In a wonderful and very interesting turn of events, ethical algorithms are suddenly all the rage. Cathy O’Neil wrote a book called Weapons of Math Destruction, in which she went through a couple of interesting case examples of how algorithms can work in an unethical and destructive fashion. Her examples came from the US, but that the phenomenon doesn’t limit itself on the other side of the pond.

In fact, just a month ago, the Economist reported on the rise of credit cards in China. The consumption habits in China are becoming closer to resembling Western ones, including the use of credit cards. And where you have credit cards, you also have credit checks. But how do you show your creditworthiness, if you haven’t had credit?

Enter Sesame Credit, a rating firm. According to the Economist, they rely on “users’ online-shopping habits to calculate their credit scores. Li Yingyun, a director, told Caixin, a magazine, that someone playing video games for ten hours a day might be rated a bad risk; a frequent buyer of nappies would be thought more responsible.” Another firm called China Rapid Finance relies on looking at users’ social connections and payments. My guess would be that their model predicts your behavior based on the behavior of your contacts. So if you happen to be connected to a lot of careless spend-a-holics, too bad for you.
​
Without even getting to the privacy aspects of such models, one concerning aspect – and this is the main thrust of O’Neil’s book – is that these kinds of models can discriminate heavily based completely on aggregate behavior. For example, if CRF:s model sees your friends spending and not paying their bills, they might classify you as a credit risk, and not give you a credit card. And if there is little individual data about you, this kind of aggregate data can form the justification of the whole decision. Needless to say, it’s quite unfair that you can be denied credit – even when you’re doing everything right – just because of your friends’ behavior.

Now, O’Neil’s book is full of similar cases. To be honest, the idea is quite straightforward. The typical signs of an unethical model (in O’Neil’s terms, a Weapon of Math Destruction) has a few signs: 1) they have little to no feedback to learn from, and 2) they make decisions based on aggregate data. The second one was already mentioned, but the first one seems even more damning.

A good example of the first kind is generously provided by US education systems. Now, in the US, rankings of schools are all the rage. Such rankings are defined in with a complicated equation, that takes into account how well outgoing students do. And of course, the rankings drive the better students to the better schools. However, the model never actually learns any of the variables and their importance from data – these are all defined by pulling them from the administrators’, programmers’, and politicians’ collective hats. What could go wrong? What happens with systems like these, is that the ranking becomes a self-fulfilling prophecy, and that changing how the ranking is calculated becomes impossible, because the schools that do well are obviously up in arms about any changes.

This whole topic of discrimination in algorithms is actually gaining some good traction. In fact, people at Google are taking notice. In a paper that was recently presented at NIPS, the authors argue that what is needed is a concept of equality of opportunity in supervised learning. The idea is simple: if you have two groups, (like two races, or rich and poor, etc.) in both groups the true positive rate should be the same. In the context of loans, for example, this means that of all those who could pay back loans, the same percentage of people are given a loan. So if groups A and B have 800 and 100 people that could pay the loan back, and your budget can account a loan to 100 people, then 88 in group A and 11 in group B would get the loan offer (both having 11% loan offer rate).
​
Mind you, this isn’t the only possible or useful concept for reducing discrimination. Other useful ones group-unaware and demographic parity. A group-unaware algorithm discards the group variable, and uses the same threshold for both groups. But for loans, depending on the group distributions, this might lead to one group getting less loan offers. A demographic parity algorithm, on the other hand, looks at how many loans each group gets. In the case of loans, this would be quite silly, but the concept might be more useful when allocating representatives for groups, because you might want each group to have the same number of representatives, for example.
Anyway, there’s a really neat interactive graphic about these, I recommend you to check it out. You can find it here.

Okay, so I promised to reveal my own results for the last week’s Rationality test, and also take a deeper look at the questions while I’m at it. So here goes.

You might guess that – based on the fact that rationality is a big theme of this blog – I would receive the score “Rationalist”. Well, you’d be half right. When I was trying out the beta version of the test (with slightly different questions I think), I got “Skeptic”. Also, the more rationalist result of the second try was a lot due to the fact that for some questions I knew what I was supposed to answer. I guess this shows there’s still room for improvement. Anyway, so what are you supposed to answer, and why? I’ll go through the test question-by-question, providing short explanations and links to relevant themes or theories. At the end, I’ll show how the questions relate to the skills and sub-categories.

Question-by-question analysis

1. LA movieWell, this is just a classic case of sunk costs. You’ve invested time and money into something, but it’s not bringing the utility (or profit, or whatever) that you thought. If you abandon it, you abandon all chance of getting any utility out from the investment. However, as far as rationality is concerned, past costs are not relevant, since you can’t affect them anymore. The only thing that matters is the opportunity cost: should you stay in the movie, or rather do something else. If that something else brings more benefits, rationally you should be doing that.

