Teach the Robots Well

Maybe it’s because I just finally read “Do Androids Dream of Electric Sheep?” the book that the movie “Blade Runner” is based on, that I’ve noticed so much artificial intelligence in the news lately. The coverage hasn’t approached the post-apocalyptic nightmare of the fiction, but neither has it cemented a shiny utopian future for those following along.

It started off innocently enough. In March, the AlphaGo program, created by Google’s DeepMind division, beat a human champion of the complex board game Go. Then it turned ugly. Microsoft’s Twitter chat bot, Tay, began broadcasting obscenity-laced racist and sexist comments that it picked up from an online group intent on taking advantage of the chatbot’s “learning features.”

On the could go either way front, Facebook announced tools that allow developers to build bots inside Facebook Messenger, enabling news personalization or automated ordering using conversational language in the messaging app; then IBM announced that its cognitive computing system Watson is learning to become more of a Sherlock. Eight universities will be involved in teaching Watson to detect cyber crime vulnerabilities over the next year.

And, finally, in news sure to make Vegas bookmakers take notice, an artificial intelligence platform based on the “swarm intelligence” of actual human beings, correctly predicted the Kentucky Derby Superfecta—that is, the first four horses to finish the race, in order.

The examples above represent just a few of the varied types of neural networks, pattern recognition algorithms, data-driven predictions, machine learning and deep learning that are all being lumped together under the AI moniker. They have important differences, but they all have a huge appetite for data—and engineers are preparing the menus.

Data Dining

Before data is mined, it must be collected. Thanks to the explosion of connected products, there is no shortage of data to feed the machines. Engineering teams are in the perfect position to help determine what that data will be. What sensors will be integrated? How will products communicate with the cloud and one another? What variables will be most telling? The larger the data set, the better for AI—whether it helps factory robots safely collaborate with humans, allows self-driving cars to recognize stop signs or lets you tell a computer what you want on your pizza.

The exciting thing about today’s AI is when a product appears to learn to perform an action based on the data it is receiving—as if it is making a human-like decision. To help make those decisions faster, processing power is being driven to the “edge” (which is displacing the cloud as the latest technology buzzword). From a design engineering point of view, that means you’ll be designing, specifying and embedding even more computing hardware and software directly into products so they can process data at the source.

“A radical transformation is underway from the cloud to the edge of every major system,” said Les Santiago, research director, Wireless and IoT Semiconductors at IDC in a press release. “In an effort to address the opportunity, both edge and cloud infrastructure needs to continue to scale and support trillions of sensors and billions of systems. Increasing intelligence at the edge will be one of the primary drivers of growth of the overall semiconductor market over the next few years against the backdrop of a maturing smartphone and PC market, and a difficult pricing environment in the memory markets.”

As Terry Jones, founder of Travelocity and founding chairman of Kayak.com, told Siemens PLM Connection 2016 attendees in his keynote last month: “Location, location, location used to be about 1st and Main, but today location is about the edge of the glass. It’s about where the customer is. It’s the edge that’s important.”

Integration Makes Technology Look Easy

It’s not enough for engineers to design the products used on the edge, they need to get a look at the edge themselves. I don’t mean camping out on the factory floor or consumers’ homes—though that would surely yield interesting data—I mean getting access to usage info, equipment failures, manufacturing logs and other data directly in the software being used to design the next generation of products.

Actionable edge information isn’t just for operations, sales and marketing. It is the key to continuously improving increasingly complex products in less and less time. All the major engineering software vendors are focused on data sharing, PLM integration and collaboration because it’s critical to innovation. To design smarter products, engineers need access to intelligence from the edge.