We interviewed TrojAI founders, James Stewart, CEO, and Stephen Goddard, COO, on how their careers, interests, and relationships led them to build their company. TrojAI develops solutions to protect artificial intelligence platforms from adversarial attacks on training data and AI models. This year, TrojAI was named one of the 100 most promising AI startups in the world by CB Insights, a New York-based startup intelligence group.
Tell us about TrojAI and how the concept for the company evolved.
James: The idea evolved from my first tech startup, EhEye. We used computer vision to automate video surveillance monitoring for the early detection of guns and fights, inspired by the Parliament Hill attack in 2014. The company was acquired by a publicly traded defense company in 2018 where I served for two years as SVP of video analytics. It was during that time I became aware of attack vectors around artificial intelligence and that AI is actually hackable. Stephen, who was board chair of EhEye at the time, encouraged me to leave the safety of my job and start a new startup. That is what gave rise to TrojAI. Ultimately, we are building protections for artificial intelligence systems.
"We have developed a solution where we can assess, measure, and track the robustness of a model using empirical measures."
Can you provide more context on the severity of this issue?
James: I can use the example of autonomous vehicles. You've got a model that's been embedded into a self-driving car that takes input from the external world. Well, there are ways to attack that AI model when it's deployed. There are adversarial patch attacks that can be perpetrated by people just standing on certain roads. Whole AI systems can be made completely blind because someone has a digital patch. At TrojAI we think of how to make the model better, more robust. We have developed a solution where we can assess, measure, and track the robustness of a model using empirical measures. A better performing model is a more secure model. A more secure model means it has been constructed responsibly and, therefore, is part of the real push for trustworthy AI.
"One of our roles is to educate, which sets the baseline to explain what we do and why it’s important."
What are some of the main challenges of working in this field?
James: We started the company knowing that attacks on training data and AI models were one of the most significant cybersecurity risks for the next decade. At the time, that risk wasn’t recognized across the industry. It’s been a challenge, and we provided an educational component to our business - to explain why this is important.
Now, enterprises across the board recognize that this is an issue, but that also means a lot of new companies are entering the space. Unfortunately, many of those companies have half baked solutions, which compromises the work that really needs to be done.
One of our roles is to educate, which sets the baseline to explain what we do and why it’s important. We keep going back to what constitutes responsible AI, what constitutes trustworthy AI, and the difference between being responsible versus being trustworthy.
It is vital to explain these concepts properly and have real solutions to address this.
How did you two meet?
Stephen: James and I met around 20 years ago on the Judo mat. I was the instructor. James came in as a student. We quickly learned that we got along well, both hyper competitive with each other, with similar outlooks in life and what we want to achieve.
We come at everything from very opposite ways of looking at the world. James often says we come at things very differently, but we almost always end up in a better place.
Soon after we met, I became a head coach of the women's rugby team and reached out to James for assistance. Our approach to running a team is very similar, whether it be in business or in sports:
How do you speak to the team; how do you motivate them? How do you coach?
From day one we've been laying the groundwork for what a collaborative environment and culture could be at TrojAI and much of it was informed from the work we did coaching together.
"Our culture is based on radical candor."
How do you develop culture?
James: Our culture is based on radical candor. We host a festivus each month based on the secular holiday from the TV show Seinfeld, where you “air out your grievances.” For us, it's a team meeting where we share absolutely everything. We spend a good hour or two talking about what's going well, and what isn’t. I think folks really appreciate it and it unites us as a team.
What have you enjoyed most about starting your own company?
James: On my way home from work recently I ran into one of our recent hires. He was at a great pub around the corner from our office with his wife. He told us he was celebrating his very first paycheck. At the end of the day, even if sometimes it is hard and stressful, we're building a strong, connected team and that’s already a success.
About ventureLAB
ventureLAB is a leading global founder community for hardware technology and enterprise software companies in Canada. Located at the heart of Ontario’s innovation corridor in York Region, ventureLAB is part of one of the biggest and most diverse tech communities in Canada. Our initiatives focused on raising capital, talent retention, commercializing technology and IP, and customer acquisition have enabled thousands of companies to create over 4,500 jobs and raise more than $200 million in investment capital. At ventureLAB, we grow globally competitive tech titans that build-to-scale in Canada, for global markets.
