Everday Integrity
Consider these scenarios:
- You are a junior data scientist who has been asked for a follow-up to a previous project. Your boss liked your work so much, he wants more. As you look over your earlier results, you notice something. You pull up your code. Could it be? Yes, there is an error here…. You have a sinking sensation. Right now, everyone is happy. If you don’t say anything, no one will ever know.
- Your phone rings. It’s a customer. They are concerned about a metric you provided. Is it possible it was wrong? You are confused. Also, resentful; after all, you’ve gone the extra mile for this client, and received scant appreciation.
- It’s the one-year anniversary of the launch of a new initiative, and you’ve been asked to compile report on its performance. This is your boss’s pet project, which has generated a lot of attention. Unfortunately, you find that the effort has been less successful than hoped. You check and recheck your results. Your boss is already annoyed with you for a mistake you made last time. Plus, you’ve had to take a lot of time off lately, and she’s commented on it.
I’ve spent nearly a decade in analytics and data science, mostly in junior or mid-level individual contributor roles. Firms I’ve interacted with have ranged from just a few employees to tens of thousands. I’ve been part of companies focused entirely on analytics, as well as part of a small team within a larger organization. I’ve faced ethical dilemmas and seen others do the same. Some situations have been almost humorous, others devastating.
Any data scientist working for any length of time is going to find a mistake in their own work, which no one else will ever know about if they don’t say anything
Any data scientist (or analyst, etc.) working for any length of time is going to find a mistake in their own work, which no one else will ever know about if they don’t say anything. I’ve faced that situation many times. I’ve also had others ask questions about my results, which I could maybe have talked my way out of. As many times as I’ve faced these kinds of situations, they are always painful and embarrassing. However, I am fortunate to have learned a few techniques to help (discussed later). When I’ve applied them and, for example, told a client about an error that would not have been discovered otherwise, I’ve had very good experiences. I’ve even been offered jobs! Never have I lost a project or had my overall competence seriously questioned. I feel that, overall, clients have appreciated disclosure and trust has increased.
Some circumstances have been much more difficult. Although I am fortunate now to work for a great company, I have had some negative experiences. On a couple occasions, I’ve witnessed people in positions of leadership lying. Suddenly, I’ve had to reckon with choices I didn’t make or even see coming. Although the specific falsehoods didn’t seem very significant or dangerous, these situations were frightening and degrading. In one case, I was personally visited by a leader to secure my compliance and praised for my "good discretion" so far.
I’ve thought a lot about these events. How could I be so easily intimidated? Is leaving a job enough? Sometimes entire rooms full of people have kept silence. I think of younger people, just starting in the workforce, immersed in those environments. Do they think this is normal? Is it normal?
AI Ethics and AI Integrity
I once asked an Ai Ethics panel about best practices for dealing with mistakes. One participant replied (transcribed from the video):
And so what we realized was the results that [we] had given to a very large company were just totally wrong. And the question was what do we do? And the person running the team at that point said, well, we already delivered those results a couple months ago, they already set their plan for the year based on those results, so there’s nothing that really be done at this point, so we’re just not going to tell them, we’ll just fix the code moving forward. I’ve never really agreed with that decision, but … you can understand how much pain that would have caused, to go to a major client and say all the models that you based your yearly plan on are bogus.
I found it interesting that this ethics panel focused entirely on issues such as privacy and algorithmic bias, and didn’t address integrity. The above only came up because I asked about it, and after a noncommittal discussion, the conversation shifted back to the original topics. But the response at least confirmed that my experiences aren’t unique.
I found it interesting that this ethics panel focused entirely on issues such as privacy and algorithmic bias, and didn’t address integrity
AI ethics topics typically discussed at conferences are urgently important and need to be addressed. However, for those of us working in "regular" jobs, integrity is another crucial discussion.
In this column, I reflect on ethical dilemmas I’ve faced, focusing on common situations where I believe falsehoods tend to occur. For this first installment, I focus on preparation. I believe that a major obstacle to integrity is a failure to plan for common situations.
The Problem of Planlessness
Reactive lying is more common than premeditated deception. People are going about their day, doing their jobs, thinking about their own business, and out of the blue comes an ethical dilemma!
If I told you that mistakes are inevitable, you would of course agree. Everyone makes mistakes! Everyone knows this! However, I have found that few organizations have processes in place to address these predictable occurrences. Individuals are also caught off guard. So, what tends to happen is that people react from emotion, like short-haired kindergartners caught holding scissors. Nothing happened! The dog did it! Or, we freeze and wait. Or, we get defensive. I’m not wrong! You are the one who’s wrong!
Ethical dilemmas are not unique to data scientists. We are not the first to be tempted to lie to save face or appease others. However, we have special vulnerabilities. We deal in information and influence decisions. Our work depends on the goodwill of our clients and society at large.
In my opinion, we can’t afford even a small amount of dishonesty. When I’ve witnessed or suspected dishonesty, it hasn’t seemed to go well for those involved. A client may never be aware of a falsehood, but still becomes uneasy with the relationship. Often, dishonesty is just one manifestation of underlying attitudes that tend to reduce trust (I will discuss some of these in a future essay).
Too often, we just don’t have a plan. Issues of integrity are absent from public discussion and many organizations don’t have clear policies. The burden often falls on individual workers to do the right thing.
Practical Tips for Individuals
In my initial examples, I included details like how much PTO you’d recently taken or how demanding a customer might be. If I just wrote my values down on a piece of paper, I wouldn’t include such minutiae. But our ability to execute on values can change radically depending on our environment.
A plan isn’t just values. Without a plan, everyone is prone to falling short of their principles. A plan must include strategies and tools that can be used in the moment to help us achieve our values.
Below, I share some strategies that hopefully can help reduce the effects of context. The tips below are (mostly) aimed at a generally honest person who would prefer to do the right thing. Your plan will be based on your own values, personalities, and situation.
Please post additional ideas in the comments!
Phone A Friend
Time is not on the side of honesty. The longer you wait, the less likely disclosure becomes. Someone like me whose stress response is to freeze is especially prone to waiting too long. However, immediately telling a supervisor or a client can be way too daunting. What’s much easier is to have a friend, relative, etc., you can call, text or email right away. The best person is someone nonjudgmental, trustworthy, and levelheaded. It’s great if they live far away or have little interest in data science.
Time is not on the side of honesty
A long time ago, I was working in a lab and dropped an expensive lens. I was alone and it was late at night. I started fretting about the cost, put the part back on the shelf, waffled. It occurred to me to call a particular person. As the lab was underground, I had to walk quite a way to get cell signal. At first, I had all sorts of things in my head — maybe the part was fine, I shouldn’t be asked to work this late anyway, these German optics companies make things too slippery. As I kept walking, I imagined telling my friend all these things. Then, I stopped and turned back. As clearly as if she’d been in the corridor next to me, I’d heard my friend’s voice, and none of my nonsense made sense anymore. I returned to the lab, emailed the lab manger, and went home to bed.
I have found that informing another person can be an instant relief. It lifts away emotion, allowing clearer thinking. Plus, your friend isn’t affected by your annoying client or distracted by the fact that you haven’t had lunch yet.
Put A Dollar Value on It
Most lies I’ve witnessed haven’t seemed to be major. They’ve been more ambiguous, and unsurprisingly I’ve often heard things like, "the error is small, it’s not worth saying anything". Or, "no one is using it anyway".
But isn’t there something disturbing about, "I am going to decide what is meaningful to you"?
But isn’t there something disturbing about, "I am going to decide what is meaningful to you"? We often don’t have complete knowledge of what our clients are doing with our products, or what they might do later. What assumptions or prejudices might a client carry into the future based on our numbers, even if they do not drive decisions today?
There is an obvious conflict of interest when the person who might be embarrassed is the one to determine whether a mistake meets some threshold for disclosure. Notably, this threshold is not set ahead of time, or pre-arranged with the client, but is determined on the fly by an emotional person who just made an error.
Additionally, easier difficult conversations are an opportunity to practice your plan. Do we really believe that a person who avoids addressing a small error will be capable of stepping up to the plate in a crisis? Integrity isn’t a finite substance that must be saved up for when it really matters.
So, I propose a simple calculation. First, estimate the dollar value to yourself of not being honest. This can be just a gut feeling, or an hourly rate for correcting and discussing a mistake. Reflect on that number a minute.
If you still have any doubts, compare this value either to the value you place on your personal integrity, or to a risk value equal to the probability of getting caught times the cost to yourself if discovered.
