In the first phase of Ware’s study, four of ten infants receiving conventional treatment died, and all nine of nine infants receiving ECMO survived. The four deaths were enough to trigger a transition to the second phase, where all twenty patients were treated with ECMO and nineteen survived. Ware and colleagues were convinced, concluding that “it is difficult to defend further randomization ethically.”
But some had already concluded this before the Ware study, and were vocal about it. The critics included Don Berry, one of the world’s leading experts on multi-armed bandits. In a comment that was published alongside the Ware study in Statistical Science, Berry wrote that “randomizing patients to non-ECMO therapy as in the Ware study was unethical.… In my view, the Ware study should not have been conducted.”
And yet even the Ware study was not conclusive for all in the medical community. In the 1990s yet another study on ECMO was conducted, enrolling nearly two hundred infants in the United Kingdom. Instead of using adaptive algorithms, this study followed the traditional methods, splitting the infants randomly into two equal groups. The researchers justified the experiment by saying that ECMO’s usefulness “is controversial because of varying interpretation of the available evidence.” As it turned out, the difference between the treatments wasn’t as pronounced in the United Kingdom as it had been in the two American studies, but the results were nonetheless declared “in accord with the earlier preliminary findings that a policy of ECMO support reduces the risk of death.” The cost of that knowledge? Twenty-four more infants died in the “conventional” group than in the group receiving ECMO treatment.
The widespread difficulty with accepting results from adaptive clinical trials might seem incomprehensible. But consider that part of what the advent of statistics did for medicine, at the start of the twentieth century, was to transform it from a field in which doctors had to persuade each other in ad hoc ways about every new treatment into one where they had clear guidelines about what sorts of evidence were and were not persuasive. Changes to accepted standard statistical practice have the potential to upset this balance, at least temporarily.
After the controversy over ECMO, Don Berry moved from the statistics department at the University of Minnesota to the MD Anderson Cancer Center in Houston, where he has used methods developed by studying multi-armed bandits to design clinical trials for a variety of cancer treatments. While he remains one of the more vocal critics of randomized clinical trials, he is by no means the only one. In recent years, the ideas he’s been fighting for are finally beginning to come into the mainstream. In February 2010, the FDA released a “guidance” document, “Adaptive Design Clinical Trials for Drugs and Biologics,” which suggests—despite a long history of sticking to an option they trust—that they might at last be willing to explore alternatives.
The Restless World
Once you become familiar with them, it’s easy to see multi-armed bandits just about everywhere we turn. It’s rare that we make an isolated decision, where the outcome doesn’t provide us with any information that we’ll use to make other decisions in the future. So it’s natural to ask, as we did with optimal stopping, how well people generally tend to solve these problems—a question that has been extensively explored in the laboratory by psychologists and behavioral economists.
In general, it seems that people tend to over-explore—to favor the new disproportionately over the best. In a simple demonstration of this phenomenon, published in 1966, Amos Tversky and Ward Edwards conducted experiments where people were shown a box with two lights on it and told that each light would turn on a fixed (but unknown) percentage of the time. They were then given 1,000 opportunities either to observe which light came on, or to place a bet on the outcome without getting to observe it. (Unlike a more traditional bandit problem setup, here one could not make a “pull” that was both wager and observation at once; participants would not learn whether their bets had paid off until the end.) This is pure exploration vs. exploitation, pitting the gaining of information squarely against the use of it. For the most part, people adopted a sensible strategy of observing for a while, then placing bets on what seemed like the best outcome—but they consistently spent a lot more time observing than they should have. How much more time? In one experiment, one light came on 60% of the time and the other 40% of the time, a difference neither particularly blatant nor particularly subtle. In that case, people chose to observe 505 times, on average, placing bets the other 495 times. But the math says they should have started to bet after just 38 observations—leaving 962 chances to cash in.
Other studies have produced similar conclusions. In the 1990s, Robert Meyer and Yong Shi, researchers at Wharton, ran a study where people were given a choice between two options, one with a known payoff chance and one unknown—specifically two airlines, an established carrier with a known on-time rate and a new company without a track record yet. Given the goal of maximizing the number of on-time arrivals over some period of time, the mathematically optimal strategy is to initially only fly the new airline, as long as the established one isn’t clearly better. If at any point it’s apparent that the well-known carrier is better—that is, if the Gittins index of the new option falls below the on-time rate of the familiar carrier—then you should switch hard to the familiar one and never look back. (Since in this setup you can’t get any more information about the new company once you stop flying it, there is no opportunity for it to redeem itself.) But in the experiment, people tended to use the untried airline too little when it was good and too much when it was bad. They also didn’t make clean breaks away from it, often continuing to alternate, particularly when neither airline was departing on time. All of this is consistent with tending to over-explore.
