In a wide-ranging discussion today at VentureBeat’s AI Transform 2019 conference in San Francisco, AWS AI VP Swami Sivasubramanian declared “Every innovation in technology is going to be driven by developers.”
Sivasubramanian made the statement while talking about growing demand for machine learning engineers and internal efforts at Amazon to train more employees to use machine learning. Facebook VP Jérôme Pesenti also stressed plans to make machine learning part of each employee’s job at Facebook. And earlier today, Amazon committed $700 million to upskilling its U.S. workers.
“Amazon developed what we call Machine Learning University. This is what we use to train our own engineers on machine learning, even if they didn’t take it as part of their own university [course work],” Sivasubramanian said. “When we externalized it as part of the AWS training and certification platform, we had more than 100,000 people register to start learning ML in less than 48 hours. Think about that: That’s the level of appetite we’re starting to see.”
One Amazon employee with no prior knowledge of machine learning used AWS SageMaker to create a computer vision system that recognizes whether a cat is carrying dead prey into the house. If the computer vision system predicts with high confidence that kitty is in fact carrying a dead animal, the system locks the pet door.
In response to a question from Wing research partner and moderator Rajeev Chand, Sivasubramanian agreed that corporations should have a C-level executive in charge of AI deployment. But he believes the first step is a sound data strategy.
“I think [companies should have C-suite leadership on AI], but I do not think the title is chief AI officer or chief analytics officer or data officer,” he said. “The number one thing I can say there is you want to get your data strategy right, because if you don’t, when you end up hiring a machine learning scientist and you expect them to come and invent amazing new algorithms, the reality is they spend a large percent of their time dealing with data cleanup and data quality setup and so forth. So getting your data strategy right is probably one of the hardest things.”
Pesenti echoed this thought shortly after Sivasubramanian, suggesting that companies anxious to get started with AI hire a chief data officer before they go in search of a chief AI officer. He too believes companies that get their data sets in order have a better chance of collecting the right data to power AI model training.
Sivasubramanian touched on myriad topics onstage, such as whether AI should look like humans (he believes it should not), how stupid AI is today, and how global corporations should think about the use of facial recognition software.
Addressing the topic of AI’s relative stupidity, a subject discussed with Google Cloud AI chief Andrew Moore on the first day of Transform, Sivasubramanian spoke about his four-year-old daughter.
“She learned to recognize a tomato by looking at a tomato probably three times, whereas a machine learning computer vision system, arguably it used to require seeing 10,000 pictures. Now it’s probably like half [that]. It’s not yet at the level of being able to deal with ambiguity in the same way that humans are,” he said, adding that AI is good at doing things humans find boring, without making errors.