|Choice Exploration and Exploitation|
We all try to make good choices, right? But that's not so easy in changing environments, where the outcomes of our choices vary
over time. Take this example: there are three restaurants you could go to, A, B and C. Restaurant A is your favorite and you'd like
to go there as often as possible. That's what we would call exploiting your favorite. On the other hand, B and C keep changing, hiring
new staff, changing prices, introducing new food, etc.
The question is, shouldn't you explore something else than your favorite once in a while to make sure that your beliefs about the choices are still up-to-date? Some studies have shown that people systematically exploit favorites and explore alternatives. That means they are more likely to explore an alternative, if it has been a long time since they checked it for the last time. We refer to this type of exploration as uncertainty-minimizing.
In our restaurant example, people should be less likely to explore B and C if they have just done so and actually be more likely to go to restaurant A the next time (if it is still considered the favorite). Then again, if they hadn't been to B or C in a long time, they are more likely to go and check these restaurants.
In my project, I mainly work with a major UK retailer and analyze thousands of customers and their grocery shopping behavior. Here, we loook at the underlying patterns of people's exploratory choices. The question is: when do people exploit favorite products and when do they explore somehting new/different?
Unfortunately, I can't provide data here, but you'll be able to find the results of this research in our respective paper about exploration in the supermarket. Many people seem to follow a systematic pattern and explore alternatives less likely the longer they have been exploiting a favorite, which contradicts the experimental findings that I have mentioned before. However, there is a difference to typical lab studies: choosing a product is subjective (you decide what's tasty etc.) while the outcomes in lab studies are mostly objective (e.g. money which's value is clearly defined). Research involving subjective decision outcomes has shown that people actually tend to make self-confirmatory decisions and that their choices can even shape their preferences. We refer to this type of exploration as coherency-maximizing. This would concur with our findings that people go on long exploitation streaks and reduce exploration with every additional exploitation.
I'm working on this project together with Brad Love.
Let's be honest, every one of us has already, maybe just for fun, imagined what would happen if zombies attacked at some point.
Well, I have. And frankly, I can only see two things one could do: hide from them or fight them.
While this might be an individual decision, its outcome is determined on a group level. If you decide to hide, you don't want every one
else to hide and especially not in the same hiding place as you do. On the other hand, if you fight, you don't want to be alone.
Well, I'm just trying to make sense of a story here that actually points at a general social dilemma of public goods.
It's about the question whether we should pay our part of something that we share with others or whether we should let others pay
for it, still receive our share but save our own resources. When the zombies attack, our shared public good is protection and we can either
contribute by fighting and protecting or just enjoy the protection of others without contribution when we hide from the zombies.
The problem here is that if everyone hides, there is no protection. Then again, it's quite tempting to rest and let others put themselves
in danger for you. The whole story becomes a struggle between self-interest and responsibility towards the group.
We launched several experiments, some of them at the London Science Museum and some of them with undergraduate students at UCL. Our experiments are embedded in the zombie apocalypse story and people repeatedly decide whether to hide from the zombies or whether to fight them. Everyone received a base-rate of points that was connected to how many people decided to fight. On top of this, those who were hiding always received a relative bonus (e.g. 10% more). This created the public goods dilemma I have been talking about: if everyone fights, the base-rate of points for everyone will be very low. However, the hiders will always get relatively more points, which is really interesting once the base-rate of points is high with many people fighting. We model different types of people according to their reactions to the actions of others. This embodies very basic but intriguing attitudes that we hold about cooperation. On top of this, we have started to apply structural fMRI scans in order to identify properties in the brains of certain types of cooperators. We still need more data, since we are working with large groups and require a lot of different groups to verify our findings. I will provide some scripts under materials, where I have managed to build simple online multiplayer games that can, for example, be used for Amazon Mechanical Turk. That way, recruiting is a little easier.
My collaborators in this project are Brad Love, Daniel Richardson, Jorina von Zimmermann and Bahador Bahrami.
Which road should I take to work: the busy highway or the less congested country road? Depending on whether we make the right
or the wrong decision here, we will consider choosing differently the next time. Fast roads become more popular, making them slower due to
an increasing number of cars. Then again, the more people use a road, the less popular it becomes. This is basically popularity as a
function of itself. Feedback and perception is however delayed and people react differently, which makes coordination in such cases
sometimes rather difficult.
We conduct experiments with large groups and observe how people use information about the group and members of the group. For example, we examine whether information about the switching behavior within the group can help groups to coordinate more efficiently. In most coordination scenarios there are equilibria, where people distribute available resources in a pareto-efficient way. In the roads example, people should spread across the available roads so that travelling on each of them takes the same amount of time. Overall, this minimizes the time that all people together spend on the road.
Research in this area, especially regarding group sizes that we investigate, is still scarce and features many open questions. Coordination is a highly relevant social topic that covers, for example, various aspects of urban development.
Here, I work together with Brad Love, Daniel Richardson and Jorina von Zimmermann.
If you have ever wondered why there are so many betting shops in the UK, then take a look at Philip Newall's work.
He explains what kind of bets are very misleading to people, but financially very attractive to bookies.
Therefore, people feel like they would place smart bets which in fact they don't. We launched a project together, where people join our betting site and place
free bets on football matches. It's in an early stage, but you can check out the website here
if you're interested.
This project is a collaboration with Philip Newall.