2. Music or jobThis problem is a classic case of how to confuse your opponents in debates. You can see this in politics, like suppose you’re talking about school shootings with a hard-line conservative: “Either we give guns to teachers or we lower the right to bear arms to 8 years old!”. Well, of course you see right away that you’re being presented a false dichotomy: there are many other ways to prevent shootings – like banning all arms altogether. But to skew the debate and try to put you in a position in which you have to accept something you don’t like, your opponent tries to lure you into the these-are-the-only-two-options trap.

3. Doughnut-making machineNow, this question is basically just simple arithmetic. However, the trick here is that the answer that immediately comes to mind is incorrect, ie. a false System 1 response. Instead, what you need to do is to mentally check that number, see it is wrong, and use System 2 to provide the right answer. The question itself is just a rephrase of one question in the classic Cognitive Reflection Test.

4. Fixating on improbable frightening possibilities
I’m a little puzzled about this question. Sure, I understand the point is that if you’re always fixating on really unlikely bad things, you’re doing something wrong. Still, I find it hard to see anyone would actually be like this!

5. The dead fireman
Now, the point in this question was to see how many possible causes you would think of before deciding on one. The idea is, naturally enough, confirmation bias. We’re too often thinking of a certain explanation, and then immediately jumping to look for confirming evidence. In complex problems, this is a special problem since as we all know, if you torture the numbers enough with statistics, you can make them confess to anything.

6. Things take longer
Well, I presume this simple self-report question is just measuring your susceptibility to the planning fallacy.

7. Bacteria dish
This question has the same idea as the Doughnut-making machine. You get a System 1 answer, suppress it, and (hopefully) answer correctly with System 2. This question is also a rephrase of a question from the Cognitive Reflection Test.

8. Refuting arguments
Being able to argue against other people is a clear sign of rhetorical skills and logical thinking.

9. Budgeting time for the essay
This question checks the planning fallacy. Often, we’re way too optimistic about the time that it takes to complete a project. For example, I’m still writing a paper I expected to be ready for submission in May! In this question, you were awarded full points for assigning at least 3 weeks for the essay, ie. the average of the previous essay completion times.

10. Learning from the past
This is a simple no-trick question that honestly asks whether you learn from your mistakes. I honestly answered that I often do, but sometimes I end up repeating mistakes.

11. BacoNation sales and the ad campaign
This checked your ability to use statistical reasoning. True enough, sales have risen compared to the previous month, but all in all the sales have varied enough to make it plausible that the ad campaign had no effect. In fact, if you pnorm(44.5, mean(data), sd(data)), you get 0.12167, which implies that it’s plausible the September number comes from the same normal distribution. This makes the effect of the ads only somewhat likely.

12. Sitting in the train
So this is the first of the two questions that check how much you value your time. Of course, the point here is that you ought to be consistent. Unfortunately, there may be valid arguments for claiming that you value time on holiday and at home differently, due to differing opportunity costs. See question 20 below for more explanation.

13. Value of time
This question simply asks whether you find it easy or difficult to value your time. Unsurprisingly, the easier you find it the higher your points.

14. One or two meals
Would you rather have one meal now or two meals in a year? This is measuring the discounting of time. Assuming that you’re not starved of food, you presumably should discount meals in the same way as money, since money can obviously buy you meals. See question 21 below for a longer explanation.

15. Continue or quit
Another one of those self-report questions, this is basically asking whether you have fallen into the sunk cost trap.

16. 45 or 90
Here’s another question about time discounting, this time with money. The same assumptions hold as before: we’re assuming you are not in desperate need of money. If that holds, you should discount the same way over all time ranges.

17. Certainty of theory
Can a theory be certain? If you’re a Bayesian (and why wouldn’t you be, right?), you can never set a theory to be 100% certain (let’s ignore tautologies and contradictions here). In a Bayesian framework this would mean that no matter what evidence you observe, the theory can never be proven wrong, because a prior of 1 discounts any evidence for or against it.

18. 100 vs 200
Another discounting question, this time with slightly different amounts of money. Once again, you should discount the same way and choose whatever you chose before. Note that here we are also assuming that 100/200 amounts are close enough to the 40/90 decision – if we had amounts in the millions, that might make a lot of impact.

19. Big assignment
The big assignment vs. small assignments is just a self-report measure to investigate your planning skills.

20. Paying for a task
This question is a sister question to the one where you’re sitting in the train. I presume that the point is that your valuation of one hour should be the same in both question. However, we can question whether the situations are really the same. In one, you’re one holiday, and sitting in a train in a new city has positive value for me. What’s more, on holiday the opportunity costs are different. I’m not really trading off time for working hours, because the point of the holiday mindset is precisely setting aside the possibility of work, so I can enjoy whatever I’m doing – like sitting in a train in a new city. In this question, you’re trying to avoid a task at home, where the opportunity costs of one hour may certainly be different than when you’re on holiday. For example, if you have a job you can do from home, you could be working, or going out with friends, etc.

21. 45 or 90
Well, this is of course part of the other time discounting questions. Here we have the same 45/90 amounts, but the time has been shifted for one year to the future. Again, you should choose whatever you chose before.

