“Do things with your own means, in your own way” — An interview with Dr Joseph Dureau, our new CTO at Snips
Snips is betting big on voice and Privacy for the Internet of Things. Dr Joseph Dureau, who joined us as our second hire, has recently been named CTO. Congrats, Jo.
What is your background?
I was originally trained as an engineer in applied mathematics, which brought me to varied subjects. I spent a year at NASA working on climatology, and another year at the French commission for atomic energy, doing machine learning. After that, I did a PhD in Statistics at the London School of Economics, on questions related to epidemics, and evaluation of public health interventions.
When I finished my PhD, I got attracted by the vitality of the tech world. I felt there was so much to learn, and so much to change through technology! I co-founded Standard Analytics, a NYC-based semantic web company focused on scientific publications. I left this company about three years ago to join Snips as one of their first data scientists.
Why did you join Snips?
First of all, I was impressed by the team. At the time, there were only 6 people (the 3 founders, a data scientist and two interns), but they allhad incredible experience, a fascinating energy, and an inspiring approach to problems. But perhaps most importantly, I knew as soon as I met them that the company had a strong ethos — exemplified by their betting big on Privacy at a time when nobody cared.
Lastly, I was hopeful that Snips would have the ambition and the means to be a European success story. Still some way to go, but I’m thrilled to be part of that adventure. ;)
Why AI and voice?
Voice is one of the most active theaters for AI today, and one in which advocating for Privacy is absolutely fundamental. After all, voice is a biometric marker, and using a voice-first device generally means having an always-on microphone in your environment.
Beyond the question of Privacy, there is also a very fun and futuristic aspect of working with voice. We are still in the early and often clumsy days of this technology: when you see Echo-like devices reacting to videos playing on our TVs / computers, it doesn’t feel like we’ve nailed it yet.
I’m looking forward to the day in which natural language interactions become ubiquitous, Privacy-preserving, and context aware. If we do our job right, voice interactions should feel like magic, make technology so intuitive that it disappears into the background.
How did you end up becoming CTO at Snips?
On one side, it’s been quite organic. Growing the machine learning team has been a very structuring experience. I joined the company as a data scientist, eventually taking on the role of VP of Machine Learning, where I was coordinating many things in collaboration with our VP of engineering. This experience and close collaboration gave me both depth and breadth on every aspect of our technology, something necessary to make sure our stack stays coherent as the product and technology matures!
Technology aside, I also put a lot of focus on making people work together, and with method. It’s vital for an AI company like ours — a company that hinges both on deep engineering and deep machine learning skills. We’ve iterated a lot on our structure, putting a lot of emphasis on versatility, autonomy, and eagerness to learn new skills. These have always been the canvas on which we’ve built our organization, and one that I will keep advocating!
Having said that, I think I have a steep learning curve ahead of me to become the best CTO my company can have. ;)
What is your vision for the technology over the next months / years?
There are so many interesting things happening, actually!
First, our early bet on Privacy is finally paying off, with European regulations forcing any company doing business in Europe to follow certain Privacy-by-Design principles. This is an important step in the right direction, and countries outside of Europe will hopefully be inspired to follow a similar path! This is particularly true in the US, where the current government doesn’t seem to consider Privacy for what it is: a fundamental human right.
From a technical perspective, we offer Privacy by design by running all of our computations directly on the device, and not in the cloud. We are the only company able to do Deep Learning directly on-device, including entry-level hardware such as a Raspberry Pi and below.
Privacy aside, on-device also means resilience to internet connectivity issues, which is a key selling point for IoT manufacturers. People are just fed up with unreliable technology, and rightly so!
Lastly, I believe strongly in giving back to the community. We have already published a number of papers and open sourced libraries, and I foresee this become an increasingly important part of what we do at Snips. Trust and standards will be critical for the development of the Internet of things.
Do you think Snips can beat the big guys? How?
Well, I know we can, and in fact we already did on some specific problems!
In the field of Natural Language Understanding, for instance, we’re starting to have results that are on par with assistants from Amazon and Apple, and better than those from Google and Microsoft!
To be honest, we didn’t really have a choice, as we made a commitment to our customers that we would never compromise on performance in exchange for Privacy. Using our technology has to be as least as good as good what our customers would get elsewhere, but with Privacy added.
What’s important to keep in mind is that you need to do things with your own means, in your own way. As a startup focused on Privacy, we don’t have any user data to train our machine learning algorithms. This ethical choice forced us to be creative: we invented a way to automatically generate a large datasets of human-grade supervised data, which we can then use to train deep learning algorithms for multiple languages.
Something else we found was that the marginal gain of having more user data decreases very fast. This is why you now see many startups able to compete against big guys, despite not having anywhere near the same amount of data.
What was your best day at Snips?
Growing from six employees to over forty now feels like we created multiple companies, each bringing their own memories. There has been many moments of euphoria, many fun moments with the team.
One very memorable day for me was the day my team shipped the first working version of our AI platform. Everything wasn’t perfect —far from it— but we had gone further than we would have dared to imagine when we started the project. It felt good turning an idea into an amazing product that anyone can now use!
