Bridging the Gap: AI and Aphasia with Dr. Abi Roper

In this episode of 'On the Tip of My Tongue,' hosts Rob Edwards and Jonathan Hirons delve into the world of AI and its applications in treating Aphasia, a debilitating condition affecting language abilities, following brain injuries like strokes. Joining them is Dr. Abi Roper from City, St. George's University, London. Dr. Roper discusses how AI tools, like speech recognition technologies and AI-driven text simplification, benefit individuals with Aphasia by aiding communication and daily tasks. She also addresses the limitations of AI, such as its inability to fully replicate human conversational skills and the importance of verifying AI-generated information. The episode highlights ongoing research and projects aimed at harnessing AI in therapy, underscoring AI's potential to support but not replace human therapists.

00:00 Introduction to AI and Aphasia
01:18 Meet Dr. Abby Roper
01:43 AI's Role in Assisting Aphasia
03:13 Speech Recognition and Practical Uses
05:25 Challenges and Limitations of AI
27:07 The Future of AI in Therapy
30:35 Conclusion and Resources

Transcript

Copy Transcript Copied

SPEAKER_02 0:00

The thing about AI, as with anything, really, is it's a tool and it can be used well and it can be used badly. And so if we don't consider how it can be useful or where it can be used by everyone, then there's just going to be a group of people who are missing out on it.

SPEAKER_00 0:16

In this episode, we delve into AI with Dr. Abby Roper from City St George's University London. Hi, and welcome everybody, and we hope you enjoy this podcast called On the Tip of My Tongue. Now, I'm Rob Edwards, and I'm Jonathan Hierons. This podcast is a follow-up to a film which Jonathan made about aphasia. So, what is aphasia? Well, it's a condition caused by some kind of injury to the brain, which is often could be a stroke or could be just falling off a bike. And it affects your ability to use language in all its forms. Speaking, writing, reading, sending text, whatever. 350,000 people in the UK suffer from a debilitating condition called aphasia. Fewer than half this number suffer from Parkinson's, and yet most people have heard of Parkinson's, whilst almost nobody has heard of aphasia.

SPEAKER_02 1:18

Hi, I'm Abby Roper, Dr. Abby Roper. I'm a speech and language therapist and a research fellow. I'm based at Sissy St George's University of London, and I do lots of research and thinking about aphasia and technology.

SPEAKER_01 1:35

Thank you. So that's a nice introduction. What we really want to talk about today, what I want to talk about today is AI. Perhaps you could give me a kind of an overview of what AI does for what can do for people with aphasia.

SPEAKER_02 1:52

In a lot of ways, AI has been helping people with aphasia for a long time. We've had things like automatic speech spelling correction on our phone, and that was an early example of AI. Now, obviously, we've got really clever models of artificial intelligence that have much more information at their fingertips. So they can do far more clever things. One thing that they can do, for example, is help you take a very complicated long piece of text, maybe a long email or a letter that you've received, and they can help you simplify that information and find what are the key bits so you can really get to the bottom of the understanding without feeling too tired out, if you like, because it takes a lot of energy when you have a phasia to do reading and extracting information. So simplifying information is one of the really lovely ways that it can support people with a phasia. Another way is it can help you polish things that you want to say. So if you're typing something, it might correct the sentences for you, or you could say, could you make it sound a bit more professional once you've written something that's a starting point? So it's a good for polishing and good for summarising. There are some other ways that it's useful. So I've done some research about speech recognition technologies, things like your Alexa devices and Siri. And one of the ways that people can find speech recognition, speakers and things like that useful is that you can practice using your voice to instruct things around you. So you might want to practice asking for a piece of music that you really like. And one of the nice things about doing that to a piece of technology rather than a friend or a therapist is that if you get it wrong, it's not embarrassing for you. If you get really fuss frustrated and fed up and you want to practice a bit more, you can even swear at it and it doesn't take offense. But you eventually, if you get it right, you get a nice reward in that you get your favorite song played to you. A lot of the ideas that I'm investigating in kind of research and in things like this have been have come from people working with speech and language therapists and telling them, This is my goal. I want to use AI in this way. And it's really being driven by people living with aphasia. If I can, there's another example. Speech recognition devices can do clever things like controlling other things in your house. So you can use your speech to turn off your lights or to turn your lights on. And that's great. If it's difficult for you to physically get out of bed to turn your lights off before you want to go to bed, you can ask Alexa to turn the lights off for you. Or you can link it up. There's a lady who I've heard from in the Aphasia Accessible Voices Network at the Strike Association, who said it's really helpful for her. She finds programming the central heating really complicated. She can now just ask her device and say, Can you turn the heating up? And it does it all for her. It does all that complicated programming stuff for her. So those are some of the examples of how it's being used and when it works well.