All these questions had the similar format:
A dollars in time t vs. B dollars in time t+T

If you’re perfectly rational, you should discount in the same way between times [now, 3 months] and [1 year, 1 year 3 months]. The reason is quite simple: if you’re now willing to wait for the extra three months but not when the lower amount is immediate, you will in the future end up changing your decision. And, if you already know you will change it, why wouldn’t you choose that option already. Hence, you should be consistent. (if you really need an academic citation, here is a good place to start)
 
The score
 So how do these questions make up your score?
If you look at the URL of your results report, you probably see something like
https://www.guidedtrack.com/programs/1q59zh4/run?normalpoints=33&sunkcost=4&planning=3&explanationfreeze=3&probabilistic=4&rhetorical=4&analyzer=3&timemoney=4&intuition=14&future=14&numbers=16&evidence=14&csr=8&enjoyment=0&reportshare=0
You can use that to look at your score by category, for example in my case:

That’s all for this week, happy rationality (or rationalizing?) to all of you! :)

As a decision scholar, I’m a firm believer in the benefits of specialization. If someone is really good at doing something, then it’s often better to rely on them in that issue, and focus efforts towards where you’re personally the most beneficial to others and society at large. Of course, this principle has to apply over all agents – including myself. With that in mind, I’m going to make a feature post about something a certain someone else does – and does it much better than me.

Enter Spencer Greenberg. I’ve talked to Spencer over email a couple of times, and he’s really a great and enthusiastic guy. But that’s not the point. The point is that he does a great service to the community by producing awesome tests, which you can use to educate yourself, your partner or anyone you come across about good decision making. What’s even better is that the tests are done with the right kind of mindset: they’re well backed up by actual, hard science. What this means is that the questions make sense – there’s none of that newspaper-clickbait “find your totem animal” kind of stuff. There’s proper, science-backed measuring. Even better, the tests have been written in a way anyone can understand. You don’t need to be a book-loving nerdy scholar to gain some insights!
​
Now, I’ve always wanted to bring something to the world community. And a while ago, I thought maybe I could produce some online tests about decision making. But after seeing these tests, I’ll just tip my hat and say that it’s been done way better than I ever could have! Congrats!
And now, enough of the babbling: go here to test yourself! (For comparison, a reflection of my results can be seen in next week’s post :)

The past few days I’ve been spending at INFORMS Annual Meeting at the beautiful San Francisco. Due to the planning fallacy, I never had the time to write something new and proper for this week - so you if you’ll excuse me, I’ll just use this week’s post to share a few interesting insights from the talks I’ve seen here during the past few days.

Statistics.com

I’ve never heard of them before, but apparently these guys offer stats courses on crisply defined domains. Each course costs a small amount of money, and lasts for four weeks. So completing one of them may be a good idea if you want to know about one specific area, rendering full-semester MOOCs somewhat useless.

Cost Transparency Increases Purchase Intention

Bhavya Mohan, from Harvard Business School, showed in their group’s experiments that – at least currently – a company can increase customers’ willingness to buy in an online store by providing information on the product’s costs.

Emotional-Motivational Responses Predicting Choices

Outi Somervuori, a colleague of mine from Aalto University, showed that negative emotions and frontal lobe asymmetry predict the endowment effect.

My Own Talk

My own presentation (available here) was about using a linear value function model to predict choices when choosing student apartments. As it turns out, whether a subject is consistent with a linear value function had very limited impact on how well the model can predict choices – meaning that assuming a linear value function is an ok starting point for a model.

ProbabilityManagement.org

These guys have a fantastic, completely free SIPmath standard for communicating uncertainty between Excel and different pieces of software. It for example enables easy calculations with distributions in native Excel, something that’s notoriously difficult. For example, in the screenshot I’ve just calculated U+U, U*U, and U^cos(U) with  a sample of 10 000 for the uniform distribution. I think this system is nifty enough to maybe warrant a full post in the future.

Decision Analysis in Life and Business

Larry Neal told about his personal experience on how business and life decisions differ. A decision analyst at Chevron (one of the most decision analysis –friendly companies out there), he faced cancer a few years ago, and ended up helping several other patients to analyze and decide on their treatment. He concluded that DA in personal decisions is much harder, because the goals are more ambiguous and the situation much more complex. Moreover, in his view our behavior is heavily influenced by cultural assumptions, and it is often impossible to recognize them yourself – but that’s exactly something as an analyst he was able to help with.

MOOCs

A panel by four professors from Stanford, MIT and Columbia was talking about MOOCs. Their response was very favorable. They all agreed that the traditional way of teaching with a professor speaking and the students listening is not adequate. MOOCs can be used to deliver value to a large audience, and even though the completion rate is low (less than 10 %), they still reach thousands of students per course.

Teaching Decision Analysis

There has been several amazing sessions about Decision Analysis. I just came back from a session about teaching and the legacy of Ron Howard, one of the giants of DA. It was certainly very inspiring and there were some good points about how DA teaching need to stay connected to the actual practice of decision making. After all, we don’t want to produce just new researchers who produce papers – the point of research is to impact the world in a larger way.

Well, that's a summary of the past couple of very exciting days! I'll be back on track with more decision making thoughts next week.