The holiday travel season is once again upon us! It's the magical time of the year that combines standing in airport security lines with incrementally losing your mind as the hands of your watch perpetually tick closer to a boarding time that somehow moved up 45 minutes since you left the house and the goober in front of you is, in the year of our lord 2022, still somehow confused about why we have to take our shoes off in security and, goddamit dude, stop arguing with the TSA and untie your laces already these tickets are nonrefundable.
AI can help fix this. It can perhaps even give regular folks a taste of the effortless airport experience that more well-heeled travelers enjoy — the private jet set who don't ever have to worry about departure times or security lines like the rest of us schmucks stuck flying Spirit.
In their latest book POWER AND PREDICTION: The Disruptive Economics of Artificial Intelligence, University of Toronto economists and professors Ajay Agrawal, Joshua Gans and Avi Goldfarb examine the foundational impact that AI/ML systems have on human decision making as we increasingly rely on automation and big data predictions. In the excerpt below, they posit what the airports of tomorrow might look like if AI eliminates traffic congestion and security delays.
Reprinted by permission of Harvard Business Review Press. Excerpted from POWER AND PREDICTION: The Disruptive Economics of Artificial Intelligence by Ajay Agrawal, Joshua Gans, and Avi Goldfarb. Copyright 2022 Ajay Agrawal, Joshua Gans, and Avi Goldfarb. All rights reserved.
Ajay Agrawal, Joshua Gans, and Avi Goldfarb, economists and professors at University of Toronto's Rotman School of Management. Their previous book is PREDICTION MACHINES: The Simple Economics of Artificial Intelligence.
The Alternative Airport Universe
Before considering the threat AI prediction may pose to airports, as with everything, there is an alternative system that can show us what the other side looks like. One example is the alternative universe of the very, very wealthy. They don’t fly commercial and so have no occasion to deal with either the old or newly designed public airport terminals. Instead, they fly privately and go through private terminals. Normally, glitz, glamour, nice restaurants, and art galleries are going to be where the very rich are. But in the world of airports, private terminals are positively spartan.
The reason there is no investment in making private terminals better places is that the very uncertainty that plagues the rest of us doesn’t plague the rich. With a commercial plane, you are tied to a schedule, and those planes will leave late passengers behind. With a private plane, the schedule is more flexible or even nonexistent. If the passengers aren’t there, the plane doesn’t leave until they arrive. If the passengers are there earlier, the plane leaves then. The whole system is designed so there is no waiting—at least, on the part of the passengers. No waiting means no need to invest in making waiting more pleasant. At the same time, the rich don’t have rules about when they need to leave for the airport. They leave when they want. If more people could have that experience, then surely the optimal terminal would be more spartan than cathedral.
You don’t have to be rich, however, to see this alternative universe. Instead, just compare the world on the other side of the arrival gates to those at departure. When arrival areas are separated from departure areas, they are spartan. You might find some light food outlets, but everything else is designed to get you out of the airport. The critical issue is how close the taxi and parking facilities are, even though you may not be in a stressful rush. Do you even remember any details of arrivals at your regular airport, other than how best to get out?
The AI Airport Threat
Airports are no strangers to AI. Air traffic control has adopted AI-based systems to better predict aircraft arrivals and congestion. At Eindhoven Airport, a new AI baggage-handling system is being piloted whereby passengers simply photograph their bags, drop them off, and pick them up at their destination—no labels required. Subject to privacy requirements, it hopes to do the same with people. All this will help you get to your flight more quickly.
None of these things, however, hit at the key drivers of uncertainty in your travel to your flight — traffic and security. Change, however, is already here with regard to traffic. Navigational apps such as Waze account for traffic conditions and can reasonably estimate how long it takes to get to any airport based on the time of day. The apps aren’t perfect, but they keep getting better.
