AI Developer Shares His Thoughts

During June, Hexagon invited me to its HxGN Live meeting in Las Vegas. The mining industry sees one vertical aspect of Hexagon, but the company’s products and tools are used in many different areas, including agriculture, construction, emergency response, etc. The company recently announced that it was collaborating with Nvidia on a suite of solutions for industrial digitalization. Together they are developing an app enabled with artificial intelligence (AI) that will let users see real-time comparisons of digital twins. The difference between it and previous digital twins is that it will not only look better, it will also behave correctly when tasked with what-if scenarios.

During a session labeled “World Building,” Nvidia CTO Rev Lebaredian discussed the development of computing power and the advancements the company is making with AI. Full disclosure: I own a few shares of Nvidia.

Rev got his start in Los Angeles working as a graphic designer for Warner Brothers and Disney. “I really love the business of creating beautiful images using algorithms and simulations,” he said. “What I really wanted to do though is do it in real-time so that we can experience these things interactively.” He joined Nvidia 21 years ago to do just that.

AI is the hot topic now, but Rev and the folks at Nvidia have been thinking about it for a long time. Modern AI, deep learning, he explained, was essentially born on Nvidia’s computers. In 2012, a group of scientists took one of Nvidia’s gaming PCs and connected it to data using a fairly old idea, neural networks. Then, they managed to do something that had vexed computer scientists for decades. Their computer was able to robustly tell the difference between a cat and a dog inside an image.

In a relatively short period of time, computing progressed from systems that were 70% accurate at telling the difference between a cat and dog to being superhuman, he explained. “Today, AI can now determine the exact species of cats and dogs far better than any humans that have this particular skill,” he said.

The key to this change was the use of algorithms. The most advanced algos in the future will not be created by humans directly, he explained. “We used to program algorithms based on some ideas of how we would want them to work and then we hit our limit,” he said. “Instead, we wrote an algorithm, pumped a whole bunch of data into it and out comes a new algorithm. We are essentially writing software to write software.” While the rest of the world is trying to wrap its head around the implications, Nvidia has been building all the infrastructure it will need to move forward with the next generation of AI.

What’s next? According to Rev, AI will begin to aggregate all readily available data from different sources. Then there’s the intelligence that comes from other sources, like the engineers that design mines and processing plants. Once AI has those inputs, it can model complex installations. Today, it’s impossible to ensure that a new processing plant is going to work until it is built. If this could be done early in the design process, engineers could solve the problems way before ground is broken. It has the potential to save billions of dollars, and thousands of hours of wasted time, but it also has the potential to replace the engineering department.