Clare Gollnick is an ex-neuroscientist who loves to debunk statistical fallacies. She helps lead a company expert in navigating the dark web and protecting information.
She was recruited to join Terbium Labs by its CEO and founder Danny Rogers in 2016 as a chief data scientist. In 2017, she was promoted to CTO. Her team’s focus is on automating the intelligence cycle, a framework (not unlike the scientific method) for figuring out hard problems, meaning that their focus is even broader than a traditional analytics platform. “We work to minimize the time between when a question is asked and when it is answered,” she says.
Women 2.0 was lucky to get a chance to talk to this articulate, brilliant tech executive recently, and sometimes she sounds as much like a philosopher as she does a technologist.
You developed an algorithm to quantify the strength of relationships in different types of network data structures. Talk about how you design measurements of something so variable and subjective.
Usefulness is more important than objectivity. To be honest, I do not believe in objectivity, anymore. As a young scientist, I desperately wanted to be objective. The reality that the pursuit of knowledge is not objective was a hard reality for me to face.
In graduate school in neuroengineering, the disconnect between what I thought science should be (purely objective) and what I was experiencing (a lot of storytelling and marketing) led me to study epistemology and the history of scientific discovery. Through this, I realized that data analysis cannot be objective. One always starts with an assumed model of how the world works before learning from data. In some circumstances, this is called as bias (though that is a bit unfair).
Through the study of empiricism, I learned that the scientific method is objective on the time scale of generations, not an individual scientist’s work. This isn’t failing of individual humans, it is actually true because our current systems of logic, induction/inference.
That’s a long way of saying: All models are wrong. Many are useful. When you work in career in data, pursue usefulness not truth. It is a much more practical approach.
Your company helps people protect their most important data. What is our most important data and how well do you think individuals and businesses assess the importance of different types of data?
This is a great question. Terbium’s patented data fingerprinting technology allows us monitor and detect data leakage and evidence of data breach without putting our customer’s sensitive data at additional risk. It is designed to prevent the type of massive data breach we saw with Equifax (among many others). The technology is based on a technique called fuzzy hashing. One of the things that makes Terbium’s solution unique is that our methodology is entirely transparent. No secrecy. No tricks. It just works.
As far as which data is important, that’s a ‘holy grail’ type of question. The value of a dataset depends on why people want to use it. The data can be the same as it was five years ago, yet because new ways of making it useful are developed, it suddenly becomes a much more valuable asset. Interestingly, this can be true of all business and personal assets. It is possible to make careful risk management decisions by thinking of data as an asset of unknown future worth.
Tell us about your team and what are its best qualities, biggest current strengths.
My team is tackling a problem with immense ambiguity and few known solutions. The questions I ask of them, and the requirements given to them, are more open to interpretation than in most engineering and data science teams. That’s what happens when your technology is designed to find answers that are not yet known. Excelling in engineering and data science even with imperfect knowledge is something that makes my team exceptional and unique.
Define “dark web.”
Dark Web: the part of the internet you cannot find using a google search or get to through a normal browser. (The definition is a work in progress)
Terbium’s crawler technology focuses on the Tor Network and the hidden services that run on it because much of the underground economy where data is traded appears there. Our search scope is much broader than just the Dark Web in order to serve our customers well.
Here in New York State, we have a congressional candidate who is a data security expert, and she has pointed out that there’s currently very little data security knowledge in congress. Do you agree, or is it even important that our representatives understand how data security works?
Representative should be experts on government and policy first, and any other type of expertise is a welcomed bonus. Political expertise is the only way anything gets done in such an immense bureaucracy. Technical expertise can be incorporated into policy by other means – asking domain experts as the bills come up through the system. Having switched fields a lot in my career, I value expertise on many topics: medicine, public health, data integrity, cybersecurity, law etc. It is unreasonable to expect our representatives to be experts on everything. Instead, I look for someone who knows how to make government work and who I trust to evaluate experts, ideas and evidence on many topics.
Can you describe your move from biomedical engineering to corporate data science?
Honestly, it was painful. It was one of the most difficult times in my life. Finding a job was hard. Hiring managers did not seem to believe that my skills were transferable from neuroengineering to data science. Now, I try hard to consider people from non-traditional backgrounds or different skills for all of Terbium’s open positions. The opportunity to give someone a chance no one else will is my favorite perk of my current position.
What questions do you wish I’d asked?
My favorite topic is on the limits of inference – how and why data fails. I mentioned it a little, but I could talk about it for hours.