Finally, psychologists Mark Steyvers, Michael Lee, and E.-J. Wagenmakers have run an experiment with a four-armed bandit, asking a group of people to choose which arm to play over a sequence of fifteen opportunities. They then classified the strategies that participants seemed to use. The results suggested that 30% were closest to the optimal strategy, 47% most resembled Win-Stay, Lose-Shift, and 22% seemed to move at random between selecting a new arm and playing the best arm found so far. Again, this is consistent with over-exploring, as Win-Stay, Lose-Shift and occasionally trying an arm at random are both going to lead people to try things other than the best option late in the game, when they should be purely exploiting.
So, while we tend to commit to a new secretary too soon, it seems like we tend to stop trying new airlines too late. But just as there’s a cost to not having a secretary, there’s a cost to committing too soon to a particular airline: the world might change.
The standard multi-armed bandit problem assumes that the probabilities with which the arms pay off are fixed over time. But that’s not necessarily true of airlines, restaurants, or other contexts in which people have to make repeated choices. If the probabilities of a payoff on the different arms change over time—what has been termed a “restless bandit”—the problem becomes much harder. (So much harder, in fact, that there’s no tractable algorithm for completely solving it, and it’s believed there never will be.) Part of this difficulty is that it is no longer simply a matter of exploring for a while and then exploiting: when the world can change, continuing to explore can be the right choice. It might be worth going back to that disappointing restaurant you haven’t visited for a few years, just in case it’s under new management.
In his celebrated essay “Walking,” Henry David Thoreau reflected on how he preferred to do his traveling close to home, how he never tired of his surroundings and always found something new or surprising in the Massachusetts landscape. “There is in fact a sort of harmony discoverable between the capabilities of the landscape within a circle of ten miles’ radius, or the limits of an afternoon walk, and the threescore years and ten of human life,” he wrote. “It will never become quite familiar to you.”
To live in a restless world requires a certain restlessness in oneself.
So long as things continue to change, you must never fully cease exploring.
Still, the algorithmic techniques honed for the standard version of the multi-armed bandit problem are useful even in a restless world. Strategies like the Gittins index and Upper Confidence Bound provide reasonably good approximate solutions and rules of thumb, particularly if payoffs don’t change very much over time. And many of the world’s payoffs are arguably more static today than they’ve ever been. A berry patch might be ripe one week and rotten the next, but as Andy Warhol put it, “A Coke is a Coke.” Having instincts tuned by evolution for a world in constant flux isn’t necessarily helpful in an era of industrial standardization.
Perhaps most importantly, thinking about versions of the multi-armed bandit problem that do have optimal solutions doesn’t just offer algorithms, it also offers insights. The conceptual vocabulary derived from the classical form of the problem—the tension of explore/exploit, the importance of the interval, the high value of the 0–0 option, the minimization of regret—gives us a new way of making sense not only of specific problems that come before us, but of the entire arc of human life.
Explore …
While laboratory studies can be illuminating, the interval of many of the most important problems people face is far too long to be studied in the lab. Learning the structure of the world around us and forming lasting social relationships are both lifelong tasks. So it’s instructive to see how the general pattern of early exploration and late exploitation appears over the course of a lifetime.
One of the curious things about human beings, which any developmental psychologist aspires to understand and explain, is that we take years to become competent and autonomous. Caribou and gazelles must be prepared to run from predators the day they’re born, but humans take more than a year to make their first steps. Alison Gopnik, professor of developmental psychology at UC Berkeley and author of The Scientist in the Crib, has an explanation for why human beings have such an extended period of dependence: “it gives you a developmental way of solving the exploration/exploitation tradeoff.” As we have seen, good algorithms for playing multi-armed bandits tend to explore more early on, exploiting the resulting knowledge later. But as Gopnik points out, “the disadvantage of that is that you don’t get good payoffs when you are in the exploration stage.” Hence childhood: “Childhood gives you a period in which you can just explore possibilities, and you don’t have to worry about payoffs because payoffs are being taken care of by the mamas and the papas and the grandmas and the babysitters.”
Thinking about children as simply being at the transitory exploration stage of a lifelong algorithm might provide some solace for parents of preschoolers. (Tom has two highly exploratory preschool-age daughters, and hopes they are following an algorithm that has minimal regret.) But it also provides new insights about the rationality of children. Gopnik points out that “if you look at the history of the way that people have thought about children, they have typically argued that children are cognitively deficient in various ways—because if you look at their exploit capacities, they look terrible. They can’t tie their shoes, they’re not good at long-term planning, they’re not good at focused attention. Those are all things that kids are really awful at.” But pressing buttons at random, being very interested in new toys, and jumping quickly from one thing to another are all things that kids are really great at. And those are exactly what they should be doing if their goal is exploration. If you’re a baby, putting every object in the house into your mouth is like studiously pulling all the handles at the casino.
