National

Axios What's Next

Assisted-driving systems don't perform well enough to consistently avoid collisions, new testing by AAA shows, underscoring why most drivers are skeptical about autonomous vehicles.  

Why it matters: Although you still can't buy a self-driving car, a growing number of new cars and trucks are equipped with Automated Driver Assistance Systems, or ADAS technology, that helps cars steer, stay in their lane and brake in an emergency. 

Driving the news: In its third round of ADAS testing, AAA measured how three vehicles equipped with driver-assist technology — a 2020 Tesla Model 3, a 2021 Subaru Forester and a 2021 Hyundai Santa Fe — performed in four collision scenarios using dummy cars and cyclists made of foam.

What they found: When faced with an oncoming dummy car at 25 miles per hour, the Hyundai and Subaru didn't slow down, while the Tesla did slow to about 3 miles per hour. But all three collided with the dummy vehicle, AAA said. 

The Subaru failed to detect a simulated bicycle rider crossing its path in five test runs, while the Tesla and Hyundai successfully stopped in time to avoid a collision. 

All three tested vehicles avoided hitting slower-moving dummy vehicles and cyclists when approaching from behind. 

The other side: Hyundai and Subaru said they are looking into AAA's findings, and Tesla did not respond to a request for comment.

The big picture: AAA has been tracking consumer sentiment toward vehicle automation since 2016, and its latest survey found that distrust continues to run high. 

85% of those surveyed said they were fearful or unsure of self-driving technology; a similar percentage said they would not be comfortable using a self-driving vehicle to carry their children or loved ones.

The survey found drivers want automakers to improve the performance of existing driver support features rather than develop self-driving cars.     

What they're saying: "You can’t sell consumers on the future if they don’t trust the present," said Greg Brannon, director of AAA’s automotive engineering. 

"Drivers tell us they expect their current driving assistance technology to perform safely all the time. But unfortunately, our testing demonstrates spotty performance is the norm rather than the exception."

Assisted-driving systems aren't collision-proof, AAA testing shows

Consumers want carmakers to focus on making today's tech better before unleashing self-driving cars

Screenshot from a crash video showing a dummy cyclist being hit by a car

Autonomous vehicles

Automation and AI

From the Valley to D.C., Big Tech players like Facebook, Google and Amazon are under more scrutiny than ever as new technology develops and privacy and antitrust concerns grow in lockstep with companies’ ambitions.

Technology

Self-driving cars are now tooling around in a growing number of cities — even giving rides to the public, in some instances — but it'll be years before summoning a robotaxi is routine. 

One big obstacle: Teaching them the rules of the road in each city, which must be done block by block and street by street.

Why it matters: Despite high-profile efforts to get autonomous vehicles on the road, the obstacles — including engineering challenges, capital requirements, regulatory issues and liability questions — are immense, raising questions about how quickly the market can move forward.

Driving the news: Argo AI announced Tuesday that it has begun driverless testing operations in Miami and Austin, two of the eight cities in the U.S. and Germany where it is developing its AV technology.

It's an important milestone on the path to commercial deployment for the Ford- and VW-backed company.

Argo is unique for testing cars simultaneously in multiple cities; most of its rivals are focused on mastering one city at a time.

Founder and CEO Bryan Salesky says the approach will enable Argo to gather richer data so it can quickly scale AVs into a viable business.

Where it stands: So far, only one company — Waymo — has a commercial robotaxi service that's up and running. 

In suburban Phoenix, you can hail a driverless taxi using Waymo's app. But the service is limited to a 50-mile area around Chandler and Tempe — you can't go downtown or to the airport, for example. 

In San Francisco, GM-backed Cruise is giving free rides to the public, but only in certain neighborhoods during overnight hours. It hopes to receive the first permit in California for a paid robotaxi service as early as June 2.

In Las Vegas, another AV developer, Motional, has been running its cars on the Lyft network (with safety drivers behind the wheel) for several years. It plans to launch a fully driverless service in Las Vegas in 2023, with other cities to follow. 

