Why you need Rebel Ideas within your Data Science team
Building on our continued focus on getting work delivered, what’s the role of rebel ideas?
Many miles of copy have been written on the advantages of diverse teams. But all too often this thinking is only skin deep. That is it focusses on racial, gender & sexual orientation diversity.
There can be a lot more benefit in having team members who actually think differently. What is called Cognitive Diversity. I’ve seen that in both the teams I’ve lead and at my clients’ offices. So, when regular guest blogger Harry Powell approached me with his latest book review, I was sold.
You may recall that Harry is Director of Data & Analytics at Jaguar Land Rover. He has blogged for us previously on the Productivity Puzzle and an Alan Turing lecture, amongst other topics. So, over to Harry to share what he has learnt about the importance of this type of diversity.
Reading about Rebel Ideas
I have just finished reading “Rebel Ideas” by Matthew Syed. It’s not a long book, and hardly highbrow (anecdotes about 9/11 and climbing Everest, you know the kind of thing) but it made me think a lot about my team and my company.
It’s a book about cognitive diversity in teams. To be clear that’s not the same thing as demographic diversity, which is about making sure that your team is representative of the population from which it is drawn. It’s about how the people in your team think.
Syed’s basic point is that if you build a team of people who share similar perspectives and approaches the best possible result will be limited by the capability of the brightest person. This is because any diversity of thought that exists will essentially overlap. Everyone will think the same way.
But if your team comprises people who approach problems differently, there is a good chance that your final result will incorporate the best bits of everyone’s ideas, so the worst possible result will be that of the brightest person, and will it normally end up being a lot better. This is because the ideas will overlap less, and so complement each other (see note below).
Reflections on why this is a good idea
In theory, I agree with this idea. Here are a few reflections:
- The implication is that it might be better to recruit people with diverse perspectives and social skills than to simply look for the best and brightest. Obviously bright, diverse and social is the ideal.
- Often a lack of diversity will not manifest itself so much in the solutions to the questions posed, but in the selection or framing of the problems themselves.
- Committees of like-minded people not only water down ideas, they create the illusion of a limited set of feasible set of problems and solutions, which is likely to reduce the confidence of lateral thinkers to speak up.
- Strong hierarchies and imperious personalities can be very effective in driving efficient responses to simple situations. But when problems are complex and multi-dimensional, these personalities can force through simplistic solutions with disastrous results.
- Often innovation is driven not simply by the lone genius who comes up with a whole new idea, but by combining existing technologies in new ways. These new “recombinant” ideas come together when teams are connected to disparate sets of ideas.
All this points towards the benefits of having teams made up of people who think differently about the world. But it poses other questions.
Context guides the diversity you need
What kinds of diversity are pertinent to a given situation?
For example, if you are designing consumer goods, say mobile phones, you probably want a cross-section of ages and gender, given that different ages and genders may use those phones differently: My kids want to use games apps, but I just want email; My wife has smaller hands than me, etc.
But what about other dimensions like race, or sexual preference? Are those dimensions important when designing a phone? You would have thought that the dimension of diversity you need may relate to the problem you are trying to solve.
On the other hand, it seems that the most important point of cognitive diversity is that it makes the whole team aware of their own bounded perspectives, that there may be questions that remain to be asked, even if the demographic makeup of your team does not necessarily span wide enough to both pose and solve issues (that’s what market research is for).
So, perhaps it doesn’t strictly matter if your team’s diversity is related to the problem space. Just a mixture of approaches can be valuable in itself.
How can you identify Cognitive Diversity?
Thinking differently is harder to observe than demographic diversity. Is it possible to select for the former without resorting to selecting on the latter?
Often processes to ensure demographic diversity, such as standardised tests and scorecards in recruitment processes, promote conformity of thought and work against cognitive diversity. And processes to measure cognitive diversity directly (such as aptitude tests) are more contextual than are commonly admitted and may stifle a broader equality agenda.
In other words, is it possible to advance both cognitive and demographic diversity with the same process?
Even if you could identify different thinkers, what proportion of cognitive diversity can you tolerate in an organisation that needs to get things done?
I guess the answer is the proportion of your business that is complex and uncertain, although a key trait of non-diverse businesses is that their self-assessment of their need for new ideas will be limited by their own lack of perspective. And how can you reward divergent thinkers?
Much of what they do may be seen as disruptive and unproductive. Your most obviously productive people may be your least original, but they get things done.
What do I do in my team?
What do I do in my team?
For data scientists, you need to test a number of skills at interview. They need to be able to think about a business problem, they need to understand mathematical methodologies, and they need to be able to code. There’s not a lot of time left for assessing originality or diversity of thought.
So what I do is make the questions slightly open-ended, maybe a bit unconventional, certainly without an obviously correct answer.
I expect them to get the questions a bit wrong. And then I see how they respond to interventions. Whether they take those ideas and play with them, see if they can use them to solve the problem. It’s not quite the same as seeking out diversity, but it does identify people who can co-exist with different thinkers; people who are open to new ways of thinking and try to respond positively.
And then try to keep a quota for oddballs. You can only have a few of them, and they’ll drive you nuts, but you’ll never regret it.
EndNote: The statistical appeal of this Rebel Ideas
Note: This idea appeals to me because it has a nice machine learning analogue to it. In a regression you want your information sets to be different, ideally orthogonal. If your data is collinear, you may as well have just one regressor.
Equally, ensembles of low performing but different models often give better results than a single high-performing model.