
Dr. John Pierce discusses how AI might be used to help people with aphasia, clinicians, and researchers, as well as the risks.
Notas del episodio
Show Notes: Episode #TK
AI and Aphasia with John Pierce
Interviewer info
Lyssa Rome is a speech-language pathologist in the San Francisco Bay Area. She is on staff at the Aphasia Center of California, where she facilitates groups for people with aphasia and their care partners. She owns an LPAA-focused private practice and specializes in working with people with neurogenic communication disorders. She has worked in acute hospital, skilled nursing, and continuum of care settings. Before becoming an SLP, Lyssa was a public radio journalist, editor, and podcast producer. In this episode, Lyssa Rome interviews John Pierce about AI and aphasia.
Guest info
Dr. John Pierce is a speech pathologist and researcher at the Centre of Research Excellence in Aphasia Recovery and Rehabilitation at La Trobe University in Melbourne, Australia. John worked for 12 years as a speech pathologist in neurological rehabilitation before completing his PhD in 2021. His research focuses on improving aphasia rehabilitation through innovative treatments and technology, with a particular interest in the potential of artificial intelligence to improve access to personalised therapy. John has published research exploring how AI is being used in aphasia assessment and rehabilitation, including the use of generative AI to create therapy materials. He is also involved in international efforts to guide the future development and responsible use of AI in aphasia research and clinical practice. Alongside his research, John developed Aphasia Therapy Online, a free therapy website available in twelve languages that was used by more than 75,000 people in 2025. His work aims to translate emerging technologies into practical, accessible tools that support clinicians and improve rehabilitation opportunities for people living with aphasia.
Listener Take-aways
Differentiate large language models from other branches of AI, and identify which have the most relevance for supporting life participation in aphasia.
Evaluate the privacy, bias, and accuracy risks of AI tools, and how accessibility concerns and culturally stereotyped output can affect people with aphasia.
Describe co-design principles that involve people with aphasia and clinicians as partners in choosing and evaluating AI tools.
Edited Transcript
Lyssa Rome: Welcome to the Aphasia Access Aphasia Conversations podcast. I'm Lyssa Rome. I'm a speech language pathologist on staff at the Aphasia Center of California, and I see clients with aphasia and other neurogenic communication disorders in my LPAA-focused private practice. I'm also a member of the Aphasia Access podcast working group. Aphasia Access strives to provide members with information, inspiration, and ideas that support their aphasia care through a variety of educational materials and resources.
I'm today's host for an episode that will feature Dr. John Pierce, a speech pathologist researcher at the Center of Research Excellence in Aphasia Recovery and Rehabilitation at LaTrobe University in Melbourne, Australia. John worked for 12 years as a speech pathologist in neurological rehabilitation before completing his PhD in 2021. His research focuses on improving aphasia rehabilitation through innovative treatments and technology, with a particular interest in the potential of artificial intelligence to improve access to personalized therapy. John has published research exploring how AI is being used in aphasia assessment and rehabilitation, including the use of generative AI to create therapy materials. He's also involved in international efforts to guide the future development and responsible use of AI in aphasia research and clinical practice.
Alongside his research, John developed Aphasia Therapy Online, a free therapy website available in 12 languages that was used by more than 75,000 people in 2025. His work aims to translate emerging technologies into practical, accessible tools that support clinicians and improve rehabilitation opportunities for people living with aphasia. John Pierce, welcome to the Aphasia Access Conversations podcast. I'm really glad to be talking with you today.
John Pierce: Thanks, Lyssa.
Lyssa Rome: We often start the show with an "aha" moment. I'm wondering what made you want to research aphasia or work with people with aphasia and why have you turned your research toward technology and AI and aphasia?
John Pierce: Yeah, I mean, as an SLP, I worked in, with all sorts of different conditions and mainly in rehab. And aphasia just had such an impact on me, just seeing the, all the various things that could impact across life and the changes in identity it caused in people, even in what we would call very mild aphasia, it still changed the way the world saw those people and therefore how they saw themselves. So when I, at a certain point, decided I wanted to get into research, aphasia was sort of a really obvious topic because there's so many things to do—so much work to do in that space.
As for technology, I mean my PhD wasn't actually on it, was nothing to do with technology. It was about intensity and multimodal aphasia therapy, but many of us find ourselves doing things that we didn't necessarily plan to and I think you know, I've always been a bit of a nerd, as my wife often tells me. So I've got a sort of natural affinity for technology and many years ago when I was a clinician I learned enough about software when I made Aphasia Therapy Online so I kind of know enough about tech to speak the language a little bit and to understand what's possible or not.
And then after my PhD I was in a post doctoral position on a project called Communication Connect led by Miranda Rose and in that project we were collaborating with a team who worked in AI at our university to develop this digital self-management platform. And once they started showing me bits and pieces of what AI could do, I mean, this was back when GPT-3 had come out, so it was before it was chat GPT, they were just showing me and I started to look into it. I just thought, there is so much potential here for aphasia. So that's how I've kind of found myself where I am now.