Meanwhile, other self-driving tech companies — like Nuro and Aurora — are focused on local delivery or long-haul trucking applications. 

Details: Argo's driverless vehicles have operating limitations of their own: 

Only employees can ride for now, and the cars are confined to specific streets and neighborhoods. 

If they encounter heavy rain, they'll pull over and stop. The software doesn't allow them to operate in severe weather like snow, dense fog or hail. 

Between the lines: AV developers put their own parameters on where and when their cars can operate safely, then they gradually expand those limits — geography, lighting conditions and weather, for example — as the technology improves and regulations allow.

Each city is different, with different road layouts and unique local driving behaviors, explains Sam Abuelsamid, principal analyst at Guidehouse Insights, a research and consulting firm.

People can figure out how to drive in a new city pretty quickly, he says, but training AV software to understand the local rules takes time.

"Humans are naturally more adaptable than software is," he says. "It's just a reality. We do a better job of adapting to changing conditions than software does."

The bottom line: A driverless car is a lot like a Roomba vacuum: With regular use, it learns to navigate your entire living room. But to find the dust bunnies in your bedroom, it needs to be trained all over again.

Image of a self-driving Argo vehicle with no one in the driver's seat.

Argo AI's driverless test vehicles are now operating in Austin, Texas (above) and Miami. Photo: Argo AI

Why driverless cars can't go everywhere yet

Like a Roomba vacuum navigating your home, they need to be trained for every city

Economy & Business

The police in San Francisco see camera-laden autonomous vehicles as potential witnesses in their criminal investigations, setting off alarm bells for privacy advocates, VICE reports. 

Why it matters: As Axios has reported, self-driving cars capture and store huge databases of images so that they can train their algorithms and become better drivers. What that means is that bystanders are often captured in the footage, raising privacy concerns.

Driving the news: The police use of AV footage as evidence came to light because of a public records request from VICE's technology website, Motherboard.

A three-page San Francisco Police Department training guide provided to the website explained how officers should interact with AVs that don't have a human driver, mentioning Waymo and Cruise, two of the industry leaders.

Some of the guidelines include: "Do not open the vehicle for non-emergency issues" and "Do not pull vehicles over unless a legitimate law enforcement action exists."

A section titled "Investigations" advised officers of the AVs' usefulness in collecting footage, saying that this "potential evidence" had been accessed several times already. 

"Autonomous vehicles are recording their surroundings continuously and have the potential to help with investigative leads."

Privacy advocates warn that law enforcement agencies are extending their "web of surveillance" using information like Ring doorbell footage, automated license plate readers and location data they purchase from private companies. 

"​​As companies continue to make public roadways their testing grounds for these vehicles, everyone should understand them for what they are — rolling surveillance devices that expand existing widespread spying technologies," Chris Gilliard, visiting research fellow at Harvard Kennedy School's Shorenstein Center, told VICE. 

The other side: "The SFPD can unequivocally state that we do not use autonomous vehicles to surreptitiously surveil people in the City of San Francisco," an agency spokesman told Axios by email.

He would not say how many times the department had requested or accessed AV video for investigations. 

What they're saying: "Waymo captures data that's relevant for training our tech, not to identify individuals," a Waymo spokesman said.

"Our policy is to challenge, limit or reject requests that do not have a valid legal basis or that are overly broad."

A Cruise spokesman said: "We work closely with law enforcement on our common goal of making our roads safer. We share footage and other information when we are served with a valid warrant or subpoena, and we may voluntarily share information if public safety is at risk."

Self-driving cars could be potential crime witnesses

Cameras that gather video footage to train algorithms might also capture crimes as they occur

Illustration of a car as seen through a series of TV screens

As voice technology in cars has improved, it has opened the door for apps like what3words, a voice-navigation tool that helps drivers get precisely where they're going.

What's happening: Nearly a dozen manufacturers have integrated the London-based company's technology into their vehicles, including Mercedes-Benz, Ford and Subaru.

