The humble CAPTCHA is a good way to keep bots honest.
CAPTCHA — created in 1997 — is a contrived acronym for “Completely Automated Public Turing test to tell Computers and Humans Apart. “It’s a type of challenge/response test used in computing to determine whether or not the user is human,” says Wikipedia.
In the past, they were simple: a string of letters and numbers in different sizes and fonts. Nowadays, they often present as a grid of nine thumbnails, with instructions to check off all images of cars, bicycles, or crosswalks.
No one is overly fond of this task — we squint at our laptop screens, trying to figure out if a grainy image contains a traffic light or not. We complete the task, only to face another screen of thumbnails.
What we’re doing is essential, though. We’re helping train an AI.
Computers are run by algorithms, while humans have real-world training. When we scan a street for motorcycles or crosswalks, we’re not thinking of sun interference or how much of the image our retinas capture. Humans have pattern-recognition skills no computer can match. We see a splotch of white paint on asphalt and decode it: crosswalk/no crosswalk.
The CAPTCHAs collect our decisions to build algorithms that will help pilot the automated vehicles of the future. Input from a single programmer won’t construct an accurate algorithm, but input from millions of users who’ve made thousands of real-world bicycle/no-bicycle decisions hopefully will. Automated vehicles need to operate on real-world roads, and they need massive databases to help them distinguish the objects they encounter and navigate them successfully.
“While AI systems can show human-level competence in low-level pattern recognition skills, they are merely imitating human intelligence on a cognitive level,” says Michael I. Jordan in an article published on CDOTrends. Jordan is a professor at the University of California, Berkeley.
That’s fine because human-level competence in low-level pattern recognition skills is what’s needed to solve a CAPTCHA. And that competence helps inform the algorithms that will pilot the trucks cruising the highways in the future.
Self-driving cars capture the imagination, but it will be some time before daily commutes can be executed by a computer-driven autonomous vehicle. The early adopters of this nascent tech are found in the trucking industry, where driverless vehicles have distinct advantages in carrying cargo along known fixed routes.
Think of an express train/local train system. The express carries commuters to a hub where they transfer to a slower local line to their precise destination. The same paradigm applies to elevators in a tall building, where an express lift takes people to a “sky lobby” with separate lifts for low/high floors or odd/even floors.
The same principle applies to the ecosystems of driverless trucks, which transport between major cities. While truly driverless vehicles aren’t on the road yet, there are already market movers in this space. California-based TuSimple “currently operates across the states of Arizona, New Mexico, and Texas,” says the firm on their website. “The company plans to expand service coast to coast by 2023 and nationwide to the lower 48 states in 2024.”
There’s a caveat: while this firm’s trucks are self-driving, there are two human operators in each vehicle. “Currently in this phase of development, TuSimple operates with a trained safety driver behind the wheel to monitor the system and a safety engineer in the passenger seat while operating autonomously,” said the firm in a statement. Other operators testing the self-driving waters, including Tesla, Uber, and Google’s Waymo, have in-cab human operators as well.
In August 2019, Reuters reported that UPS, the world’s largest package delivery company, bought a minority stake in TuSimple, and has been testing the startup’s autonomous trucks since May 2019 on a busy freight route in Arizona.
According to Reuters, the firm’s other investors include Chinese online media company Sina Corp. and graphics giant Nvidia Corp. And in April 2021, MarketWatch reported that TuSimple Holdings Inc “has set terms of its initial public offering, in which the self-driving trucks maker could be valued at up to USD 8.3 billion.”
Bumps in the road
There are potholes on the road to driverless vehicles. In November 2019, San Francisco-based startup Starsky Robotics was named as one of the CNBC Upstart 100: “an original list of the brightest, most intriguing, young startups promising to become the great companies of tomorrow.”
“As of June 16, Starsky began operating truly driverless semi-trucks on the Florida turnpike,” reported Rachel Premack in Business Insider. However, the vehicles did have a human driver, “It’s just that he or she is remotely controlling the vehicle, and several others, as many as 500 miles away.” The setup was likely similar to a pilotless drone with a video feed and controls for the operator.
Unfortunately, according to Wikipedia: “In November 2019, over 85% of staff were laid off after the company failed to find further investment, as concerns mounted over the financial stability of its freight-hauling arm. By March 2020, the company sold off the remaining assets, including patents relating to operating remote vehicles.” It’s a safe assumption that some of that patented technology will find its way into driverless vehicles of the future.
A humanity effort
Needless to say, the first road accident involving a driverless truck will be headline news, complete with references to Terminator movies. It will include some statistics about accidents caused by vehicles piloted by actual drivers, but the clickbait aroma will remain: “Robot truck crashes!”
Still, we do our part. Our human frustration swells as we peer at those hard-to-figure thumbnails, but our human intelligence helps add a few more bytes to the car/not-car algorithms. And that’s a good thing.
Stefan Hammond is a contributing editor at CDOTrends. He is an avid follower of cybersecurity issues, robotic, and macro tech trends. You can reach him at [email protected].
Image credit: iStockphoto/wildpixel