How ChatGPT and other new AI tools could reshape industry

PARIS -- Generative AI, the technology behind the popular chatbot ChatGPT, is poised to drive innovation across the industry, from design to self-driving testing to infotainment to factory processes, an executive at tech company Nvidia says.

"AI was under the radar in a lot of cases, but now it really is becoming accessible to everyone," said Danny Shapiro, vice president automotive at Nvidia.

Suppliers including Nvidia are developing ways to integrate generative AI into their products, such as the California-based company’s Omniverse industrial metaverse platform that helps automakers optimize their industrial processes and Drive, its autonomous vehicle development platform.

Other uses include automotive design, infotainment and even configurators that let potential buyers create their own cars in the virtual world. 

Generative AI is a set of algorithms that can create new, realistic content based on "training data" such as text or images. 

So-called chatbots such as ChatGPT accept text or image inputs -- known as prompts -- to create text outputs, with use cases that include writing software code, summarizing articles or analyzing data sets. 

Other tools such as Midjourney, DALL-E and Stable Diffusion can create images from natural language inputs.

The development and release to the public of these easy-to-use tools are AI’s "iPhone moment," Nvidia founder Jensen Huang said last month at Nvidia’s annual GTC developers’ conference.    

"Now, everybody can become a programmer not by learning a programming language, but by using English to describe something they want [the computer] to do," Shapiro told Automotive News Europe.

For the automotive industry, generative AI can simplify and speed up the design process for parts and even the entire vehicle. AI tools can be taught the laws of physics or thermodynamics to modify existing designs to be more fuel efficient or aerodynamic, Shapiro said -- or they can draw on the body of designs that already exist to generate models based on certain parameters. 

"That design then flows into the engineering side," he said, adding that AI tools could be used, for example, to help reduce weight without compromising safety. The tools could remove material from a component virtually, then run a simulation analysis to create different permutations of a particular part or structural element that are lighter or stronger. 

Car enthusiasts around the web have been using tools such as Midjourney to redesign their favorite models, create photo dioramas or create their own dream cars, often giving instructions for the exact prompts that they used. 

Generative design models, however, still need to be refined by humans, Shapiro and others say, and the output quality depends largely on the quality of the prompts.

Another use of generative AI in the design space is to improve online configurators, as more and more of the car-buying process goes online.

"A user can say, 'Show me a blue, two-door model with a tan interior'," Shapiro said, in much the same way a designer could ask to generate a compact crossover with a certain drag coefficient that resembles a specific existing model.

In a similar way, generative AI can help create more efficient factories -- an area where Nvidia and other suppliers are working with automakers to create "digital twins" to test out designs and processes, and iron out any bugs before even the first dirt is shoveled.

"Think of a factory as just millions of parts," Shapiro said. Generative AI is "creating a more efficient part and more efficient ways of putting these parts together," he said, "and in the case of a factory, how the layout and flow works."

Generative AI can also enhance infotainment systems by allowing more natural voice commands. "One of the things that is really awesome about these new tools is that they are ‘context aware,’ so every query isn’t a new, unique thing, but rather maintains an understanding of the dialogue," he said. 

For example, he said, a driver could ask what the weather is in Detroit; a follow up question could simply be "And in London?" – because the voice assistant, enabled by generative AI, would already know that the subject of the conversation was the weather. 

Mercedes has shown a ChatGPT-enabled voice assistant, and GM is reportedly working on an in-car personal assistant that uses generative AI.

Perhaps no other automotive development field has been dependent on AI as autonomous vehicles, which need to react to a seemingly infinite number of scenarios. Data from these events are used to train AI models, but the difficulties are that real-world collision data is scarce, and it isn’t feasible to recreate them in the real world, Nvidia says.

One solution is to turn to generative AI. "We will use synthetic data generation to create all kinds of scenarios that could exist," Shapiro said. Once this dataset has been created, in addition to training AI models, it can also be used to validate self-driving systems. "We need to make sure that the data system or the self-driving system can handle those situations, and we can do that in simulation," he said.

Shapiro added that these uses for generative AI will take several years to be refined and applied. The speed of deployment depends in large part on being able to connect cars to the cloud, he said. 

At the same time, providers of design and engineering tools such as Dassault, Autodesk and Adobe are quickly integrating generative AI.

Shapiro said it was too early to accurately quantify benefits from generative AI, but he added that in factory productivity, for example, the combination of digital twins and new AI tools "will make a huge impact in terms of efficiencies, cost reductions and time savings."

"AI doesn’t need to sleep, and its faster and more capable than any human at specific tasks," he said. "If you can know how to direct that a specific problem, you can achieve much better results."