By dividing the world into a grid of 10-foot by 10-foot squares, each designated by a unique combination of three random words, the technology helps drivers navigate to places where an address isn't enough.

Why it matters: The app could enable rideshare drivers to find their customers more easily, for example, or delivery drivers to find the right door to an apartment building.

Drones could also use what3words to make sure packages are delivered to the exact location in a backyard. 

Between the lines: Some places, like stadiums or industrial parks, have a  single address but multiple entrances and parking garages. 

Some pop-up venues may have no address at all.

Some places have no street address, such as beaches, national parks or hiking trails.

How it works: To pinpoint your location, you can share your three random words — your what3words address — with other app users.

"We're not trying to replace street addresses," the company's chief marketing officer, Giles Rhys Jones, tells Axios. "It's more like a super zip code." 

Yes, but: For what3words to become the universal navigation tool the company aspires it to be, it needs to be integrated into rideshare and delivery apps like Uber and DoorDash.

Some services in the U.K. like DHL, DPD and Hermes are using what3words, but so far none of the big delivery companies in the U.S. have signed on.

Some emergency dispatch centers in the U.S. are using the technology, however. 

Carmakers adopt what3words app for voice navigation

Firm divided the world into 57 trillion squares and assigned each one a random three-word address.

Image of a 10-foot square block of pavement in NYC, identified by a three-word address.

Three words identify the precise location where a taxi might meet you. Image courtesy of what3words

Automobiles

One way to speed the adoption of self-driving vehicles is to give them automated express lanes, where they can zip past regular traffic without having to tangle with human-driven cars. 

Why it matters: For all the billions companies are spending to develop automated vehicles, very little has been done to prepare U.S. roadways for these advanced cars, trucks and buses. That's starting to change, as business and government leaders turn their attention to new digital infrastructure projects.

Driving the news: A startup company, Cavnue, just raised $130 million in Series A financing to build the nation's first network of connected and automated vehicle (CAV) corridors, starting in Michigan.

CAV corridors would have roadside sensors to communicate with modem-equipped vehicles, advising them of road conditions up ahead, such as deteriorating weather, road construction or even potholes that pop up during rush hour.

The company's investors all have an interest in the future of roadways. They include a high-tech infrastructure company (Sidewalk Infrastructure Partners), a global carmaker (Ford), a freight logistics provider (Landstar) and a global toll road operator (Globalvia's Openvia). 

The big picture: There's more attention on smart roadways since Congress passed the bipartisan infrastructure deal last November, opening the door for more federal funding, Laura Chace, CEO of ITS America, tells Axios.

Companies are working on a variety of technologies, including systems that warn cars about pedestrians (and vice versa) or that use AI and predictive analytics to manage traffic more efficiently.

A Safe Streets and Roads for All program will fund road tech projects designed to reduce the nearly 40,000 traffic deaths that occur each year.  

Cavnue is the first to focus on designing dedicated roadways for connected and automated vehicles, CEO Tyler Duvall tells Axios. 

Unlike other infrastructure companies that operate private toll roads, Cavnue is partnering with state and local governments to deploy the systems. 

Its first project, with the Michigan Department of Transportation, is a 25-mile segment of Interstate 94 near Detroit.

Initially, the CAV corridors would be available to a mixture of cars, but eventually they would be restricted to partly or fully automated vehicles. 

Those with Level 2 hands-free assisted-driving systems, like Cadillac's Super Cruise or Ford's Blue Cruise, would allow drivers to take their hands off the wheel and their eyes off the road while traveling in that chute. (On today's highways, the systems require drivers to pay attention.)

In the future, fully autonomous cars, trucks and buses would travel at steady speeds in the CAV corridors without any human interaction. 

The corridors would have "dynamic pricing," depending on the time of day or use case, and motorists would be billed monthly.

What they're saying: "We think [the CAV corridor] can only enhance and accelerate the benefit we'll get from these connected and automated vehicles on our roads," Chace says.

Self-driving cars will need smarter roadways

Cavnue is designing dedicated roadways for connected and automated vehicles, starting in Michigan.

