Guest Post: AI Needs to Reside in the Vehicle to Work Well

0
Jun 20, 2023

Mercedes is running a beta program where those that opt in will be able to access ChatGPT from their vehicle by interacting with the voice assistant already present in MBUX-equipped vehicles. But rather than the cloud-based service that Mercedes is going with today, it should be looking at implementing ChatGPT directly in the vehicle. 

  • Mercedes owners in the US can enroll for the program by accepting an update for their car. 
  • The test is due to run for three months and is being supported by Microsoft’s Azure OpenAI Service, which is an API to which clients can connect their services to have generative AI functionality. 
  • Mercedes is able to implement this service very easily because all it is really doing is providing a prompt for the vehicle assistant to fill in, send it to the cloud and then read out the results. 
  • This means that all of the inference or processing of the request will be done in the cloud with the voice assistant doing nothing more than acting as a front end to provide the voice functionality. 
  • The vehicle is a use case where generative AI could have a disproportionately large impact. This is because a touch-based icon grid is a substandard user experience no matter who provides it. 
  • The problem that the car makers have is that their icon grid is much worse than Apple, Gooxgle or Tesla. 
  • Furthermore, in 2016 and 2017 we concluded that voice was the leading contender to improve the digital experience in the vehicle but that voice was not good enough to create an acceptable user experience. 
  • This is why vehicles are still limping along with smartphones embedded in the dashboard. 
  • We have also concluded that generative AI represents a significant step forward in the ability of machines to communicate with humans and provide a user interface for a digital service. 
  • Consequently, generative AI offers a significant opportunity for vehicle makers to win back the digital initiative that they have ceded to the digital ecosystems. 
  • This is extremely important as vehicle makers’ ability to monetize the market for in-vehicle digital service will be contingent on their ability to remain relevant in the digital vehicle experience. 
  • This is why Apple and Google are coming aggressively after the vehicle and so far, the OEMs have mounted feeble resistance or offered complete capitulation. 
  • The problem with this approach is that the only way to implement generative AI effectively in the vehicle is to put it directly in the vehicle. 
  • This is because reliability and speed are critical, and in this example when the network goes, the service goes with it. 
  • Furthermore, it is unlikely that there will be any real integration with the vehicle, meaning that telling ChatGPT that one is feeling hot is likely to result in silence rather than the air-conditioner being turned up. 
  • Using ChatGPT as the benchmark implementation in the vehicle will have a profound impact on the cost of the vehicle’s electronics as well as its power consumption which in an EV is a deal breaker. 
  • There are rapid developments going on in the open-source community that may make this a lot easier to achieve, but implementing large language models outside of the data center remains a work in progress. 
  • Despite the current limitations, the potential for generative AI to help OEMs to overcome their digital shortcomings is substantial and represents one of the best opportunities the OEMs have had for a long time. 
  • The risk is that if no one uses it as a result of the way the Mercedes experiment is implemented, it will lead to the (wrong) conclusion that putting it in the vehicle is a waste of time. 
  • This would lead to the squandering of another opportunity, resulting in digital irrelevance and greater commoditization. 
  • We remain pretty pessimistic about the outlook for the OEMs. 

(This guest post was written by Richard Windsor, our Research Director at Large.  This first appeared on Radio Free Mobile. All views expressed are Richard’s own.) 

Summary

Published

Jun 20, 2023