The apps free passengers from having rules that tell them how early they need to leave for the airport. Instead, they can add that flight time to their calendar, and an app tells them the best time to depart and schedule their time accordingly. Even better, in the near future, the uncertainty in the actual time a flight leaves will be taken into account. Rather than just telling you when you need to leave based on a scheduled departure, the app will tell you when to leave depending on the flight’s predicted actual departure. Again, there is residual uncertainty, but the leap from having no information to having more precise information could save hours of waiting time. Similarly, many Uber riders who previously thought they wouldn’t care about knowing the predicted arrival time of their taxi now cite that information as one of the most valuable features of the service. Uber uses AI to make that prediction. AI could also predict security line wait times. Put it all together, and you can use the AI to decide when to leave for the airport rather than rely on rules. As with everything, there will be some who leap at this possibility ahead of others. At Incheon and many other airports, waiting isn’t bad anymore, so maybe you don’t need to make an informed decision.
Those developing an AI-driven navigation app or flight departure predictor have no direct interest in the earnings of in-terminal airport activities. However, the value of their AI applications depends critically on how many people do not want to wait at airports. Thus, if airports are currently less costly to wait in, the value of those apps is diminished. The security line prediction is another matter. Airports claim that they want to improve security times and reduce uncertainty. But as economists, we don’t think their incentives are aligned with passengers. Yes, improving security times leaves more time to spend at the facilities past security. But, at the same time, it will reduce uncertainty and cause people to tighten their airport arrival times. Combined with AI that solves the other uncertainty for passengers in getting to the terminal, will the airports want to eliminate the uncertainty under their own control?
Our broader point is not about airports but about rules. Rules arise because it is costly to embrace uncertainty, but they create their own set of problems. The so-called Shirky Principle, put forth by technology writer Clay Shirky, states that “institutions will try to preserve the problem to which they are the solution.” The same can be said of businesses. If your business is to provide a way to help people when they wait for a plane, what’s the chance you are going to ensure they don’t have to wait for planes?
If you want to find opportunities by creating new AI-enabled decisions, you need to look beyond the guardrails that protect rules from the consequences of uncertainty and target activities that make bearing those costs easier or to reduce the likelihood of bad outcomes that the rules would otherwise have to tolerate.
We can see this in the long-standing protection farmers employ in England — building hedgerows. A hedgerow is a carefully planned set of robust trees and plants that serve as a wall between fields. It is extremely useful if your field is full of farm animals, and you do not want to employ a person to ensure they do not wander off. It is also useful if you do not want heavy rainfall to erode soil too quickly or if you want to protect crops from strong winds. Given all this protection against risky events, we are not surprised that this practice was the origin of the term “hedging,” which evolved to have a broader insurance meaning.
But hedgerows come at a cost. By dividing farmland, they make it impossible to use certain farming techniques — including mechanization — that are only efficient for large swathes of land. After World War II, the British government actually subsidized the removal of hedgerows, although in some cases, that removal was excessive, given their role in risk management. Today, there is a movement to restore hedgerows, led most prominently by the Prince of Wales. In many situations, costly investments are made to cover or shelter a would-be decision-maker from risk. Miles of highways are cocooned with guardrails to prevent cars from going down embankments, hills, or into oncoming traffic. Most are, fortunately, never used, but each allows a road to be built in a way that might have otherwise not been sufficiently safe, given the fallibility of human drivers.
More generally, building codes precisely specify various measures to protect those inside buildings from uncertain events. These include fire, but also damage from weather, weak building foundations, and other natural phenomena like earthquakes.
What these protection measures have in common is that they typically generate what looks like over-engineered solutions. They are designed for a certain set of events — the once-in-a-lifetime storm or the once-in-a-century flood. When those events occur, the engineering seems worthwhile. But, in their absence, there is cause to wonder. For many years, Freakonomics authors Steven Levitt and Stephen Dubner pointed out how life vests and rafts on aircraft — not to mention the safety demonstrations of each — appeared wasteful, given that no aircraft had successfully landed on water. Then, in 2009, Captain Sullenberger landed a US Airways plane with no working engines on the Hudson River. Does that one example of a low-probability event make the precautionary life vests worth it? It is hard to know. But we cannot conclude that the absence of a possible outcome causes us to assess the probability of that outcome at zero.
Levitt and Dubner’s main point, however, is that while it is often possible when protection measures are employed to assess the likelihood or change in the likelihood of underlying uncertainty over time, it is not possible to measure whether the investments made to reduce the probability of a consequence are excessive, as the very risk management strategy employed takes away that information. It is entirely possible that too much is wasted on something that, for other reasons, is no longer high risk at all.