Rendering of a highway with dedicated lanes for automated cars, trucks and buses

Rendering of a connected and automated vehicle (CAV) corridor. Image courtesy of Cavnue

The Department of Commerce has appointed more than two dozen experts to a committee that will advise President Joe Biden on artificial intelligence policy.

Why it matters: It's the Biden administration's first substantive step to shape U.S. policy on AI, as Europe, China and other countries leap ahead with their own rules.

Thursday's announcement comes as lawmakers and the public continue to grapple with AI's role in everyday life. House lawmakers launched an investigation into facial-recognition scanning company ID.me Thursday, per the Washington Post.

Driving the news: Twenty-seven people from across the private sector and academia, who were nominated by the public, will make up the National Artificial Intelligence Advisory Committee.

Creating such a committee was a direct order from the National AI Initiative Act of 2020, passed as part of 2021's National Defense Authorization Act.

The committee's recommendations will "serve as building blocks for U.S. AI policy for decades to come," U.S. deputy secretary of commerce Don Graves said in the announcement.

Congress has directed the committee to "submit a report to the President and Congress after the first year, and then again every three years, that provides their findings and recommendations..."

Details: The committee includes executives from companies like Google, Workday, Salesforce, NVIDIA, Microsoft and Amazon Web Services along with academics from Carnegie Mellon, Stanford and Cornell.

Part of the group's focus will be on "matters related to the use of AI in law enforcement," including bias and data security, per the announcement. 

Biden administration takes step forward on AI policy

It's the Biden administration's first substantive step to shape U.S. policy on AI.

Animated gif of a computer screen that reads "A.I" with the "I" as a blinking cursor

Artificial intelligence

Autonomous trucks could one day replace more than 90% of all highway trucking, which could have a profound impact on as many as 500,000 long-haul truckers, a new study found. 

Why it matters: Automation, and its potential impact on human labor, is a widespread concern for workers in many industries. While labor markets continually evolve, it often takes time for displaced workers to adapt to new jobs requiring different skills.

The big picture: Long-haul trucking is a grueling job, with a high turnover rate. Last year, the industry faced a record shortage of 80,000 drivers. 

Driving the news: The study by researchers at the University of Michigan and Carnegie Mellon University assessed the impact of autonomous trucking on operator-hours, depending on how the technology — now being tested in Sun Belt states — is deployed.

They looked at operator-hours, not jobs, because many truckers are independently employed. 

Using federal data about trucking shipments, and the operator-hours used to fulfill those shipments, they explored various deployment scenarios and how those might evolve over time.

What they found: The impacts of automation might not happen all at once. 

“If automation is restricted to Sun Belt states (including Florida, Texas and Arizona) — because the technology may not initially work well in rough weather — about 10% of the operator hours will be affected,” said study co-author Parth Vaishnav, assistant professor of sustainable systems at the University of Michigan School for Environment and Sustainability.

If automated trucks were deployed across the U.S., in all types of weather, then up to 94% of operator hours would be affected — the equivalent of 500,000 truck driver jobs, their research found.

The findings are consistent with a sprawling 2020 MIT study, which concluded that technological advances in AI and automation will have an enormous impact on the workforce, but it may take decades for those effects to be fully felt.

How it would work: Many automated trucking developers envision the use of transfer hubs, where cargo trailers would be handed off between humans and robots. 

Autonomous trucks would carry freight on the highway, between the hubs. 

Human drivers in conventional trucks would then take over on local streets to the final destination.

What they're saying: Drivers displaced by autonomous trucks could find work closer to home on short-haul routes, or even in newly created logistics jobs at transfer hubs, AT developers say. 

But the U-M study found that those jobs would not be sufficient to fully replace long-haul truckers' earnings.

Short-haul jobs typically pay less than long-haul jobs, the study noted.

Editor's note: This story originally published on March 28. 

