On Monday, March 11 at the 81st Annual Truckload Carriers Association (TCA) Conference at The Wynn in Las Vegas, program director Chris Henry moderated a wide-ranging audience discussion about the risks and opportunities posed by autonomous trucking technology.
One of the anxiety-laden themes in the discussion was embodied by contradiction between the American Trucking Associations’ (ATA) contention that the organization advocated for driver-assist, not driver-less technology, and McKinsey & Company’s view that completely automated trucks could generate cost savings of 45 percent. As one audience member pointed out, 45 percent cost savings imply an elimination of driver wages, benefits and recruiting – in other words, true driverless vehicles. The question was whether autonomous technology developers could justify the cost of research and development if drivers were not actually eliminated.
“In our view,” said Henry, “the only true protection against massive consolidation is keeping drivers in the truck as long as possible.”
Henry appeared to be referring to a scenario where Level 5 autonomous trucks would complete the total commodification of trucking capacity – the additional capacity brought into the market by more productive trucks would crash rates, and then trucking companies wouldn’t have anything to distinguish themselves from each other. The industry would shift from being labor-intensive to capital-intensive, and overall margins would contract, forcing consolidation.
The audience tended to engage on a more granular level.
One point of contention seemed to be whether productivity gains could actually be realized from Level 4 autonomous technology. Several audience members opined that if a driver must stay attentive, must be ready to take over the operation of the truck in the event of adverse road conditions, sensor failure or a software glitch, over the course of the day the driver’s attention level would be comparable to driving a truck today.
A driver prepared to take over at a moment’s notice, in other words, would not be truly resting, and therefore there would be no compelling reason to argue for a revision in the hours of service regulation. In theory, the cost of developing and implementing the technology could be justified on pure safety grounds, but the trucks themselves would not necessarily generate more revenue.
Two audience members drew an analogy to airline pilots, saying that the autopilot feature on aircraft has been around for decades, but that human pilots were still retained for the difficult tasks, like landing and taking off, and that the general public would probably not be willing to get onboard a plane that had no pilots. Riffing on airlines in general, Chris Henry pointed out that the two recent Boeing 737 Max 8 crashes appear to have been caused by the same software bug, and asked what that meant for the prospects of Level 5 autonomous trucking.
The consensus seemed to be that not only was over-reliance on software a safety threat, but that, in the case of a highly connected trucking ecosystem, it also opened up cybersecurity vulnerabilities.
At some points the conversation even got philosophical. One respondent brought up the trolley problem, the famous thought experiment in ethics. Imagine that you’re driving a train, and ahead of the train, you see five people on the tracks. You have the opportunity to switch to a side track, but there is one person on that track. The dilemma comes from the question of whether it is morally permissible to affirmatively choose to kill one person to save five lives.
In autonomous vehicle technology, the dilemma becomes even more complicated. One respondent said that Mercedes-Benz designs its driverless technology to prioritize the safety of the driver over anything else; a Mercedes-Benz SUV might choose to hit a family in a Ford Focus rather than a Class-8 truck to protect the driver of the Mercedes-Benz, even though that choice leads to more fatalities.
Another respondent pointed out an approach to training machine-learning algorithms to deal with the trolley problem that generated problematic results. Humans were given images of people and objects on the road and had to pick one of them to steer the car toward in the event of an emergency, and those choices were fed into an algorithm so that the autonomous vehicle would at least behave similarly to a human driver. The problem was that the people inputting the preferences in the algorithm tended to prefer saving affluent-looking white people to saving working-class-looking people of color. Clearly, training a driverless car to merely replicate humans’ unconscious biases could introduce even more vulnerabilities to litigation than autonomous vehicles already have.
The conversation was a robust discussion of a wide array of issues around autonomous technology, informed by the expertise of industry professionals from both truckload carriers and original equipment manufacturers. What remains clear is that there are many more questions to be answered before autonomous trucking impacts freight markets in a meaningful way.