Lyssa Rome: All right. So before we dive into this more, we're going to acknowledge for the moment and for this conversation that we're somewhat setting aside some of the big picture concerns about AI and the environment, future of humankind. I do want to talk some about the risks and challenges of AI and other technology though, especially as they relate to aphasia.
John Pierce: Yeah, and you know, I'm certainly well aware that there's a whole spectrum of attitudes and beliefs about AI across the world and no doubt with the listeners. So right up one end, there are honestly people out there who think AI is all positive and it's going to be the end of us having to work, and 100% positive. And then at the other end, people who think it's all negative and they refuse to engage with it. And probably most of us are in the middle, and that includes me. I'm in the middle somewhere.
There are environmental challenges, there are ethical challenges, and unfortunately the people running the show are for-profit companies. And even where they're acknowledging risks and negative outcomes, they still seem unwilling or unable to slow things down. So it's definitely real. I guess the key for me is that this technology probably isn't going away, even if we all took a vote, it's just not going away. So if it's not going away, why shouldn't we be using it for people with disability or chronic conditions? Like, how can we make it do what we actually want and not just increase profit margins and get rid of staff? So that's kind of the perspective I take. I certainly don't say that it's all good, but I think they can be good.
Lyssa Rome: Yeah, that makes a lot of sense.So maybe we should take a step back and talk about a general overview of AI. When people think about AI, I think they first think of large language models or LLMs, bots like Claude and ChatGPT and Gemini. Maybe you could situate those more familiar types of artificial intelligence in a larger definition.
John Pierce: Yeah, you're right that these chatbots that we all know are almost synonymous when we say AI, but that's kind of a branch of a branch of AI really. It's a slippery term, but I mean, artificial intelligence means kind of having computers do things that normally a human can do, so that kind of evolves over time.
Just some other kind of branches of this technology that we forget is also AI. Chatbots are part of, they're built on large language models. Large language models are part of natural language programming, which is a branch that really has AI computers trying to process and understand language in the ways that humans do instead of just really structured with code. So that one's got quite an obvious application to aphasia. But it falls within a branch that you might call generative AI, where AI might produce images and video and attempt to produce music, produce human sounding speech, and of course text.
Some other sort of goals of AI is around perception. So computer vision is one where, you know, if you've got a doorbell that tells you what's at the front door or tells you who's there, that's AI, detecting gestures or even just understanding speech.
And then perhaps a branch that people kind of forget about is around machine learning, which is using computers to plan and make decisions. So things like, you know, feeding a model a heap of data and asking it to diagnose or assess aphasia, that would be machine learning and it would have nothing to do with a chat bot necessarily.
Another would be social intelligence. So can we have computers sort of detecting emotions? Can we get them to understand theory of mind and inference? That could be really helpful for people with perhaps right hemisphere stroke or brain injury.
And probably the most distant from AI is robotics or embodied AI. So computers moving around the world as we've all seen, kind of terrifying different robots.
But I guess a lot of that's starting to merge. like we're seeing chatbots that produce images, understand images, use speech, use text, a lot of this is becoming multimodal AI. But I guess it's just important to know that AI doesn't have to be ChatGPT or Claude.
Lyssa Rome: Yeah, that makes sense. So what are some of the promising use cases for AI for people with aphasia specifically, maybe also for speech therapists in assessment or diagnosis and treatment? How could people with aphasia or their care partners use AI for self-management?
John Pierce: I guess I'm coming at it from the SLP perspective, but it's also important that we soon understand what do people with aphasia want. And I'm working with Matthew Berryman, who's a stroke survivor and software engineer. He's interviewing people with aphasia about what do they do with AI now? What do they want to do in the future? But LLMs, these large language models, fundamentally, they are predicting the next word in a sentence. Even though they can do pretty amazing things now, that is the mechanism that underlies what they do. So to do that, they've kind of needed to get the hang of how humans use language, the patterns and the meaning of what comes before. And so as a result you know most of us have seen they can understand the intent, usually, of human language, produce output that usually passes as human. And those are both things that people with aphasia may struggle with, sort of comprehending language and producing the next word. So that capability is there's a very obvious application and opportunity to help people with aphasia.
So there's a number of groups looking at could AI help within conversation? And we've explored this a little bit. So could it fill in the blanks when someone can't think of a word, could it correct an error either out loud or on a screen for the person with aphasia, could it recast a sentence without the errors or expand on it? Or maybe a system could generate images during a conversation to help with comprehension. So there's a lot of within conversation tools that you could imagine.
Another one on the SLP side, I think is discourse analysis. It's been known for a while that looking at just word-level language is really pretty reductionist and doesn't give us the full picture of communication. But at the same time, SLPs don't have time to transcribe a conversation and then analyze it with whatever metrics. So AI can or should be able to soon do both of those fairly accurately. And I think we'll be seeing tools for that very soon where you can sort of immediately analyze a complex conversation and not just words or sentences. That's pretty exciting for the discourse researchers out there.