Consider an example like the one quoted earlier, where erroneous results had been given to a client. A leader makes the disclosure decision, but an entire team is aware. If the vendor is honest with the client, in the worst-case scenario, they get fired. Let’s say that’s worth $1 million. But what is the potential cost of attempting secrecy? If it becomes public that a firm knew about an error and didn’t correct it, they might lose all their clients now and in the future. Maybe $25 million dollars? What are the odds of the falsehood remaining secret as time passes and the employees go onto other positions? For dishonesty to make financial sense, the probability must be below 4%, which doesn’t sound like a good bet. And 4% is probably an overestimate, considering the potential losses due to reduced staff morale and attrition.
This dollar value technique might sound questionable or even shocking. It’s totally self-interested. Shouldn’t we want to do the right thing because it’s the right thing to do? Or out of genuine empathy towards those who might be affected by our actions?
For a generally honest person, this method is intended to engage the analytical mind and interrupt panic, not to actually weigh dollar values. A dollar value helps put a situation in perspective — maybe it isn’t end of the world after all!
For those of below average integrity, the dollar value technique might at least put a lower bound on lying. I’ve seen people lie when honesty would have cost one or two difficult conversations or a few hours’ work. A floor on dishonesty might seem like a small gain, but many issues arise from everyday situations; moral dilemmas are right skewed.
Get To Know Them
The previous tips were reactive, to be used in a moment of crisis. This next strategy is preventative, and simple: As much as possible, get to know the people who are affected by your work. Do this long before an issue comes up.
I am not suggesting talking more to the person paying you. The goal is to meet people potentially affected by your model or analysis. For example, if your work involves building prospect models for a sales team, maybe spend a day shadowing a salesperson so you can see what their day really is like. Or, you might ask to sit in on focus groups with customers.
The immediate goal is to increase empathy and reduce our ability to tell ourselves that "it’s no big deal" or "the damage is done". But there are myriad benefits. Observing people who generate data and/or use our models or Analytics is eye-opening. I’ve gained a lot of insight into the reliability and meaningfulness data while, for example, watching a user click past a ridiculously long list of options to select "other" every time.
For individual contributors who spend a lot of time heads-down, I think that this sort of experience can be a refreshing change and increase engagement. Before the pandemic, I was fortunate to get to spend a day shadowing a service position. I was amazed by the complexity of the job, as well as the patience and kindness of the clients. I felt more motivated in my work after that, and especially wanted to make sure any projects I worked on would make that job easier, rather than adding additional hassle.
A great employer will provide these kinds of opportunities, but in many places an individual data scientist might have to make them happen. I think it’s worthwhile to make the effort — even if you arrange one meeting a year, that’s more often than most! It may be inappropriate or impossible to talk to the actual people affected, but you may be able to find someone closer (e.g. teachers if the data involves students).
Redefine A Successful Relationship
Some data scientists define project success as "The client (or boss) is happy and I am happy". Happiness is great, but this framework starts to break down when mistakes occur. Admission of an error is something that probably will make both the client and data scientist unhappy.
This definition also tends to focus on the immediate client — the person approving the project or paying the data scientist. Those not in that room tend to recede. Sometimes, people like end users, data subjects, etc., might as well live on the moon.
Sometimes, people like end users, data subjects, etc., might as well live on the moon
It’s nice to make clients happy, and if the person paying you is never pleased, there’s obviously a problem. But we might crack open the door to the possibility that a project can still be successful, even if the client is unhappy.
Love bombing clients may not even be the best way to win their regard. Focusing too much on client happiness can sometimes lead to an intense honeymoon period, followed by disappointment. I think that a desire to do excellent work, a focus on client success rather than happiness, or just a simple good faith effort, often pleases them more long term.
Even if a mistake is irreparable, and the client can’t do anything about it except to feel unhappy and disillusioned with you, I think they should be informed. Learning from misfortune and mistakes is an essential part of the human experience, and it’s not right to knowingly interfere in this process. Perhaps because disclosing errors is an inherently respectful act, clients often appreciate the effort.
Final Thoughts
Whatever your ideals, living up to them tomorrow means effort today. As Eliyahu Goldratt said, "bad luck is when lack of preparation meets reality."
If you enjoyed this article, stay tuned for future installments of "Everyday Integrity", coming soon!