Automation could send long-haul truckers driving into the sunset

Study finds self-driving trucks could eventually affect up to 500,000 drivers

Image of an Aurora self-driving truck on the highway

Aurora-powered PACCAR trucks are autonomously hauling freight for FedEx in pilot testing in Texas. Photo: Aurora

Trucks

Transportation

Customer service chatbots are becoming kinder, smarter and even more helpful, thanks to huge leaps in artificial intelligence.

Why it matters: Chatbots can be aggravating if their formulaic responses don't provide the help you're searching for — what customer service experts refer to as "the spiral of misery," per the New York Times. 

More capable AI promises to make those encounters less robotic with personas that employ "soft skills" like empathy to read between the lines and defuse tension. 

That could make customers happier and allow companies to spend less time answering questions on the website and more time focused on other things.

What's happening: Chatbots, computer programs intended to replicate human conversation, have been around for decades. Now, however, they are becoming far more ubiquitous — and personable. 

New machine learning techniques have made them much better at carrying on their end of the conversation, via both text and voice. 

Improvements in natural language processing mean bots are better at understanding and producing language.

They can even interpret the intent of a customer's inquiry and analyze what's transpiring during a chat session, explains Bern Elliot, a technology analyst at Gartner.

For example: "It sounds like you need to make a deposit, and you need to locate an ATM — is that correct?" he tells Axios.

Some of the most successful chatbots even have human personas like Nanci (GM Financial), Sydney (health insurance company Anthem) and Erica (Bank of America).

State of play: "This is fundamentally a tech-enabled surge in interest," Elliot says. "That doesn't mean they can't go wrong. They can and they do. Design is really important, and picking the right use cases is really important."

Chatbots serve as a successor to the automated phone trees still widely used by many organizations. ("If you want X, press 1... If you'd like to speak to a customer service representative, press zero.")

Chatbots interpret users' questions and reply from a library of pre-programmed answers.

Yes, but: Conversations can be stilted, and bots are only as good as the data on the back end.

Apple's Siri and Amazon's Alexa, for example, can search the entire internet to answer questions. But most chatbots are mining their company's own limited databases for answers to customer inquiries.

The more data they're trained on, the better they are at providing relevant answers. 

What to watch: New user-friendly AI products make it easy for non-techies to create custom chatbots without writing any lines of computer code themselves.

How it works: Recruiters at the University of Illinois used startup Juji's "no-code" software to create a custom AI chatbot to manage mounting inquiries about its fast-growing online MBA program.

They started by loading commonly asked questions and answers into a spreadsheet that Juji turned into a friendly chatbot. 

"In the beginning, it was a little bit slower," Myranda Crist, the university's assistant director of recruitment and admissions, tells Axios. "But we didn't have as much information on the back end. We couldn't possibly anticipate all the questions. As there were more questions, it's gotten more conversational."

The intrigue: Juji's software uses machine learning to infer a user's unique characteristics — are they an extrovert or an introvert, for example — in order to personalize each interaction, founder Michelle Zhou tells Axios. 

The bots are designed to be empathetic, and maybe even tell a joke if they detect unease, she said. "We want to make the person feel like, 'I've been heard'."

Our thought bubble: An efficient chatbot can be a blessing, but it's never going to be easy for customers to feel truly "heard" by a piece of code, even one shaped by AI. And cracking jokes trips up actual humans all the time. 

Chatbots are getting smarter — and nicer, too

Advances in AI are helping the programs read customers' mental states and interact with a little empat

Illustration of a friendly robot bearing flowers and an ice cream cone. 

Bots

Former Google CEO Eric Schmidt hit Capitol Hill this week to urge lawmakers to create a digital service academy that would train Americans in artificial intelligence and cybersecurity in exchange for government service.

Why it matters: The looming threat of cyberattacks from U.S. adversaries such as Russia underscores the need for tech-savvy employees through the federal workforce.

Driving the news: Schmidt envisions a four-year, digital training academy for civilians who would then work for the federal government for five years.

Schmidt told Axios in an interview that getting AI and cyber talent into the government is "crucial":  "The systems that are coming are so complicated that the government, who's well-meaning, will not be able to manage them. You're going to need better talent."

