From Speech Recognition to Autonomous Vehicles

karen kaushansky
2 min readJun 24, 2016

Twenty years ago this month at Nortel Networks, I started designing “Voice Activated Dialing” for home phones and the first generation of cell phones. Speech technology was such that it allowed only twenty address book entries, where each entry was painfully trained as if hardcoded. There was a case then for entering a star code on your landline phone to access twenty address book names: memorizing a code to make it easier to dial 20 of you closest friends and family.

Fast forward ten years to 2006, and I was working on VoiceDial for AT&T while at Tellme Networks: pay $4.99 a month to push a code on your mobile phone to access your web-based address book. Now you could drive AND call friends hands free at the same time.

In 2011, Siri was introduced to the world, and with its improvements over time, now a simple “Hey Siri, call mom” and my mom is on the phone.

Looking back, what can we learn? Experiences that begin changing the world are both compute-intensive and even start with labor and/or friction. They are often hand-tuned and carefully monitored. Those who recognize the need or foresee the value, are willing to invest the effort either because of the promise or the alternatives are inferior.

So here I am: back at the beginning again. Autonomous vehicles will go through a similar evolution but this time, with machine learning and artificial intelligence, it won’t take twenty years . Today we pre-map our streets to know where to drive; we collect and tag data to train our models to differentiate joggers from skateboarders. Hard traffic scenarios need to specifically be modeled out. But the robots have started to learn for themselves. In Anca Dragan’s work, robots observed how humans drive and then taught themselves to inch backwards at a four-way stop to signal their intentions.

What are the initial friction points? Well I expect people will need to wait a little longer, drive a little slower, or pay a little more for a robot taxi, for example, with the benefit of a better, more personalized and safer service than what exists today. The larger impact will be a transformation of our streets and our cities.

It’s all starting to come together; last week Local Motors introduced Olli, an autonomous, electric bus powered by IBM Watson. What excites me the most? I can hop a ride and ask things like “Are we there yet?” or “ What’s the vehicle doing?” and it will answer back. I just wonder with 12 people on board, whose mom will be called if I ask.



karen kaushansky

Senior Conversation Designer at Google. I tend to work on things 5–10 years out: speech recognition, biometrics, autonomous vehicles, conversational interfaces.