Dusseldorf, Vienna If BMW has its way, revolutionary times will dawn in software development. “Become part of the ‘iPhone moment in artificial intelligence’ as our magician who makes things possible with generative AI.” With these words, the car manufacturer is looking for a job advertisement for a project manager for the area, which did not even exist a few months ago .
The novel AI language models, which gained notoriety earlier this year with the emergence of the AI-based bot ChatGPT, could open up new software development opportunities for the auto industry. According to a study by the management consultancy Boston Consulting Group, car manufacturers could use generative AI to accelerate their own software development by up to 55 percent and at the same time reduce costs by up to 95 percent.
These are fantastic prospects for car manufacturers such as BMW, Mercedes and Volkswagen. Because so far, despite investing billions in software development, they have lagged behind competitors such as Tesla or new Chinese electric car brands. They are dependent on tech companies like Google and suppliers like Bosch, Continental and ZF. The software commitment of the German car manufacturers is mainly characterized by problems and delays.
“We are convinced that generative AI can make a value-added contribution to increasing productivity and the pace of development,” says BMW. The car manufacturer is already using AI language models for its digital user manual, the “Car Expert”.
“It finds the right spots and looks for the answer, summarizes it intelligently and makes it easily accessible in human-like dialogues,” says BMW. The Dax group is also experimenting with image-generating AI. This could support design processes and vehicle development processes.
>> Read also: VW and Mercedes are turning around when it comes to software – too much effort, too little know-how, too high costs
The development departments of Volkswagen’s software unit Cariad are also looking for ways to integrate generative AI into software development and process automation, says David Ilgener from Cariad. The manager heads the area at Volkswagen’s software unit that takes care of the introduction of new tools for software development. “For example, we found that these language models can also program surprisingly well in the automotive environment, although they are not explicitly trained for this,” says Ilgener.
VW’s main motivation is likely to be to speed up software development. In the recent past, Wolfsburg had massive problems there. Important model launches had to be postponed because software was not finished or was faulty.
The notorious lack of talented coders is one reason for this. The entire German auto industry has been suffering from this shortage of personnel for years.
More code with the same number of developers thanks to AI
This skilled worker vacuum could at least partially eliminate generative AI, according to the study by BCG. “With generative AI, code is not only developed in half the time. Also, more code can be generated with the same number of collaborators,” the authors write. According to an analysis, 90 percent of developers would feel more productive thanks to the use of language models, and 96 percent noticed a significant acceleration in repetitive tasks.
From BCG’s point of view, generative AI could, for example, take over the testing and validation of developed software, an important area for suppliers. After all, the millions and millions of lines of software code for car manufacturers are still being developed by suppliers. The automotive supplier Continental announced that it could already use generative AI to improve the process through to product completion.
An employee tests trips with the Hamburg Truck Pilot.
(Photo: Thies Raetzke)
The automation of complex processes also plays an important role. One area would be the so-called labeling of raw data. This is necessary in the development of automated driving. Objects such as cars, streets or pedestrians have been traced and categorized in camera images in a largely manual process. An AI is trained with this labeled data. According to BCG, an image-generating generative AI could automate this process with up to 25 percent greater precision.
However, the industry is still in a discovery process, also because the technology suffers from inadequacies. The language models still tend to “hallucinate”. The models would work plausibly and linguistically adept, but not always correctly, says BMW. At Cariad, the possible applications are classified so that the developers can estimate in advance which software products are suitable for the use of generative AI.
>> Read also: Working on the digital assembly line: Clickworkers train AI – for $40 a month
Elmar Pritsch, partner at the consulting firm Deloitte and responsible for car software there, sees the use of the new language models as just the beginning. “Generative AI will not immediately become a game changer in software development in the automotive industry,” says Pritsch. The car manufacturers would first have to create the conditions for this, including standardized software and hardware architectures. “AI-generated code can then act on this and be used much more efficiently.”
Read more about artificial intelligence:
Hardly anyone in the industry expects that the majority of the software in the automotive industry will be developed automatically with the help of the models in the future. “We assume that the tools will be an important tool for our developers and, for example, free them from repetitive programming tasks or participate as co-developers,” says Ilgener from Cariad.
According to Pritsch, it will still be some time before generative AI is widely used in the automotive industry. In contrast to electronic items such as smartphones, there are also higher safety requirements, since failure can result in traffic accidents with injuries.
More: VW boss Oliver Blume begins restructuring Cariad
First publication: 06/11/2023, 1:36 p.m.