AI for AAC: Can AI make assistive communication more human?

One of Zoe’s most memorable recent conversations was with an AI bot. A new colleague of hers had created it to handle basic conversations on his behalf. As someone with a speech impairment, he had grown tired of answering the same questions when meeting people for the first time.  

Before she started working with him, Zoe decided to test the bot. They talked out loud for half an hour, with the bot using a digital clone of her colleague’s voice. She learned a lot about him in this time, and it felt almost as if he was learning about her, too. When she met the man himself, she had the sense of having a shared history with him, though he, of course, knew nothing about her. 

“Anyway, we laughed about it,” said Zoe. “The whole thing was an experiment, but super interesting.” 

As a global product manager responsible for software, Zoe, along with her team, does a lot of experiments. They explore different forms of AI to see how they can enhance communication apps like TD Snap and TD Talk.  

While AI can’t yet understand the history between two people, there’s a lot it can do to make assistive communication more personal and easy to use. We called Zoe to get her take on how Tobii Dynavox is leveraging AI technologies to create a better experience for people with complex communication challenges. 

Note: Some answers have been shortened for conciseness.

We’ve been using AI for over a decade in our products. Why does that experience matter? 

It matters because it explains the approach we take to development. We only use AI when it actually solves a real user need. There are lots of user needs, and not all of them require fancy technology. Sometimes the most basic, logical solution can be there instead of a fancy algorithm. It's super important to have that expertise around it. 

How do we decide which AI technologies to adopt? 

It’s really based off user need, but then there is what the market demands as well. 

What’s your favorite way that we’re using AI these days? 

It's definitely in the development process itself. When we have an idea, it's quite easy to get AI to sort of mock it up. But maybe something a little more obvious is voice technologies. 

What can AI do for digital voices these days? 

If you don't create your voice yourself, if you don't clone it, but you want something that's unique to you, you can tweak something until you think it sounds just like you. A nice example is we've been able to make gender neutral voices, so a blend somewhere between a male and a female.  

How has AI made AAC more comfortable to use for people with complex disabilities? 

We've explored something that you could refer to as computer vision. The system is able to pick up at a pixel level what image it's supposedly looking at. And that helps when you have challenges with very small details.  

What are some steps we’ve taken to ensure we’re using AI ethically? 

We haven't put a lot of generative AI into our language systems for our pediatric users. We want them to be able to learn their system and know how it is going to behave each time they interact with it, so that they can learn and acquire language. 

For example, TD Snap Core First and the Motor Plan use a form of AI, deep neural networks, to be able to correct word form based on grammar. 

There’s a little bit of a predictive nature, but it's not generative AI.  

What are some aspects of communication that AI will never replace? 

Body language. We've had some people discuss creating an avatar, because what happens with people with motor impairments, at least, is that their facial muscles are not animating the way that they previously did. But that is very difficult to replicate with AI.  

Our relationships are built on so much context. We’ve tried exploring context with audio. If the AI can hear the conversation, it could try and get there. And then we thought we could layer it with some of the history. But it doesn't know your full history. It doesn't know that one time when you took an airplane somewhere with your family and something really funny happened. And so much of that can be built into interactions that we have with loved ones. There's layering that happens. 

The human experience is so complex. How do you replicate that? 

Learn more about the use of AI in assistive communication