A 5by5 conversation with Mitchell Bobman, Analyst at Gensler about how is access changing based on algorithms and what can we do to get it right?
Interview by Thomas Brandenburg and Twisha Shah-Brandenburg
Companies and institutions have spent the past decade hailing the concept of ‘big data’ – always assuming more data is a good thing and hiring talent to help monitor it. Now that we have all of this data, and a greater computational ability to analyze it, there’s an emerging need to simplify and make sense of it all. —Michell Bobman
Is it possible to create a completely unbiased algorithm?
As we’ve been starting to feel the long-term effects of how algorithms have been influencing things like public and private policies, it’s becoming more apparent that embedding bias within algorithms is nearly unavoidable.
With that said, it doesn’t make algorithms some uniquely flawed tool we use to make decisions, bias is embedded in nearly every stage of the decision-making. The charge, rather, is coming to an understanding that these algorithms aren’t end-to-end sources of truths and we should be extremely critical of them.
What is the role of diversity in the planning and creation process of algorithms?
Given the role that bias can play in the creation of algorithms, diversity helps to mitigate this bias. In a computational process, there can be a number of different ways to get from A to B. The more ways that are accounted for, the deeper the level of understanding that can be embedded in an algorithm or computational design.
Blind orchestra auditions were introduced to keep biases in check so that the focus is on the music and not the demographic information that can make the decision-making process subjective. What might we learn from this as we design algorithms that make decisions?
Flying blind vs. knowing too much can be a fine line to walk. So if you have a dataset with numerous variables, how do you determine which are extraneous? While a common early stage in designing algorithms is cleaning and structuring the data, an initial data analysis can often be expedited or overlooked. If variables within a dataset are confounding, and some of them are known to induce demographic bias, then you can start to simplify the algorithm down to the subjective essentials.
What are the long-term effects of feedback loops? How might data scientists/designers and engineers think about and monitor their data?
I tend to think of it more as a snowball effect rather than a feedback loop. When a protocol or algorithm is developed, there tends to be an ‘if it ain’t broke, don’t fix it’ approach if it fulfills all of its short-term requirements. This can range from something as simple as how a company counts how many employees it has or something as complex as how it might deploy algorithms to hire and fire them. For a small company with relatively manageable data, this might not be cause for concern. But when you’re a Fortune 100 company, this bad data management can snowball into major efficiencies and negative consequences. With that in mind, keeping the now, near, and far in mind can help anticipate and mitigate negative long-term effects.
What is the future of data science? What signals are you looking at that is making you excited and worried?
Companies and institutions have spent the past decade hailing the concept of ‘big data’ – always assuming more data is a good thing and hiring talent to help monitor it. Now that we have all of this data, and a greater computational ability to analyze it, there’s an emerging need to simplify and make sense of it all.
As computational power and digital literacy continue to increase throughout the workforce, I think we’re moving towards a future that relies less on data science specialists and more on generalists who can turn the data into strategy and insights.
See Related 5 x 5s