This article is part of the TechXchange: Generating AI.
What you'll learn:
- The combo of generative AI and a voice assistant that forms the foundation of the platform.
- What are AI hallucinations?
- Survey results on driver acceptance of such a platform.
Artificial intelligence is certainly receiving lots of attention these days. It’s enabling more tasks to be automated, whether driving in a car and using a smart device or through customer service at a restaurant or retail location (for example, Wendy's restaurants are using a Google AI to power their drive-thru ordering system).
On that front, SoundHound has launched SoundHound Chat AI for Automotive, an in-vehicle voice assistant that gives automakers and their customers access to voice AI technology. The platform is said to be the first product to offer an in-vehicle voice assistant that combines generative AI capabilities, like ChatGPT, with an established voice assistant. The result is a hands-free solution that gives drivers and their passengers access to an array of information domains enabled by complex conversational capabilities.
SoundHound is said to integrate with dozens of knowledge domains, pulling real-time data like weather, sports, stocks, flight status, restaurants, and more. That’s combined with large language models like OpenAI’s ChatGPT to deliver accurate, timely, and comprehensive responses.
Integrating generative AI and large language models with software engineering modules helps solve AI hallucinations and delivers real-time content from knowledge domains.
AI Hallucinations
Using a proprietary approach, SoundHound Chat AI for Automotive is said to avoid the kind of misleading and unpredictable responses now associated with some large language models (LLMs). Known as “AI hallucinations,” these incorrect answers can result from technology that relies exclusively on generative AI. Such incorrect results nevertheless sound amazing and thus can be quite dangerous.
SoundHound said it reduced the undesired AI hallucinations by over 90% to a negligible amount compared to a prominent off-the-shelf model. The company also claims the platform can detect when machine learning is providing an answer that's incorrect. Or when there’s the need to go to external APIs or get real-time information from other sources.
Applying a mix of software engineering and machine learning, SoundHound’s said its platform is able to intelligently select the correct response from the most appropriate domain. It could be a generative-AI-powered answer, or the kind of response LLMs can’t handle, like real-time questions about navigation, weather, sports, stock, and flight status.
Two AI Models
The company employs two models. One is called CaiLAN, the other is called CaiNET. CaiLAN is conversational AI language, while CaiNET is conversational AI network.
The first one is the software engineering approach, and the second is a machine-learning approach. Together they get the strength of machine-learning models that can scale really well and the benefit of software engineering approach that doesn't suffer from unpredictable and undesired consequences.
In Q1, the number of cars shipping with SoundHound's technology more than doubled year-over-year, according to the company, and they expect the growth to accelerate as more brands sign up. Strategic partnerships include collaborations with Harman, LG, and Qualcomm.
Survey Takeaways
To accompany the launch of SoundHound Chat AI for Automotive, SoundHound also released a preview of key takeaways from a new research study into the technology preferences of regular drivers. This CARAVAN survey was conducted by Big Village among a sample of 1,008 U.S. adults ages 18+ who drive at least once a week. The survey took place from April 5-16, 2023.
Among the major findings were:
- 50% of all regular drivers said they would be likely or very likely to use an in-car voice assistant with generative AI capabilities.
- 75% of regular drivers who frequently use in-vehicle voice control said they would be likely or very likely to use an in-car voice assistant with Generative AI capabilities.
- Over half (57%) of regular drivers under 35 said they would be likely or very likely to use more sophisticated in-vehicle voice technology, with ChatGPT-like responses.
- Almost a third (28%) of drivers who currently rarely or never use their in-car voice assistant would be likely or somewhat likely to use it, should they be able to access generative AI.
- The most popular uses for voice assistants included navigation, radio, local search, making and answering calls, and text messaging.
The most desired queries requiring more sophisticated generative AI functionality included vehicle maintenance and rules of the road, as well as questions about sights or attractions at their destinations, recommendations for scenic routes, and other road-trip/car-outing related questions.
The full results of this consumer study will be discussed at a webinar on June 1, entitled “How Generative AI Will Help OEMs Build Great In-Vehicle Experiences.”
Read more articles in the TechXchange: Generating AI.