Generative AI has taken the world by storm this past year. Many industries both inside and outside automotive are looking to leverage its power to increase profits.
However, while generative AI can be applied to many use cases, other forms of AI may be better suited in some instances.
In this insight, SBD explores different automotive use cases and the type of AI which can best support it.
What is happening?
The latest advancements in Generative AI (Gen AI) have demonstrated significant potential across various business processes, prompting OEMs to reevaluate their AI strategies and applications.
AI can support a wide range of business activities. But it can be categorized into two main objectives, consumer experience enhancement and OEM business process optimization.
Most OEMs today prioritize integrating AI into modern car features like Virtual Assistant and ADAS. Yet, we see many opportunities for in-car services and OEM business processes, especially in R&D.
While Gen AI has attracted lots of attention recently, it might not be the best fit for all use cases.
We anticipate there will be numerous AI initiatives within OEMs spearheaded by diverse teams. This will need high level prioritization and coordination to ensure maximum efficiency and best Return on Investment is achieved.
Mercedes-Benz is becoming a leader in automotive AI by investing more in AI than others, including an employee AI training program.
Why does it matter?
The relationship between AI models can be complicated. At a high level, these models can be viewed as subcategories of one another with each one providing additional functionality. Understanding which model to apply to each use case is critical for effective implementation.
Machine Learning: A form of AI that can be trained to perform tasks which may be automated. It can also identify and classify data in a structured form (i.e., in tabular form). Automotive applications include predictive maintenance, remote diagnostics in cars. Machine learning needs less data and runs smoothly on modern vehicle MCUs/MPUs.
Deep Learning: A subset of machine learning which uses a neural network to analyze data. It outperforms machine learning when unstructured data is used (image classification, natural language processing, video, etc.). Deep learning is generally more about recognizing patterns, making decisions, or extracting features from data.
Generative AI (Gen AI): A subset of deep learning. Gen AI can read data, learn from it, and generate new synthetic instances. This form of AI works well for a situation which requires adaptive outputs.
Where next?
SBD has evaluated some use cases related to the automotive industry and determined which ones should be of high priority based on implementation time and investment.
The LLM-powered virtual assistant will transform in-car experiences with its deep grasp of human intentions, multi-objective capability, and swift responses. However, integrating LLM-powered virtual assistants into vehicles and making a seamless experience on the road requires more fine-turning effort.
Predictive Maintenance and Advanced Vehicle Diagnostics can be facilitated using traditional ML algorithms, eliminating the need for Gen AI. However, gathering such data demands a robust information infrastructure for both the vehicle and its service network.
Customer Service Agent on either an OEM website or app can be a quick win to leverage Gen AI with very little fine-tuning and can be implemented in weeks.
Smart Factory takes a longer time and more investment to build and fine-tune with digital twin technology. However, this is an ideal moment to begin as most OEMs are preparing their new EV plants.
What to watch out for?
While AI can be overhyped, it is still a very powerful tool when implemented correctly. However, the entry ticket can be expensive depending on the model used and there are uncertainties and concerns when applying it into enterprises. SBD has the following suggestions:
AI will be embedded into different layers of technology stacks for future vehicles. It will be a combination of different small and large models running from chip to cloud. Monitoring industry trends will help OEMs to refine their AI strategy.
The operational cost of LLM-Inferencing remains high and should be a primary consideration for large-scale implementations of Gen AI. However, it does save cost on data annotation during the model training.
Watch edge AI players. Research is ongoing into model compression methods to optimize complex large models so they can run effectively on embedded systems.
Some players in the chart offer both AI tool chains and training datasets. OEMs lacking in data might consider sourcing it from external parties.
How should you react?
Prepare
All intelligent AI models need to be trained with quality data. Look internally for where this data can come from or explore third party options to acquire data.
Prioritize
Identify weak points within your company where automation and efficiency can be improved. Consider different types of AI models to use depending on the use case.
Monitor
See how the AI model performs and adapt as needed. If AI was implemented in a consumer facing use case, respond to customer feedback to improve the model/system.
Interested in finding out more?
Most of our work is helping clients go deeper into new challenges and opportunities through custom projects. If you would like to discuss recent projects we've completed relating to Artificial Intelligence, contact us today!
Also, be sure to view our related content:
