Cruise
- Artificial intelligence has the potential to completely change how companies like Walmart and McDonald's operate, but one challenge can be figuring out how well early-stage projects are advancing.
- Hussein Mehanna, the head of artificial intelligence and machine learning at Cruise, says a key metric is how many models an organization is testing in a given month.
- While he declined to say how many the self-driving-car company has ongoing, Mehanna said it's more difficult for autonomous driving models to scale to a level more akin to Facebook's 3 million monthly tests.
- Click here for more BI Prime stories.
Artificial intelligence holds the promise to revamp how much of corporate America operates, but one problem can be determining whether early initiatives are on the right track.
For Hussein Mehanna, the head of AI and machine learning at self-driving-car company Cruise, a key metric of how advanced a company is in pursuing the technology is the number of models it runs each month.
When he served as a director of engineering at Facebook, for example, Mehanna said the amount went from less than 100 in a month to over 3 million as the social media giant invested more heavily in the technology. For comparison, Walmart - which is often hailed as one of the leaders in the push to adopt AI - has roughly 100,000 projects running at one time.
Cruise is almost certainly running less than that - he declined to say exactly how many per month - but it's a goal of co-founder Kyle Vogt to get models trained and operational "almost as fast as compiling code," according to Mehanna. "But we're not that fast."
He noted that other Silicon Valley-based transportation companies are only running 20 a month, though he also declined to say which firms he was referencing. But one challenge for a self-driving-car company like Cruise is the sheer amount of data needed to power the AI models.
"The size of the model, the amount of data is far larger than what other companies have," Mehanna told Business Insider. "It's multi-sensor data over multiple frames, so it's pretty huge. This is the largest in my career so far."
And Mehanna has had a long career, one that has correlated very closely to the overall rise in interest in machine-learning tech. At Cruise - which was purchased by General Motors in 2016 - he is overseeing the effort to build AI models that can anticipate the activity of other drivers, as well as pedestrians.
The company is currently testing out hundreds of its autonomous vehicles in San Francisco. Those cars collect tons of data - hundreds of gigabytes per second - that are then used to train the models.
"After we train these models, we evaluate them and prune them, again as fast as possible, test them offline through simulation," said Mehanna. "The models that match and survive, we put them back on the road and we do this loop again and again."
And along with Facebook, Mehanna clocked six months as director of engineering at Snapchat. Before that, he was a senior software engineer at Microsoft as the tech giant was "investing the most in machine learning, even ahead of Google."
"Microsoft is really typical in starting things early but never winning at the end," said Mehanna. "We were heavy in machine learning, and that's where I learned my early chops."