Another one is around assessment. A lot of papers are looking at automating, detecting the presence of aphasia, the severity, working out the subtype, or just describing the error profiles. Perhaps that could be helpful for tracking progress over time, or particularly in progressive aphasia where there's very specific subtypes, helping with that differential diagnosis. That's one possibility.
In a similar vein, there's a branch of machine learning called unsupervised learning where you give a huge amount of data to a system. It can have many, many, many variables, like more than humans can keep in their head at the same time. And unsupervised learning, you don't tell it what you're looking for, you just say, can you find some patterns? Can you find some clusters of similar kind of presentations? So it's possible. A couple of papers have tried this. You give it a heap of aphasia data, like scans of brains and assessment results and transcripts of conversation. And you say, what are the patterns? What are the different sort of clusters of aphasia? Because it doesn't seem that likely that the Boston classification is giving us the full picture. There might be other ways of arranging different presentations in a way that's more meaningful. So that's a space where we can see big changes in the way we understand aphasia.
Another is sort of around recovery prediction and selecting therapy. There's a few teams across the world who are gathering huge databases. And the idea is that they learn from that and then putting in a new person, they can say how much is this person likely to recover, which therapies are most likely to be effective? You know, what's the best approach to take? And assuming that was quite accurate, that could be helpful for people to know what's coming up.
I think in terms of sort of impairment level aphasia therapy software, you know, some apps already adjust difficulty using AI automatically and I think we'll see more of that and more automated feedback so that when you practice a word or a sentence, the app actually tells you what it thinks it heard and gives you feedback.
And perhaps clinicians have already started using AI to produce words and sentences and images for therapy that are more meaningful for the person. And that's particularly helpful, I think, in people with milder aphasia, where they need more complex things to practice than just words and sentences.
Lyssa Rome: I know in my practice, I use AI quite a bit to produce accessible reading material that's at the right level and as on topics interesting to my clients. So easy to have someone just choose an article that they wish they could read and make it the right reading level for them. I'm curious about other possible use cases that are more participation oriented.
John Pierce: Well you've touched on one there that's interesting, like a lot of us are sort of using AI to summarize long texts when we don't feel like we want to spend the time reading it but for people with aphasia that's fantastic for accessibility. So it might be emails or long texts or websites that can be put to a level that they can and more comfortably read. So an example, Sarah Wallace in Australia is leading a project to automatically simplify websites, with a push of a button, to the preferred kind of level of the person with aphasia, which is just fantastic.
Yeah. I guess on the flip side of that, we're hearing people with aphasia also are using LLMs to produce text or put a draft in and have that corrected or expanded. And I know there's a group in Sweden who looked at a co-design tool to help with understanding emails, responding to emails, but still keep keeping the person's own voice involved. So that's really interesting.
A couple of other examples, practicing conversation. Early on we saw technology called EVA Park where people would practice communicating in kind of a virtual environment like a 3D game. The next level of that is, I'm working with the team in Adelaide looking at immersive virtual reality to really simulate the environment, practice with an AI agent, and sort of feel a lot more confident about communicating before you go in and order a coffee or whatever it is you want to do.
And perhaps the last one is mood management. We know that depression, anxiety and other mood challenges can just be huge in aphasia and they're just so prevalent. And yet services out there are not very accessible and it can be really hard to assess mood. So one thing we've developed—we co-designed this with some stroke survivors—it's an app called Mood Compass, and it's just a really aphasia-friendly kind of mood tracker. So it just uses simple faces and people can adjust the face until it reflects how they feel. AI kind of keeps an eye on things over time and flags when maybe their mood has been gradually declining or there's a sudden drop. And it suggests, "Hey, you might want to look at this video. You might want to, you know, can we email this, your carer or important other person?" Or if it's really bad, it might say, "Let's export your mood history and we recommend you print it and take it to your doctor." So I think some of that self-management and people with aphasia being able to take control of them. mood and other symptoms is a really big opportunity. Yeah.
Lyssa Rome: What are some of the barriers to putting these kinds of tools in the hands of speech therapists and people with aphasia now?
John Pierce: Yeah, I think there's a couple of major bottlenecks and two of them are technical. One, I think the big challenge is speech recognition. So it's just very hard for AI to understand. understand anything slightly different about people's speech. And in aphasia, things like word errors, whether it's neologisms or phonemic errors, pauses, incomplete sentences and repetition, it just doesn't know what to do with them, when a human does. So you know, I think that's a challenge. And like in the case of a word that doesn't exist, a neologism, speech recognition will just find the nearest real world, because that's what it's trained to do. It'll say, "Oh, you must have meant this. "When actually, you know, a human would say, "Oh, that's not what you meant. And it's obviously not a real word." And that's to say nothing of having an accent, having dysarthria, or slurred speech, or apraxia. So I think AI has a lot to learn there. And if we can improve that, we might see a lot more tools actually landing in the hands of the people who want them.
And the other one is just having the data. So to train really good machine learning and even speech recognition, you need a lot of samples. And we do have something called Aphasia Bank, which is a godsend. It's got hundreds of samples of people with aphasia, but probably to train AI, we n