Podcast: Play in new window | Download
Episode 14: Ethical Obligations of a Data Translator – Show Notes
Data Translator is a new title coming up in the business world over the last few years. This role is an intermediary between those requesting data science work and the data scientists. It’s sort of like a business analyst but for analytics projects.
To be honest, my first thought when I saw the articles about this job, was flash backs to Office Space. Specifically…
Really, though, this is a crucial role for those looking to data science for answers to get useful answers the first time around. It’s also helpful for the data scientists in telling the story that their hard work has uncovered.
Since this person is translating and story-telling and interpreting, what ethical burden does the data translator take on for the business? Where can they best help ensure ethical practices. Join us in exploring this new position and its ethical obligations for data science.
Additional Links on the Data Translator
Analytics Translator: The new, must-have role McKinsey study following the uptick in data and analytics translator roles in the business world.
Forget data scientists and hire a data translator instead? Forbes article describing the data translator role and the skills needed to fill it.
Multimodal Explanations: Justifying Decisions and Pointing to the Evidence White paper from Park, Hendricks, et. al. displaying techniques used for image recognition algorithms to explain the reason for their object detection or classification.
Episode TranscriptView Episode Transcript
Marie: Hello, this is Marie with the Data Science Ethics Podcast. I’m here with Lexy and we’re going to talk about something we’ve been seeing on LinkedIn and a few different articles. That’s the idea that there is a need for a role in data science of somebody that is basically monitoring all of the different algorithms and being that bridge between the business user and the data science team. That would be the data translators, so when we look at a data translator, Lexi, what are you seeing in terms of how that impacts the team?
Lexy: The translator is someone who doesn’t necessarily always go in and start coding and algorithm, but they’re really great at setting up the problem for another data scientist to take and run with and what I see is happening is as business people hear about and see all the possibilities with machine learning and data science, they want more, but they don’t quite know how to convey what they want to the data scientists
Marie: Or even the request of, “Hey, we have this data. Do something with it.”
Lexy: Yeah. Oh all the time. Somebody is the person that is the arbiter essentially of what is and isn’t possible without interrupting the flow of development for the data scientists. Trying to help the business prioritize the requests, helping them to fully vet what they want, what they expect the algorithm to do so that when you go to start the actual data science work, there’s a much better sense of where you’re going with it and what you need to do with it so that you don’t have this back and forth. It gets very frustrating kind of on both sides to the business and the data scientists. If those questions can be asked upfront by a data translator, you have much better requirements to go on as a data scientist.
Lexy: The other side of the data translators taking the results, bringing it back to the business and helping them see the benefits of the algorithm, the uses, the specific ways that that algorithm is telling them to change the business.
Marie: It sounds like in general, this is the type of role that is probably already being covered in some aspect on most teams and this is just a way of codifying that need. Saying this is an important role within the team and more people should specialize in this role. It doesn’t have to be necessarily somebody that’s doing the data science also has to wear this hat. They can partner with somebody that wears this hat and can be that bridge.
Lexy: What I’ve found is that there’s often a desire, there’s this appetite for analytics. An organization that doesn’t necessarily have data science in house or doesn’t have a lot of analytics background in house, doesn’t have that function and doesn’t necessarily have the talent internally that will do that function of translating analytics into decisions. Over time, that should be something that every manager becomes to some degree. It won’t always be someone who can go all the way back to the data scientist and tell them, I need you to run this type of model that’s going to give these types of outputs and here’s exactly how the data needs to be tagged and I’m going to work with you on setting up your entire experiment. A manager, senior manager, executive manager should be able to articulate their need for data, their need for an algorithm to be developed over time.
Lexy: As you start getting into a more mature organization. Really working through the process multiple times is how that happens. I think within the organizations that have a more mature structure, it makes sense to have it, like you said, codified into a role that people could aspire to become. There are ways that kind of progression paths into a role like that. It focuses much more heavily on understanding the business and understanding the processes that lead into the business so that you know where you can influence based on the algorithms that you develop to then be able to translate more effectively.
Marie: It sounds like not only that, something that is going to become a role in and of itself that we’ll see more in this space moving forward, but it’s also going to be something, especially at higher levels of organizations that those business leaders are going to have to get more comfortable with and it’s going to be part of their tool belt, not because it needs to be how everybody approaches their job, but it needs to be another way of understanding the business in order to be able to articulate what you need an algorithm to do, what you need the data science teams to work on and making sure that bridged not only is helping the business do what they need to, but that the people that they’re going to be reporting back to ultimately understand their role in the process in terms of how to best define for the team, what problems needed to be defined and best working with the teams to make sure that everybody is efficient and that they’re giving the right direction up front.
Marie: Do you feel like the role of data science translator is going to stay the same even as machine learning and ai change or do you feel like it’s going to evolve alongside it?