SPEAKER_01 5:18

I wish I could get someone to do my heating because I can't understand it either. That's interesting. Sometimes with people with aphasia, don't speak at all or speak very little. Now, is there a point where AI can't really help? Because uh Alexa, I assume, can't always understand what somebody is saying.

SPEAKER_02 5:41

That's quite right. Speech recognition used to be not very good, and it used to be okay if you had a very kind of normal sounding voice and you didn't really have an accent or you weren't really young or really old and you didn't ever have any problems tripping over your words. But if you had any of these other kind of ways of speaking, it was a bit rubbish and it could be really frustrating. The latest models, because it has been used so widely now, the latest models are getting much better at recognizing speech, even if somebody has a phasia. However, it's not always great if you need to pause and think about what you want to say. It thinks you finish talking, but actually you just need to pause and have a little think.

SPEAKER_01 6:29

Yeah.

SPEAKER_02 6:31

And that's something that happens a lot more when you have a phasia, is you need a little bit more time. But sometimes the devices can either jump in or switch off when you pause. Yeah. And you're like, wait, I haven't finished speaking yet. Another thing is that we found that you can actually use speech devices, even if it's difficult for you to speak yourself. Some people have a little thing on their phone which will speak out loud for them. So some people, you might have seen Lee Ridley, lost voice guy on Britain's Got Talent. So he doesn't use traditional vocal speech, but he uses his iPad to speak out loud for him. You can use your iPad to speak out loud and instruct your speech recognition technology. So that's quite a nice interaction that you can use all the technology together if you're feeling pretty advanced.

SPEAKER_01 7:24

On one of our podcasts, we had a guy who's a musician who's got very poor, no, sorry, no he's got no speech at all. Because he used to be on radio, he's there's this um this is called speak unique. And what it does is it catches all his words from previous things he's done on radio and puts it together for speech for himself. When it speaks for him, it's in his voice, and I think that's so clever.

SPEAKER_02 7:58

That's a really lovely example of technology being able to be and used for good. We can take recordings of people from when they were able to speak more clearly or more fluently and use those, even with very little information now. We can use those to train a synthetic voice that you can use. So if your iPad is speaking for you, it sounds like the version of you that you know rather than a computerized voice, which is what the case used to be.

SPEAKER_01 8:26

Yeah. It's incredibly clever, isn't it? We're talking about the uses of AI. Not all AI is, as we all probably know, is is not done particularly well. It was a the chatbot is the thing I'm thinking of. One's one's bank or whatever. Do you think that gives AI a bad name? What because we're you we're talking about the the good that AI can do. Do you think people might take uh the other side of it and say, well, actually, it's all nonsense because it doesn't help me when I go over to the bank and get my money out or whatever it might be.

SPEAKER_02 9:02

I think that's definitely the case. I think it's that it feels to me the more people I speak to about chatbots, the more universally they're disliked. That seems to be an example where people very often have an unhelpful experience. And I think the thing about AI, as with anything, really, is it's a tool and it can be used well and it can be used badly. And much technology is like this. We can have some apps on our phone that are really easy to use and really user-friendly. And then sometimes you go to a website and you're having to do something and it's horrible. And it's not that websites in themselves are bad, it's the way they've been designed and the way they've been used. And often I think we are used to expecting a certain sort of conversation if we want some help. And chatbots don't really give us the sort of conversations often that we want. They don't give us helpful answers. They give us answers that feel frustrating or like they've missed the point of what we're trying to say.

SPEAKER_01 10:05

Sounded like somebody talking to you, but it's actually a QA type.

SPEAKER_02 10:09

Exactly. And it's so it feels like you're being hoodwinked because it's as if you're having a conversation, but it's not convincing. So if we were looking at that same information on a web page on just like a frequently asked questions page, we would reconcile, we could go, oh yeah, it's okay, because I know that I'm just looking through some information, but chatbots give us this feeling that there's a human at the other end of it, and we want to speak back to them like they are a human, but they don't give us a real experience of conversation. I think bad experiences with what is AI can make people feel really put off. And I think that's why it's important for us all to recognize that it's a tool and it's being used in lots of different things, and we can have good and bad experiences, and it's not good for everything, it won't take over all of our jobs because there's lots of things that only humans can do.

SPEAKER_03 11:07

Yes.

SPEAKER_02 11:07

But chatbots being one of them, a chat interaction is much better from a human than it is.

SPEAKER_01 11:12

Yes, of course it is. Yes.

