The technological advances being developed in Artificial Intelligence (AI), Machine Learning, Computer Vision, the Internet of Things (IoT), and the complicated cohort of computing are impressive. Things that were deemed impossible merely a few years ago are being surpassed and antiquated on a daily basis. Such is the nature of technological development.
Still, automated technologies are designed by human engineers and, while capable of performing analytic tasks impossible for human beings, they’re still serving functions designed for humans, by humans.
Among these functions, the race for the first fleet of fully autonomous vehicles has technology and automotive companies either joining forces or competing for the coveted claim for first place. Regardless of who are allies or enemies, they’re all up against the same issues of addressing physical challenges with advanced technologies. In short, getting from point A to point B through an intricate network of sensors and programming. Most importantly, doing it safely.
Autonomous vehicle programs haven’t been without hindrances to their progress, the most prominent of which was Uber’s self-driving car killing a woman in Arizona earlier this year. They pulled their autonomous test vehicles off the roads in several states while the investigation went on to conclude the human safety operator was watching television on his phone. While tragic, this highlights the importance and need of vigilant human safety operators.
What we hear less about is how successful most autonomous vehicle programs are doing, racking up millions of miles and hours on the roads, free from human intervention. The progress is there, but there’s one consistent difficulty that plagues human and AI drivers alike: construction.
Now, construction may throw a wrench in the speed of your commute, and where AI beats human drivers is the inability to be angered by minor inconveniences. However, where human drivers are beating AI is in navigating the variability of construction zones. Autonomous vehicles have a difficult time with construction zones because even the most advanced algorithms can’t predict the intricacies of human variance and don’t know how to respond.
There are numerous ways road construction impedes autonomous cars and the variables range from human beings to physical roadscapes to technology to industries progressing at different rates. So, buckle up and don’t forget to read the signs.
Human beings stump systems
“Ok, Google, show me dresses from Free People.”
“Sorry, I couldn’t find any dressing for fee purple.”
How many times has something like this happened to you? We’ve all been frustrated by any number of voice assistants not understanding what exactly we want. Perhaps it’s rare, perhaps it happens all the time, but it presents a fundamental technological lesson. Human beings stump systems, no matter how advanced they may be.
Now, these systems are progressively getting smarter, but there’s still a long way to go. The application is that while your mobile assistant might not understand when you ask for a certain song, they’ll go silent and you can do it manually. When an autonomous vehicle doesn’t understand what’s in front of it at, say, 45 MPH, the risk is exponentially higher.
In construction zones, the chaos of human movement in an already disrupted roadway seriously hinders AI in autonomous vehicles.
When it’s a construction worker directing traffic, the non-verbal cues recognized from human to human aren’t understood by computer vision. Reflect back on these technologies being programmed by humans to understand human environment. Meanwhile, human beings often fail to understand some of our own idiosyncrasies. Translating them to AI becomes a challenge that we’re not sure can be solved.
Autonomous vehicles are programmed to understand roads in ideal driving conditions. Throw in massive construction vehicles, a variety of different signage, traffic cones, merging lanes, and the general disruption of order and the problem is clear. Artificial intelligence isn’t equipped to adequately deal with chaos.
Road traffic is an example often used to demonstrate the concept of chaos theory. A very basic definition of chaos theory is that in sensitive systems, very small changes can change the course of the entire system. Applied to traffic, one random bad lane switch can trigger an accident that closes several lanes of a highway. Could that have been predicted? Potentially, but not well enough to be certain. Add the randomness of human action and put it on the road with autonomous vehicles. Even with the most powerful predictive analytics, artificial intelligence can’t predict every small change and how it will affect a whole system. Change the system’s physicality with a construction zone and you’re asking AI to do more than it can (as of yet).
Industries with differential progress
Roy Amara’s law aptly describes human expectations of technological progress:
“We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.”
We’re living in the short run, giddy over the seemingly unending possibility and promise of autonomous vehicles ruling the roads. While technology firms are clawing at solutions and racing one another, they’re barely considering the impact beyond the technical achievement and the first place ribbon once they successfully iterate.
This is one of the important points that makes the construction industry and autonomous cars not play so nicely together. The two industries are progressing at wildly different rates. While construction is slower to change, technology companies are thinking about moonshot ideas without considering if they’re compatible with the real world they’ll operate in.
A major component of this is that road work isn’t always the most well-planned. Of course, it has plans that need be followed, but the nature of roadways and the work involved has to allow for variability. In the line of building physical infrastructure, things change and human workers adapt to those changing needs. Artificial intelligence in autonomous vehicles is programmed schematically to understand fairly ideal road conditions with minimal variance.
“There are times when we don’t know what the site will look like until we show up in the morning. Or we’ll finish one section, then move things around to work on another area,” says Kevin Curtis, a New Jersey construction worker.