Lexy: I think it’ll always be there. The data translator really is there to take what an algorithm is doing and make it make sense for the business, so when we take an algorithm to decision makers and say, we think that based on this algorithm you need to make the following changes. That will always be. Not everyone immediately looks at an algorithm that says, I know exactly what I’m going to do with this. That’s pretty rare actually. It takes an understanding of the business side, so when you look at the traditional drew conway Venn diagram of a data scientist being the mix between hacking skills, which is kind of programming math and stats and business acumen. The data translator is very heavy on the business acumen side. Maybe has some of the math and stats so that they understand how those statistics get leveraged by the business and they can go back and forth between the data scientists and the business in much the same way a business analyst does with data engineering project for a systems engineering process of versus what the business wants and the requirements that they have.
Lexy: There have been some attempts recently to have AIs explain themselves, so that’s intriguing. One of the big qualms people have with AI in particular is that when you get into something like a neural net, which is what a lot of those deep learning algorithms are based on, it’s difficult to peek behind the curtain on what it’s doing and you can’t really peel back the covers and say, this is the algorithm. This is how it figured it out. This is what it’s doing. This is the importance of specific factors and so forth. You kind of have to back into it now because there’s been so many people who are unwilling to trust a black box algorithm. Meaning you put the data in, something happens inside the box and you get something out. A lot of people are not willing to trust that type of system and now because the system is effective, there’ve been a number of groups that are trying to essentially make the algorithm self-documenting and self explaining.
Lexy: There was one that I saw where there was an image and two different AIs fighting about what the image was. One was saying it was a cat that was in front of a wall. The other was saying it was a dog lying on the ground. The algorithms were arguing based on the shape of the ear and the placement of certain lines until one of the algorithms gave in and let the other one went and said, Ah yes, I see your point. Essentially.
Lexy: I could see in the future an argument like that essentially playing out, but for the entire system of a neural net. That said it would be a lot to read. As much as I think the data translator will always be there. I think we’ll get better self-documenting algorithms and systems in the future and that that will be a key differentiator, especially in probably the next five to ten years as these systems become more common and more people want to use them.
Lexy: Where I think this really verges into data science ethics is that you have to rely on that data translator to have the best interest of the business and the ethics that you want within your business at heart. For sure. You have to ensure that you don’t lose what you gained in trying to implement the best ethical practices into that data science process as you can in that translation. The possibility that someone could have an unconscious bias and then project that into their translation back to the business is very real. So for example, if you looked at the outcome of an algorithm and it says something about certain populations and that person looks at those populations, they put their own unconscious filter of what they see in that population into it, and they may position the algorithm or position the outcome of the algorithm to the business in a very specific way. Based on their own biases, and so I think that it’s important in evaluating a data translator in trying to find people who would serve as the role of data translator in your business that can try to be more objective, that can try to be more inclusive, equitable, and understand the broader context of an algorithm so that you don’t end up pigeonholed into a biased view.
Marie: I think that also points to a responsibility as a data translator to then look at how people are implementing an algorithm within the business, making sure that they haven’t misunderstood something so it’s not only the unconscious bias that you might have in how you present it to the business, but also understanding any unconscious bias that the business might have if they accept the algorithm and then take it and run with it internally. It’s really having an intimate understanding of who you’re working with and what they’re doing. Not just from the business side but also the data science team side so that way you make sure that everybody’s on the same page and as you mentioned Lexy implementing the types of ethics that your company has agreed upon.
Lexy: Which also means that you have to have some ethical basis that your company is agreed upon, which is a whole other topic. Like you said, there is a lot of responsibility in that role of data translator to bring the best, most ethical result to bear on the situation and that is a very different ask than solve the problem. It means don’t go for necessarily the highest accuracy, the best statistic of measurement of an algorithm. It means find the best option for the business given the full set of constraints under which we want to operate from an ethical standpoint, give us that, tell us from that perspective what we need to do.
Marie: Or even understanding that sends the context under which all future will be framed. So even if somebody doesn’t say that as part of the ask, you know that that is part of the ass going forward.
Lexy: Thank you so much for joining us as we discuss the data translator and its impact in the business world. This is Lexy and Marie. We’ll catch you next time on the Data Science Ethics Podcast.
Marie: Thanks so much.
We hope you’ve enjoyed listening to this episode of the Data Science Ethics podcast. If you have, please like and subscribe via your favorite podcast App.
Join in the conversation at datascienceethics.com, or on Facebook and Twitter at @DSEthics where we’re discussing model behavior. See you next time.
This podcast is copyright Alexis Kassan. All rights reserved. Music for this podcast is by DJ Shahmoney. Find him on Soundcloud or YouTube as DJShahMoneyBeatz.