U.S. Markets closed
  • S&P 500

    3,298.46
    +51.87 (+1.60%)
     
  • Nasdaq

    10,913.56
    +241.26 (+2.26%)
     
  • Russell 2000

    1,474.91
    +23.09 (+1.59%)
     
  • Crude Oil

    40.04
    -0.27 (-0.67%)
     
  • Gold

    1,864.30
    -12.60 (-0.67%)
     
  • Silver

    22.99
    -0.21 (-0.91%)
     
  • EUR/USD

    1.1639
    -0.0037 (-0.3142%)
     
  • 10-Yr Bond

    0.6590
    -0.0070 (-1.05%)
     
  • Vix

    26.38
    -2.13 (-7.47%)
     
  • GBP/USD

    1.2744
    -0.0007 (-0.0586%)
     
  • BTC-USD

    10,769.81
    +72.81 (+0.68%)
     
  • CMC Crypto 200

    230.19
    +12.36 (+5.67%)
     
  • FTSE 100

    5,842.67
    +19.89 (+0.34%)
     
  • Nikkei 225

    23,204.62
    +116.80 (+0.51%)
     

The Single Most Groundbreaking Disruption

Special To ETF Trends

This article was originally published on ETFTrends.com.

By Stephen McBriden

America’s top researchers were stumped. How do you teach a computer to “see?”

By 2012, technology had advanced a great deal. For $199 you could buy a tiny supercomputer called the iPhone 5. You could talk to it—and it would talk back. It could hail you a cab, give you driving directions, or play a movie. You could even video call someone living on the other side of the world for practically free.

In fact, the iPhone 5 was more powerful than the NASA computer that sent Neil Armstrong to the moon. But computers had one big glaring weakness: they were still laughably bad at recognizing images. Take a look at these pictures:

[caption id="attachment_383261" align="aligncenter" width="630"] Source: HuffPost.com[/caption]

A toddler could tell you half of them show a chihuahua dog, and the other half show a blueberry muffin. In 2011, researchers showed them to the world’s best image-recognition computer and asked, “Is there a dog in this picture?”

The top performer got it wrong 30% of the time. Computers have always been able to count numbers and read text. But they could never master vision. In fact, about the only thing a computer could do with a pile of photos is sort them by size. If computers couldn’t tell a dog from a muffin, what hope was there for world-changing disruptions like self-driving cars and drones that need computers to see?

Doctors have long hoped to create an “all-seeing” computer that could save lives. According to IBM, 90% of medical data is image-based. In other words, doctors take X-rays to see broken bones, ultrasound scans to check a baby’s health inside a mother’s womb, and MRI scans to diagnose potentially cancerous tumors.

But human doctors often misjudge what’s in these images. Researchers at the Mayo Clinic revealed 88% of the time, diagnoses are revised upon a second evaluation. In other words, doctors get the prognosis wrong all the time. This comes on the heels of a 2015 revelation by the National Academy of Medicine that diagnostic errors cause up to 10% of all patient deaths in the US. Unfortunately, none of these world-changing disruptions are possible without computers that have a set of working eyes.

The Big Breakthrough

She didn’t know it at the time, but in 2010 Fei-Fei Li changed the world forever. Li was studying for a PhD in computer science at the University of Illinois. She was dumbfounded at how shockingly bad machines were at identifying objects.

In 2010 Li founded the “Olympics” of image recognition–ImageNet. In short, it was a challenge to her peers to build a machine that could “see.” ImageNet contained a giant database of over 14 million photos.

Teams had to teach computers to identify millions of jumbled photos of everything from cats to churches to traffic lights. The winning team in the first year had an error rate of 30%. Results weren’t much better in 2011. And then in 2012 Alex Krizhevsky, a University of Toronto student, made a big breakthrough. He created a computer program called “AlexNet” that mimicked how our brains recognize objects.

AlexNet rewired how computers “see” and swept away its ImageNet competitors, cutting the error rate in half. In short, AlexNet handed computers the gift of sight. This was the first time in history a machine could identify objects like a person. Research teams raced to build better versions of AlexNet. And in just seven years the winning team’s score jumped from 70% to 99%.

ImageNet was solved. In fact, the competition shut down in 2017 because it was no longer even a challenge. Pretty incredible when you consider that just seven years prior an error rate under 25% was considered the “holy grail.”

Here’s the thing: teaching a computer to identify a dog is just the first baby step in this disruptive megatrend. Once you teach a machine to see dogs, you can teach it to see pedestrians and cyclists—which is crucial for self-driving cars.

And asking a computer, “Can you see a dog in this picture?” becomes the same as, “Can you see a broken bone in this X-ray?”

A New Superpower

The ImageNet breakthrough was the catalyst for the computer vision boom happening all around us. As you read this letter, fully robotic self-driving cars are cruising around the suburbs of Phoenix, Arizona. These robocars are operated by Google’s (GOOG) self-driving car arm, Waymo. In fact, they’ve already driven over 20 million miles on American roads.

Waymo taught its cars to “see” by using computer vision programs that learn from every mile driven. Waymo’s fleet of robo-taxis have gotten so good, it now runs a fully-driverless ride-sharing service. Here’s a picture of one zipping around the streets of Phoenix:

[caption id="attachment_383262" align="aligncenter" width="630"] Source: Cnet.com[/caption]

The 180-year-old farming equipment maker John Deere (DE) is the most unlikely disruptor in the world. But it’s using computer vision to dramatically slash chemical use on crops.

Farmers used to have to decide whether to dose crops with chemicals on a field-by-field basis. But Deere’s See and Spray system can distinguish between healthy and unhealthy crops. It allows for targeted bursts of chemicals to be directed at individual plants—slashing herbicide use by up to 90%.

Computer vision is also pioneering mind-blowing breakthroughs in medicine. In 2018 researchers from the US and Europe taught a computer to diagnose skin cancer more accurately than leading doctors. In short, they showed the computer more than 100,000 images of potentially cancerous skin lesions. The machine had an error rate of just 5%, versus 14% for the team of 58 dermatologists.

A team of scientists at Stanford replicated these results. They fed 14,000 images of skin lesions into a computer and asked it to diagnose them. Two board-certified dermatologists were asked the same. The team concluded: “In every test, the network outperformed expert dermatologists.”

Computer vision start-up Gauss Surgical has created a device allowing hospitals to measure blood loss during surgeries. Right now, doctors rely on their own judgement of blood loss by eyeballing the surgical field. As you can imagine, this is highly inaccurate, and has led to several deaths.

Surgeons simply hold a blood-stained sponge in front of Gauss’s iPad-like device. Its computer vision program then determines the volume of blood lost based on the state of the sponge.

In short, computers have finally learned to see. And this new superpower is going to unlock a new world of disruption.

The Great Disruptors: 3 Breakthrough Stocks Set to Double Your Money"

Get my latest report where I reveal my three favorite stocks that will hand you 100% gains as they disrupt whole industries. Get your free copy here.

POPULAR ARTICLES AND RESOURCES FROM ETFTRENDS.COM

READ MORE AT ETFTRENDS.COM >