Not being able to assess and synthesize the variable human elements ubiquitous in roadside construction puts the lives and safety of those construction workers at risk. They’re already at risk enough with human drivers. While risk is high with human drivers, it only increases with self-driving vehicles.
Another problematic aspect defined by differing technological progress is access to and sharing of data. The Depart of Transportation is broken down into pieces by state, county, and even smaller, making systematic monitoring of construction unreliable at best.
“No national database tracks the location of ongoing construction zones, forcing computers to rely on information posted by a wide assortment of state and local agencies — which may or may not be keeping real-time updates of their progress.”
Autonomous vehicles rely on consuming all of the data they have access to. Where, say, construction companies could be held to data storage standards across the nation, allowing autonomous vehicles to access and analyze construction data, making AI better able to learn from the information provided. However, this doesn’t exist, leaving autonomous vehicles to fend for themselves or depend on human operators. Either way this compounds potential safety risks.
For autonomous vehicles and every aspect of roadway travel to mesh harmoniously, they need to start looking into collaborative safety standards. Where the two industries are operating apart from one another, they share a very real space with each other. You’d think that with some 10 million autonomous vehicles being on the road by 2020, that the Department of Transportation, various players in the construction industry, and autonomous vehicle developers might team up to build the future of travel together. Unfortunately, it’s been rather slow on the uptake.
Are there solutions?
Now that we’ve explored the tip of the safety conundrum iceberg for autonomous vehicles and roadside construction, where are we headed next? One thing is certain, construction will always be there and autonomous vehicles will become more and more prevalent moving forward. What solutions might create a more symbiotic relationship between the two?
We already mentioned that industry collaboration and data sharing is the foundation of this. The how is a bit more tricky. Yes, AI might not be able to read every variance of human beings, but it’s still really smart.
Michigan will install 17 miles of the first smart highway in the United States. Utilizing IoT, there will be electronics along the highway, signs and other pieces of hardware, that can electronically communicate with vehicles. Using dedicated short-range communication (DSRC), a technology that will be required in all vehicles by 2020, the road, signage, and other cars will be able to interact with one another.
“As vehicles become increasingly connected, the infrastructure must also be updated, not only for safety, but for reliability with this new technology,” said a Michigan Department of Transportation representative.
In the form of signs that can still be read by human drivers, signage will include the equivalent of QR codes for cars to read as well. With the advent of these technologies, they can be applied to roadside construction. Imagine handheld signs, cones, other construction vehicles, and even human wearables that can communicate with autonomous vehicles. The Internet of Things brings a multitude of possibilities in the future of connectivity, and the safety implications can be substantially positive.
Still, this is only an experiment with 17 miles of Michigan highway. We’re a long way off from having a nationwide network of smart highways, let alone autonomous vehicles on all of them. As it stands right now, the simplest solution for autonomous cars might be avoiding construction zones altogether. As if a detour is detected on the GPS, when construction zones are ahead, self-driving cars treat them as detours and simply find another way. It’s a bandaid on an issue that will require a more elegant solution sooner rather than later, but still a viable option.
As the race continues, the infinite variability of human-machine interaction will only expand, but the solutions will expand with it.
Let’s not get ahead of ourselves
“With autonomy, the edge cases kill you, so you’ve got to build out for all the edge cases. Which makes it a very, very difficult problem,” Uber’s CEO Dara Khosrowshahi said.
Roadside construction is an edge case in the realm of autonomous vehicles. Concern for human safety is always the pinnacle of development and progress, but trying to address the variably infinite edge cases human beings imbue is the essence of chaos theory.
Still, the work continues and as humans and machines share more spaces, infrastructure will have to follow suit. The paradoxical ability for human beings to adapt and evolve yet be so apprehensive of change is the embodiment of slowed progress.
Autonomous technologies, among millions of others, aren’t going away; they’re only getting better and more common. Human beings aren’t going anywhere either. In fact, we’re the ones developing this technology, so shouldn’t our own physical safety be at the core of development? It’s Amara’s law in action again. We’re seeing the technological short term versus the long term. By taking pause in the developmental pace of technologies in one sector, we stand to lay firmer foundations for long term progress when sectors develop collaboratively. In this case, autonomous vehicles, roadway construction authorities, and governing bodies shouldn’t be developing independently of one another. Because they’ll never be operating independently of one another.
Infrastructural and technological symbiosis will only stand firmly on this concept. Where does the safety of the construction worker lie? Learning and developing with the very machines that would cause danger. Where does advanced AI begin to better evaluate human safety edge cases? With the construction groups that would present autonomous vehicles with those edge cases.
Invariably, the learning and development processes of these two interconnected facets of humanity and technology must be done together. Otherwise, what’s the point?