1. Many thanks to IBM Watson for sponsoring our first-ever Inside AI Live conference last month in San Francisco, and to our sponsors Talla, Half Court Ventures, and LAUNCH. Today we're presenting the talk, "Autonomous AI and Tolerable Error Rate," from Tobenna Arodiogbu, co-founder and CEO of Scotty Labs.
San Francisco-based Scotty Labs is a driving company spun out of SRI International that's building an autonomous trucking solution. Unlike other companies in the self-driving industry, Scotty is not a believer in the idea of getting the driver out of the truck as soon as possible, says CEO Tobenna Arodiogbu. "There are a bunch of steps we need to take before we get there," he explains.
At Scotty Labs, Arodiogbu says they always have to think about how much testing is enough to bring their autonomous trucks to market.
In the case of building autonomous vehicles, the tolerable response time is under 100 milliseconds. That means there are no shortcuts; you have to have a really good product, he says. For the past two years, Scotty has been building its products, testing them in simulations, testing them on private roads and then public roads.
Rather than thinking about ways to increase the acceptable error rate or response time, Scotty narrowed the operational design domain of the products.
"We're not trying to build something that works everywhere," Arodiogbu says. "We're not trying to build something that takes the truck driver out of the truck."
The specific problems the company wants to tackle first are, how do they extend driver hours of service, and how do you increase the comfort and safety of the driver? That means their testing is limited to specific roads and highways and weather conditions, with no driving at nighttime or above 55 mph.
In addition, if you're building a product with a low tolerable error rate, that means you need a lot of capital, Arodiogbu says. "All AI products are not created equal," he adds.
To watch all of Arodiogbu's presentation, check out the full video here.