Initially, the program could be an approved curriculum at universities before becoming a separate academy.

The big picture: The federal government faces a severe shortage of workers skilled in artificial intelligence and cybersecurity, according to a Government Accountability Office report from November.

There are more than 38,000 open, public-sector cybersecurity positions according to a job-tracking database. 

The Biden administration is trying to accelerate efforts to fill cybersecurity positions and speed hiring for technical roles within the government.

The intrigue: Schmidt told Axios he met with members of the For Country Caucus, a bipartisan group of military veterans in Congress, to push them on including the idea as part of the National Defense Authorization Act, an annual must-pass defense spending bill.

Flashback: A digital service academy is a recommendation from the final report of the National Security Commission on Artificial Intelligence, which Schmidt led.

The report calls for both a U.S. Digital Service Academy to train current and future employees, and a civilian National Digital Reserve Corps for technically-skilled Americans to work 38 days a year for the government.

Yes, but: It's unclear if the measure will gain traction with lawmakers.

Sen. Kirsten Gillibrand (D-N.Y.) called for a Cyber Academy in an op-ed in December, and secured provisions in last year's spending bill she said lays the groundwork for such a facility.

Editor's note: This story has been corrected to remove references to "crypto" and "cryptocurrency" in Schmidt's description of the academy's focus. Schmidt tells Axios he misspoke and meant "cyber." 

Ex-Google CEO promotes digital West Point

The looming threat of cyberattacks underscores the need for tech-savvy federal employees.

Illustration of a group of computers featuring chalkboards with equations on them

Congress

Data privacy

A self-driving forklift may be the next wave of warehouse and factory innovations, saving companies on labor and, ideally, making industrial settings safer for everyone. 

Why it matters: Demand for logistics help is soaring, but forklift operators and other skilled workers are in short supply. Robotic vehicles are one way that manufacturers and logistics companies can address labor shortages while running their operations more safely and efficiently. 

Notably: While autonomous vehicles aren't yet ready for deployment on U.S. roadways, they're already being employed in factories and other industrial settings.

Driving the news: OTTO Motors, whose autonomous mobile robots are widely used in automotive plants, says its new self-driving forklift can pick up and carry up to 2,640 pounds and navigate crowded factory floors using many of the same sensors and software being developed for self-driving cars.

The forklift's software makes real-time decisions about where it's going and what it's picking up, and can steer around people and other obstacles.  

It can also help avoid accidents linked to poorly trained forklift operators who speed, make sharp turns or drive with elevated loads, causing tipovers.

Forklifts were the source of 78 work-related deaths and 7,290 nonfatal injuries resulting in days away from work in 2020, according to the National Safety Council. 

70% of all industrial accidents are caused by operator error, per the NSC.

Where it stands: Automated guided vehicles (AGVs) are fairly common in today's industrial settings, but they're not as useful or cost-effective as autonomous mobile robots (AMRs) like the OTTO Lifter. 

AGVs carry materials on a fixed route and are guided by wires, magnetic strips or sensors embedded in the floor, which can be expensive to install and update. 

AMRs are now being put to work in many factories because they're better at navigating dynamic environments.

Leading players, besides OTTO, include MiR, Fetch, Omron, Balyo, Seegrid and Geek+.

Driverless forklifts may be "poised for a breakthrough," as factories look to supplement — or replace — human forklift operators, reports DC Velocity, a supply chain publication.

By 2025, more than 2.5% of all industrial trucks will be automated forklifts, predicts Ash Sharma, a London-based analyst with Interact Analysis. 

Self-driving forklifts aim to make factories safer

Driverless forklifts help address the labor shortage and can move supplies around factories more safely.

Sections

Axios Local

Axios gets you smarter, faster with news & information that matters.

About

Subscribe

Automation and AI

The big picture

The disinformation threat from text-generating AI

AI is industrializing

How the automation economy can turn human workers into robots

AI and automation are creating a hybrid workforce

The next wave of job automation will be virtual

Automation is 2020's least understood issue

All Automation and AI stories