SPEAKER_02 11:14

There are some jobs that AI is very good at, and usually these are very prescribed jobs, and they can do them very well. But if you ask them to do something that has a bit more of a human flavour or something that maybe veers off of that original job that it's got slightly, it becomes very bad.

SPEAKER_01 11:29

When we're doing the other side of this podcast, we're putting together. My co-host said, Can you have a genuine conversation with somebody? And I don't know the answer. I'm sure you do the answer. What is the answer?

SPEAKER_02 11:44

The answer is you can have a very tightly scripted conversation with AI, one that you know, almost as if you're both following a very predictable set of questions and answers. And in speech and language therapy, that we know that therapy for aphasia, which works on practicing script interactions, like you can have a you can have an interaction, you say, Oh, I want to practice going to the shop to buy my local paper, and what might that look like? And I can rehearse it by taking turns with a computer. And that was before AI, but that was the computer would say its bit, you'd say your bit, computer, and it would take it in turns. But that is when you know exactly how that conversation's gonna go. In a real life conversation, we don't know what the other person's gonna say, and we can't predict, and we have to take it in turns and respond. And that's the brilliant thing about it is that I get a turn to choose what happens, and then you get a turn to choose what happens, and it's shared. And I think there is now a voice feature on ChatGPT, for example, where you don't need to do any typing and you don't just say a quote and then it responds in text. You you say something to it and it chats back to you as if it's a real person having a conversation. And I was just having a play with that before I spoke to you because I was thinking about can we think about this being useful? One of the ways that I think it's difficult and that I don't think it's very helpful at the moment for people with aphasia is that it does that thing where it interrupts you if you pause. So if I stop talking for a moment, it will jump in. It doesn't adjust or it doesn't automatically give me lots of time for speaking and thinking. And then another thing that it does is it over-describes, it gives you lots of information at the same time.

SPEAKER_01 13:46

You need, yes.

SPEAKER_02 13:47

Exactly. Or it assumes a certain kind of tone of voice that might not be the sort of person that I want to have a conversation with. Right. So the person that I was chatting with had a very kind of breezy, I could imagine them working in a convenience store in the States or something, and they had an American accent and were saying, Right, have a great day. Sure feeling we're gonna do that, and just using terms and that were a bit different to my British sensibilities. So I think we are on the cusp of things are emerging really rapidly and changing really rapidly. So I think those things will improve, but it's not it's not at a point at the moment where it can flexibly have a good conversation.

SPEAKER_01 14:23

Well, do we want it to? I uh I suggest maybe what you said when you don't, it's like kind of verbal tennis, you go backwards and forwards, you don't know what's happening next. That's the thing that's the enjoyment of conversation, isn't it? We don't really want that taken away, I would have thought. And maybe it's one of those things that you'll say, leave it, leave it alone and just carry on. All the other things sound sound, I think, are brilliant, actually, particularly Chat GPT, has a lot of good things about it and would help. It helps me, to be honest.

SPEAKER_02 14:57

Great, exactly. And I think one of the important things is that it's being used by everyone now. And so if we don't consider how it can be useful or where it can be used by everyone, then there's just going to be a group of people who are missing out on it. And so if we don't think about can we can it be useful and can it make our lives easier in some ways, either because you've got aphasia or because you're very busy or because you're very tired, or any of these reasons why we might want a little bit of help shortcutting something that would take us a long time. One of the things I'd like to find out from you is when you use ChatGPT, how confident do you feel about the answers that it gives you?

SPEAKER_01 15:40

It's a good question. I would have said, maybe if I tried it, and I have tried it in the past, I would have said, maybe. I think it's a lot better now than it was, but I would still double check it. I wouldn't take it as being read that whatever it's telling me is is the truth. Or sorry, it's not necessarily the truth, but it is been in some way managed.

SPEAKER_02 16:05

Yeah, I think that's really important, even that basic thing. So any of us who are using it recognize that it's good, but if it's really important thing, we need to double check whether it's right. Yeah, yeah. Yeah. So we can't take it on trust that the answers that it's giving us are 100% accurate. And so if it's something soft and frivolous and it doesn't really matter what the answer is, or we haven't got a big decision to make off the back of that answer, it's great. We can use that, and whatever answer it comes up with, that's just a helpful bit of information. If it's something that we're using to make a really big decision about our health or our finances, then it can give us a bit of information, but it's really important that we double check what the accurate thing is. And that's why knowing what the where the information's come from, so you can ask it, tell me what web page you've got this information from. That can be a good way of checking it up. Or you could even ask it, what does the BBC say about this? Or what does the NHS say about this? Yeah. So that you have information from a source that you can trust to back it up.

