What’s the Computer Think?

What massive computing power will learn from design data.

Jamie Gooch I grew up watching reruns of the “Jetsons,” “Lost in Space” and the original “Star Trek.” Maybe it was because the shows were already old (“Lost in Space” wasn’t even in color) or because I was watching them around the same time I was getting my first taste of BASIC programming—but I took for granted that having a meaningful conversation with a computer was in my not-too-distant future.

Fast forward 30-some years and I can indeed ask a computer for its opinion, but while Apples’ Siri, Microsoft’s Cortana and Google Now’s voice assistants are useful, they’re not capable of scintillating conversation. They are much more powerful versions of the BASIC programming If-Then statement. If I ask about movie show times, then they display the show times for theaters near me. If I ask how to optimize a design, the results aren’t as useful. It’s not like they’re thinking.

But that is changing as machine learning algorithms become more sophisticated and massive computer power becomes more accessible. Via deep learning, computers have taught themselves to read Chinese, identify cats by watching YouTube videos and—particularly disturbing to me—write articles that make sense. (Editor’s note: This article was written the old-fashioned way.)

Automated Design Optimization

Many of the advancements in artificial intelligence stem from the fact that computers can quickly process massive amounts of data. For example, 1,000 computers were connected and fed 10 million YouTube stills for three days to identify and learn cat faces. So what? Google didn’t tell the computers to learn cat faces; the computers decided that on their own. That experiment took place two years ago. Google is now using the same type of technology to enable computers to “read” massive amounts of text online in order to deduce the meaning of words. That scintillating conversation with your phone may take place sooner than you think.

Cool? Yes. Scary? Maybe. But what does it have to do with design? In this issue’s special focus on lightweighting, you can see a number of examples of software that suggest design ideas that people probably would not have thought of themselves. Design engineers supply the design data and constraints, and the software suggests different, optimized ways to meet those parameters.

The optimization algorithms still depend on the information being provided by the design engineer. But what if the software could come up with that information on its own? What if the data being fed to computers wasn’t YouTube stills or words, but 3D parts catalogs, your previous models and data captured from real-world products similar to what is being designed? What would massive computing power be able to glean from such data?

“We’re developing a system that learns the same way we do … and the outcome is a tool that works in a lifelike manner and supports the way we solve problems naturally,” said Autodesk CTO Jeff Kowalski during the Autodesk University 2014 keynote last month in Las Vegas. “We need to stop telling the computer what to do and instead tell the computer what we want to achieve.”

Kowalski admitted that this idea—part of what Autodesk calls generative design—is not new, but the computing power to make it a reality is finally here. From the many real-world computing examples I heard a few months ago in New Orleans at the Supercomputing 2014 conference, I don’t doubt it.

“The biggest change has been the ability to work with much richer data,” said Autodesk CEO Carl Bass in his AU 2014 keynote. “Data that better captures the complexity of the real world.”

Automated Manufacturing

To be clear, there is no product commercially available right now that can learn what makes a good design and suggest a 3D model of it. Even if that technology is eventually created, such a design might be impossible to manufacture via traditional means. That’s where 3D printing comes in.

Considered together, the democratization of high-performance computing, design and simulation optimization technologies, and advances expected in additive manufacturing and industrial automation paint an amazing picture of the future. But what is the role of the design engineer in such a future? The current levels of automation in manufacturing didn’t cause the massive unemployment that many predicted, but it is often cited as a reason for stagnant wages.

Will some design engineers be replaced by technology? Only time will tell. One thing is certain: If tomorrow’s technology can autonomously design and produce optimized parts, it will certainly be able to write an article about the process. I hope to be retired by then.

Share This Article

Subscribe to our FREE magazine, FREE email newsletters or both!

Join over 90,000 engineering professionals who get fresh engineering news as soon as it is published.


About the Author

Jamie Gooch's avatar
Jamie Gooch

Jamie Gooch is the former editorial director of Digital Engineering.

      Follow DE
#13410