that it’s only a matter of time before AI steals all our jobs.

But the idea of humanoid machines is nothing new. It began with the “heartless” Tin Man from The Wizard of Oz. By the 1950s, a generation of scientists and engineers were convinced we’d soon co-exist with clever robots.

The term artificial intelligence was coined in 1954 at Dartmouth during the world’s first AI conference. Attendee Marvin Minsky, who later founded MIT’s AI lab, said “In 3–8 years, we will have a machine with the intelligence of a human.”

A couple of years later, Stanford created its AI project “with the goal of building a fully intelligent machine in a decade.”

This idea gripped Hollywood, too. Ever watch the sci-fi classic 2001: A Space Odyssey? The 1968 movie is best remembered for the intelligent supercomputer, HAL 9000. HAL could think just like a human and had the ability to scheme against anyone who threatened its survival.

Soon novels like I, Robot packed our bookshelves. We got stories of robots gone mad, mind-reading robots, robots with a sense of humor, and robots that secretly run the world.

Even the US military was convinced, so it pumped billions of dollars into AI research. In the ‘50s, we imagined bionic men would soon be running factories. Within a decade, cyborgs would be doing our housework. We were promised a new breed of machines.

70 years later, what did we get? Dishwashers, air conditioners, and microwaves!

Q3 2020 hedge fund letters, conferences and more

How Do Robots Learn?

Despite many lofty predictions and billions of dollars in funding, we never got machines with human-like intelligence. You have to dig into how machines learn to see why the idea was a flop from the get-go.

“AI” is a term that’s shrouded in a weird mix of hype and complexity. But the core idea of artificial intelligence is a machine that learns and thinks just like you or I. Most importantly, it learns all by itself, without human intervention.

Of course learning doesn’t come naturally to robots. To overcome this challenge, scientists created neural networks in the late 1950s. In short, neural networks are computer programs that mimic how the human brain works. They are made of thousands—sometimes millions—of artificial “brain cells” that learn through analyzing examples.

Say you’re creating a machine that can recognize cats. First, you’ll feed tons of cat pictures into the neural network. After analyzing, say, 1,000 examples, it starts to learn what a cat looks like. Then you can show it a real cat it’s never seen before, and it will know what it is.

Scientists who believed neural networks would breed intelligent computers were right on the money. Problem was… they lacked the raw materials needed to fuel their ambitions.

Remember, machines learn through analyzing examples, or data. And it turns out you need to feed them with truly enormous amounts of data to kindle any kind of intelligence. So machines need to see hundreds of thousands, if not millions, of cat pictures before they “learn” what a cat looks like. But in the ‘60s and ‘70s, we didn’t have that much data. The internet wasn’t invented, so we had almost no digital text or images. Books, photo libraries, and documents were still in the physical world, which meant converting them into digital files was inefficient and expensive.

And get this: the lack of data wasn’t even the greatest hurdle to building intelligent computers. Designing computer programs that mimic the human brain was genius. The drawback was neural networks needed hyper-fast computers to function.

And by 1995, even supercomputers were shockingly slow. For example, it took a giant “render farm” of 117 Sun Microsystems running 24/7 to produce the original Toy Story. The machines worked non-stop for seven weeks to produce the 78-minute film.

A Match Made in Heaven

After 40 years in the wilderness, two huge breakthroughs are fueling an AI renaissance.

The internet handed us a near unlimited amount of data. A recent IBM paper found 90% of the world’s data has been created in just the last two years. From the 290+ billion photos shared on Facebook, to millions of e-books, billions of online articles and images, we now have endless fodder for neural networks.

ips contain billions of “brain cells” called transistors. The more transistors on a chip, the faster it is.

Your phone is more powerful than the render farm that produced Toy Story. The 117 Sun Microsystems had 1 billion transistors, combined. There are 8.7 billion packed onto the chip inside the latest iPhone!

And in the past decade, a special type of computer chip emerged as the perfect fit for neural networks.

Do you remember the blocky graphics on video games like Mario and Sonic from the ‘90s? If you have kids who are gamers, you’ll know graphics have gotten far more realistic since then. Here’s each Lara Croft from the Tomb Raider series since 1996:

Source: Epic Games

This incredible jump is due to chips called graphics processing units (GPUs). GPUs can perform thousands of calculations all at once, which helps create these movie-like graphics. That’s different from how traditional chips work, which calculate one by one.

Around 2006, Stanford researchers discovered GPUs “parallel processing” abilities were perfect for AI training. For example, do you remember Google’s Brain project? The machine taught itself to recognize cats and people by watching YouTube videos. It was powered by one of Google’s giant data centers, running on 2,000 traditional computer chips. In fact, the project cost a hefty $5 billion.