In Stay Sharp Episode 99, The AI Revolution in Digital Engineering, co-hosts Jonathan Scott and Juliann Grant are joined by Alex Bruskin, an AI expert with 25 years’ experience in PLM and digital engineering. Much of Alex’s work focuses on optimizing the connection between IT and engineering, including data transformation and validation, which makes him the perfect guest to discuss what’s really happening with the emergence of artificial intelligence (AI) in digital engineering for manufacturing.
How We’re Working with AI Now
Alex begins by making the point that AI isn’t new. Since the 1960s, it’s been around in some form, helping people optimize complex processes. What really made the current AI revolution possible is the convergence of a few factors: the immense computing resources accessible now, a mathematical foundation, and the availability of large quantities of data.
Part of why we require so much compute power and AI’s ability to process it all quickly is because society is experiencing an information overload. “It’s essentially a crisis of complexity,” Alex says. “Protein-based intelligence,” aka humans, he says, “Have reached certain limits of how much they can perceive, how much they can process, and how much they can share with other humans efficiently.” AI is filling the gap throughout society, but he notes, “It’s nowhere as important as in engineering because engineering is expensive and the cost of mistakes might be too high to bear alone.”
In addition to bulk processing, AI is also critical as a predictive resource. But that only works if your AI is trained on enough of the right information, including your designs, your business practices, your industry. “If you can steer the design to the right data, if you can validate that it is doing the right decisions, then you can use it,” he says. Right now, AI is predicting language—most of them are large language models, after all—but the idea is that there could be a language of engineering, and AI could be trained to predict the next logical item in an engineering model.
AI, Data, and Human Involvement
As we expand the use cases for AI, we understand how to make it useful for more types of data and more types of business processes. What’s crucial about generative AI these days is that it’s not doing all the work, Alex explains. For instance, society learned quickly that probability or statistics-based AI isn’t the right tool to solve math problems. As a result, AI vendors now train generative AI to call different functions, such as an arithmetic engine. “AI isn’t the right solution for everything, but it is a good wrapper,” Jonathan says. “It is a good way to interact. It’s a good way that it can determine the right solver to call at the right time to help deal with those really complex problems. Sometimes we may call it AI, but it may be a number of pieces of technology that we put together, that AI helps orchestrate.”
AI can also be combined with other tools to make sense of a standard parts database with decades of poor part descriptions or “dirty data.” Alex explains, “We attach [AI] to semantic search with elastic search, and then suddenly we can search for ‘big’ and find ‘large’. We can look for ‘stainless steel’ and find ‘S.S.’, because semantically, it’s very close. We can provide information to the user and let them decide.” Alex and team did exactly that by creating a proximity score. “We say, okay, this value looks like 90% what you need, and this looks like 89%, and so on and so forth. A human eventually decides what part is closest to what he or she’s searching, for” he says.
In this and other cases, a human still needs to be in the loop with AI, according to Alex. “Context means everything in looking at some of this data,” Juliann notes, and what the AI doesn’t understand could be of critical importance. Jonathan adds, “If you are trying to train your AI and you’re asking it to work against data that has errors or mistakes or data that’s not connected, if you ask the AI to make the inferences, it could make the wrong ones. And then if you make decisions based on those bad inferences, that’s not good.”
Learn More About the State of AI and Digital Engineering
The full podcast contains more discussion and details on a variety of topics, including why Alex compares AI to iPhones, how AI can improve the process and the success rate of data migrations by helping extract and insert data, how even drawings could be fed to AI to extract their data, what AI use case Alex is the most excited about, and more.
Be sure to check out Stay Sharp Episode 99: The AI Revolution in Digital Engineering and join us each week for a new podcast.
Follow the Razorleaf Podcast, Stay Sharp in Digital Engineering on:
