I recently interviewed Shyamala Prayaga, Autonomous Digital Assistant Vision Lead at Ford. She discusses some of the unique challenges faced when adding voice applications to vehicles, why humanizing privacy and ethics is so important, and some of the data privacy issues related to voice applications. Shyamala also participated on the panel “Meeting the challenge of Data Privacy & Ethics for AI” during the Data for AI Week Virtual Conference. The panel is available to watch on replay If you were not able to make it live.
In an interview for this article, Shyamala goes into more detail about various topics discussed in both the podcast and panel.
Many people may not think of Autonomous Voice Assistants in vehicles. What are the use cases for voice assistants in vehicles?
Shyamala: Like I always say, “when the vehicle becomes a driver, the voice becomes a companion”. When people are driving in autonomous vehicles, the natural modality to speak is voice, hence the need for a digital assistant becomes more of a necessity than a desire. Right from answering your vehicle-related queries to entertaining you, there are many use cases autonomy will adopt from traditional digital assistant use cases. With growing trends in digital assistants, an autonomous vehicle will open more and new use cases for customer care, concierge and commerce-related features. I would not say these use cases are unique to autonomous vehicles, but autonomous cars open more opportunities for the utility of these features.
What do you see as some of the unique opportunities automotive companies have around AI and voice assistants?
Shyamala: The foundation of any artificially intelligent product is data. The more data AI has about the users, and their surroundings; the smarter and personalized the experience will be. Imagine a voice-enabled tour guide recommending a must-try restaurant on your next trip to Shanghai? or Imagine a voice-enabled assistant booking your favorite concert ticket while you are on a business trip to Boston. Artificial Intelligence can provide you with that level of experience within the automotive space. Another opportunity for the digital assistant is portability. Again, not this is unique to the automotive industry, but artificial intelligence will enable the portability of digital assistants everywhere. Wherever we go, our digital assistant will follow us, and AI will play a major role in portability.
What are some of the unique challenges automotive companies have around AI and voice assistants?
Shyamala: Although there has been a lot of advancements in the voice recognition space, still the speech recognition engines currently work for perfect conditions which do not exist. For example, most of the time voice recognition works perfectly with an American English accent and no background noise. The moment either of these lacks, voice assistants fail at different levels. Both are unique challenges with automobiles to enable utility of voice assistants because the car is mobile, there is a lot of background noise. Some we can control, others we cannot. For example, we can control background noise to some extent but we cannot control how many accents the speech recognition engine supports. Good news is automotive companies are solving the background noise problem by installing noise cancellation microphones everywhere to enable speech recognition. And it tends to be working well. There has been some extent of analytics around voice recognition failures, which is being used by speech recognition companies to feed in more and more accents, colloquial and slangs.
Another challenge is connectivity. Connectivity is critical for an intuitive and smart digital assistant. Most vehicles come with connectivity today, but the rollout is slow due to the longevity of vehicle ownership. Those vehicles which do not have connectivity, fall back on embedded digital assistants. And the challenge with embedded assistants is they may not always have up to date information.
What steps do you suggest for companies to take, especially in the automotive industry, to get their data in a usable state for AI/ML?
Shyamala: So, imagine you scheduled a conference, and around a thousand people signed up. On the day of the event, only three hundred people joined. So technically, 700 registrations were unusable. The same is the state of any data needed to train the AI/ML. So, when sourcing data for training AI/ML models, we need to expect to have a lot of junk or unusable data. AI needs quality data to train the models. The first step is to try to get quality data in the first place. Then there is a lot of multi-stage filtering, which should be done to ensure the data is of high quality and usable. We also need to filter out duplicate data or incomplete data. Then we can leverage a combination of AI training methods.
What steps should be taken in regards to ethics and responsible use of AI?
Shyamala: Anyone working in the artificial intelligence-led project has a social responsibility and should consider ethics in the first place.
AI is dependent on a lot of data. Data about user location, their payment information, their personal identifiable information (PII) and so on. So much data comes with an equal amount of responsibility. Anyone working on AI should consider getting user consents before collecting any data. They should also consider a clear strategy for data storage, usage, sharing and deletion in place. Users should be provided choices of what data they want to share and what they do not. When people are trusting brands with their data, the brands have equal opportunity to foster trust through ethical practices.
You focus on humanizing privacy and ethics. Can you share what this means and why it’s important?
Shyamala: I believe most companies think about consents and privacy as a legal responsibility to cover themselves. That is the reason you would find some companies having multi-page terms and conditions in place for the users to agree. We all have been in that position at some point, and we all know, not many people read terms and conditions. There may be a few exceptions, but most do not.
Ethics, for me, is transparency, inclusivity, integrity, fairness and responsibility. I believe that companies should step up to humanize privacy and ethics.
By adopting user-centered design methods, we can involve users in the process of designing consents and privacy practices. Like when we design user interactions, we research our end users, their needs, goals and shortcoming and then test the interactions with the users to ensure it is usable. We should adopt the same practice as an industry to humanize privacy and ethics. We can embrace the six principles of human-centered design – empathy, ease, transparency, relationship, confidence, and delight to humanize privacy and ethics. When we put users first even for privacy and ethical design, we will not only build a usable solution, but a trustworthy solution people would want to use again and again. It will also build trust in the brand.
What are some of the unique challenges for ethics and responsible AI for voice assistants in vehicles?
Shyamala: We live in a connected world, where everything requires tight integration and communication. For example, if the user wants to close their garage through voice, right from the speech recognition system processing your spoken words, to natural language understanding, identifying your intent and sending for fulfilment a lot goes into it. There is a handoff at every step of the process to different engines. Some requests require third-party integration as well. If the request is limited to you, then it is simple, but the moment a third party is involved there are challenges. They may have their data sharing policies. How do we ensure secure data retention and deletion? I assume this is a challenge which comes with any integration, and may not be unique to automotive in general.
How important is AI to the future of automotive companies?
Shyamala: Many decades ago, the goal of automotive was to transport the user from point A to point B. Thanks to ever-evolving technology, user expectation has changed.
They do not expect automobiles to be just a mode of transport any more. They want it to be smart, connected and personalized, just as their smartphone. Artificial intelligence plays a significant role to enable a personalized or connected experience in the car. Imagine your vehicle adjusting the air conditioner based on the outside temperate or voice assistant fulfilling playing your favorite music. I believe with the growing user expectation from their cars, AI will play a vital role to enable it.
What is the long term vision and roadmap for chatbots and voice assistants in vehicles?
Shyamala: When Siri was launched in 2011, people were still learning how to use voice assistants. Thanks to Alexa, 2014 changed the mental model of users and how they interact with voice assistants. In the last six years, we have seen significant growth in voice adoption everywhere in hotels, hospitals, at home, drive-throughs and even in the car. According to the voicebot.ai 2020 studies, 129.7 million people have used the voice assistants in the car. This number is expected to grow even further.
The voice assistant space is growing every day with new and unique use cases. I think in the near term automotive companies will develop more vehicle-specific use cases. In the future I assume there will be arbitration between multiple voice assistants to enable the user experience and utility. I do feel as we will evolve use cases, there will be advancements to speech technologies. Maybe speech recognition will be able to co-exist and recognize noise instead of relying on noise cancellation microphones. Maybe voice assistants will be able to understand user emotions and respond accordingly. There is already a lot of research going on in this space.
What AI technologies are you most looking forward to in the coming years?
Shyamala: Indeed, level 5 self-driving cars. I am looking forward to riding in self-driving vehicles which can drive itself freely without being geo-fenced. Autonomy is where the industry is heading, and the future looks bright.