More generally, our intuitions about rationality are too often informed by exploitation rather than exploration. When we talk about decision-making, we usually focus just on the immediate payoff of a single decision—and if you treat every decision as if it were your last, then indeed only exploitation makes sense. But over a lifetime, you’re going to make a lot of decisions. And it’s actually rational to emphasize exploration—the new rather than the best, the exciting rather than the safe, the random rather than the considered—for many of those choices, particularly earlier in life.
What we take to be the caprice of children may be wiser than we know.
… And Exploit
I had reached a juncture in my reading life that is familiar to those who have been there: in the allotted time left to me on earth, should I read more and more new books, or should I cease with that vain consumption—vain because it is endless—and begin to reread those books that had given me the intensest pleasure in my past.
—LYDIA DAVIS
At the other extreme from toddlers we have the elderly. And thinking about aging from the perspective of the explore/exploit dilemma also provides some surprising insights into how we should expect our lives to change as time goes on.
Laura Carstensen, a professor of psychology at Stanford, has spent her career challenging our preconceptions about getting older. Particularly, she has investigated exactly how, and why, people’s social relationships change as they age. The basic pattern is clear: the size of people’s social networks (that is, the number of social relationships they engage in) almost invariably decreases over time. But Carstensen’s research has transformed how we should think about this phenomenon.
The traditional explanation for the elderly having smaller social networks is that it’s just one example of the decrease in quality of life that comes with aging—the result of diminished ability to contribute to social relationships, greater fragility, and general disengagement from society. But Carstensen has argued that, in fact, the elderly have fewer social relationships by choice. As she puts it, these decreases are “the result of lifelong selection processes by which people strategically and adaptively cultivate their social networks to maximize social and emotional gains and minimize social and emotional risks.”
What Carstensen and her colleagues found is that the shrinking of social networks with aging is due primarily to “pruning” peripheral relationships and focusing attention instead on a core of close friends and family members. This process seems to be a deliberate choice: as people approach the end of their lives, they want to focus more on the connections that are the most meaningful.
In an experiment testing this hypothesis, Carstensen and her collaborator Barbara Fredrickson asked people to choose who they’d rather spend thirty minutes with: an immediate family member, the author of a book they’d recently read, or somebody they had met recently who seemed to share their interests. Older people preferred the family member; young people were just as excited to meet the author or make a new friend. But in a critical twist, if the young people were asked to imagine that they were about to move across the country, they preferred the family member too. In another study, Carstensen and her colleagues found the same result in the other direction as well: if older people were asked to imagine that a medical breakthrough would allow them to live twenty years longer, their preferences became indistinguishable from those of young people. The point is that these differences in social preference are not about age as such—they’re about where people perceive themselves to be on the interval relevant to their decision.
Being sensitive to how much time you have left is exactly what the computer science of the explore/exploit dilemma suggests. We think of the young as stereotypically fickle; the old, stereotypically set in their ways. In fact, both are behaving completely appropriately with respect to their intervals. The deliberate honing of a social network down to the most meaningful relationships is the rational response to having less time to enjoy them.
Recognizing that old age is a time of exploitation helps provide new perspectives on some of the classic phenomena of aging. For example, while going to college—a new social environment filled with people you haven’t met—is typically a positive, exciting time, going to a retirement home—a new social environment filled with people you haven’t met—can be painful. And that difference is partly the result of where we are on the explore/exploit continuum at t
hose stages of our lives.
The explore/exploit tradeoff also tells us how to think about advice from our elders. When your grandfather tells you which restaurants are good, you should listen—these are pearls gleaned from decades of searching. But when he only goes to the same restaurant at 5:00 p.m. every day, you should feel free to explore other options, even though they’ll likely be worse.
Perhaps the deepest insight that comes from thinking about later life as a chance to exploit knowledge acquired over decades is this: life should get better over time. What an explorer trades off for knowledge is pleasure. The Gittins index and the Upper Confidence Bound, as we’ve seen, inflate the appeal of lesser-known options beyond what we actually expect, since pleasant surprises can pay off many times over. But at the same time, this means that exploration necessarily leads to being let down on most occasions. Shifting the bulk of one’s attention to one’s favorite things should increase quality of life. And it seems like it does: Carstensen has found that older people are generally more satisfied with their social networks, and often report levels of emotional well-being that are higher than those of younger adults.
So there’s a lot to look forward to in being that late-afternoon restaurant regular, savoring the fruits of a life’s explorations.
*The basic summary of this section: git while the Gittins’s good.
3 Sorting
Making Order
Nowe if the word, which thou art desirous to finde, begin with (a) then looke in the beginning of this Table, but if with (v) looke towards the end. Againe, if thy word beginne with (ca) looke in the beginning of the letter (c) but if with (cu) then looke toward the end of that letter. And so of all the rest. &c.
—ROBERT CAWDREY, A TABLE ALPHABETICALL (1604)
Algorithms to Live By Page 7