January 8, 2019
Faye Francy, executive director of the Automotive Information Sharing and Analysis Center (Auto-ISAC), led a conversation about the impact of machine learning, deep learning and AI on the autonomous vehicle (AV) ecosystem. “They work together to bring great things — and possibly nefarious things — to the auto industry,” she said. Inivision AI chairman Seamus Hatch noted that the three terms aren’t interchangeable. “We’re many years behind the singularity,” he said. “It’s a machine trained to solve a specific problem faster and more accurately than a human.”
SafeRide Technologies’ Gil Reiter described his metaphor as AI as the big topic, inside of which machine learning resides, and then, inside that, deep learning.
Harman Connected Services vice president Andrew Till said his company subscribes to a connected world vision. “Machine learning is firstly about how we align people to have portability of the experience from one environment to the other,” he said. “If you set your thermostat in the house, is there a way to bring that into the car?” His company is also looking at the crucial area of vehicle-to-vehicle communication, which is part of the smart cities paradigm.
Reiter and Karamba Security chief executive Ami Dotan are focused on cybersecurity issues with AVs, particularly hackers’ use of AI. “Hackers use it to define abnormalities and figure out how we use Google or Alexa,“ said Dotan. “They can then use that for pfishing. AI is a Cold War balance on both sides.”
Reiter pictures the car as a mega-computer running on the road. “The IT industry had three decades to evolve its technology to the point where it’s fairly secure to do anything online, and still there are cyberattacks uncovered every day,” he said. “The challenge is that the same revolution in IT is happening here, much faster. AI can help you bridge the cybersecurity gap and find unknown vulnerabilities you’d never find any other way.”
Hatch pointed out the dangers of focusing too much on neural networks. “It’s incredibly intellectually compelling,” he said. “You want to wire up every sensor in the car and create an end-to-end solution. But it can be unworkable.” So-called rare events happen when millions of people are driving hundreds of millions of miles. “There’s the potential for things going really wrong,” he said. “Straight procedural logic might be a better solution than using AI in these cases to understand what’s going on. End-to-end [AI solutions] are doomed to failure.”
One of the current issues in the automotive industry is the push and pull of sharing information. “There’s a huge advantage in terms of cost and safety and bottom line revenue to sharing data,” said Hatch. But, noted Dotan, companies are loathe to give up the information that differentiates them from the competition. “Why should the company that invests $100 million in AVs share data with another company that invested $10 million?” he asked.
Till said that his company “wants to accelerate and enhance sharing.” “If we don’t share in this industry, we’ll be forced to share in an intervention,” he said. “Through sharing and contribution of code, you can accelerate R&D efforts.”