SPEAKER_01 17:10

I think one of the things I I've learned using it is that you have to be quite specific about what you're asking it. You can't give it a woolly answer. Oh, a woolly question because you'll get a woolly answer, or you'll get the answer that that they think you want. But I think you have to be quite specific, otherwise you'll you don't know where that's going to go.

SPEAKER_02 17:31

You're quite right. And I think there is there is an art to learning how to ask questions. All of us get wrong at first, and then we use it a bit more and we go, hold on. And we learn how to do it better with a bit of practice. And it can be helpful to do that with somebody who's more confident or to have some help or just some knowledge that it won't necessarily understand you at the beginning, but you can ask it to give you a bit more detail or change the way you ask it.

SPEAKER_01 17:58

My questions are getting longer and longer because uh you have to put in all the things you think you're gonna be told, or the answer is going to be the way you think you're gonna get from it. Anyway.

SPEAKER_02 18:09

That's just what you're saying though. That's really interesting because what you're having to do is come up with one really long complex idea that holds in lots of ideas at the same time and put it in as one prompt. Another way you can do it is just put in a general idea at the beginning, and then when it comes out with its general answer, you can add a little prompt to refine it down. So you can do it in stages, but you're doing like a really linguistically challenging version of that. Just test your test your language, definitely.

SPEAKER_01 18:38

Yes, and I want to get to the the guy gets the answer quicker.

SPEAKER_02 18:42

Yes, yes, and it yeah, that's right. You can do it more efficiently if you can put it all in that.

SPEAKER_01 18:47

It's a bit of a game I have with thinking how can I get this boil it down to exactly what I want from it.

SPEAKER_02 18:54

But I used to have that same experience when I was writing tweets on Twitter. How can I get this really complex thing down to 140 calories? Oh, absolutely. It was so it does use it hones your it hones now, you could, yeah. But it does hone your language abilities, I think.

SPEAKER_01 19:10

Yes. I suppose it depends on how confident you are with language, really, isn't it? Something like that.

SPEAKER_02 19:19

If you type it in and it's got if you type in something and it's got speech, uh it's got spelling errors in it, usually it doesn't matter. It ignores it and corrects it to the right thing, which can be really encouraging and helpful.

SPEAKER_01 19:35

But not everybody can spell.

SPEAKER_02 19:37

Precisely. Yeah, lots of us. Yeah, lots of us can't spell, or lots of us make mistakes when we're tired, or if we're typing in a foreign language, or spelling is or or if they're really complicated words that we haven't used before. It's one of those things that we all kind of exist somewhere on that continuum.

SPEAKER_01 19:57

The other thing is that somebody's given me a poem, and someone with quite apparently severe aphasia, and but has tried to get their message across using a poem. And I'm given to understand that the poem was made partly through using AI. Now, some people would say, oh no, you can't muck about with things like poetry, but it's a really good poem, actually. And I think it gets the job done because he wants to it is he wants to get his message about him and how he feels about aphasia, and he's obviously using a tool, as you say, to get that message out.

SPEAKER_02 20:46

Exactly right.

SPEAKER_01 20:47

Okay, do you think that's a reasonable 100%?

SPEAKER_02 20:52

I use spell check on Word when I'm writing a document. To me, it's no different to that. I know that if you think about there is the potential for that for AI to do that for somebody who's got real challenges with language, and I think that is the perfect place for this kind of generative use of AI. There is also the potential for somebody who doesn't have problems coming up with language to use it, and that's okay too, but it's less of a valuable use of that ability. And I think AI can do wonderful things, and I think if but it at the same time, it has an economic cost, it has a carbon footprint, and I think if it can be used to really make genuinely make a difference for somebody to be able to express themselves who might otherwise not be able to express themselves on a topic. And if they are happy with the output that it has produced and feels that it does represent their feelings, I think that's such a valid use of this kind of technology, way more important than a Marketing company just using it to send me spam emails and all these things. I think it's a tool, and I think supporting people with language difficulties to shape and express their thoughts and understand and communicate is like the ideal way that this technology should be used.

SPEAKER_01 22:18

Let's go back to when perhaps when you did your training, things have changed for just that very reason. It's obviously technology has moved on, but does it help the person, the speech and language therapist? Does it help them or does it hinder them?

SPEAKER_02 22:36

It's a really lovely question. And I think in the time that I've been working in technology and speech and language therapy, which is about calming on for yeah, yeah. 17 or 18 years I've been thinking about this in a research speech therapy way. And when I first started, there was lots of concern that the computers were gonna do us out of a job, which I think is a valid concern for anyone in any job position, really. But actually what we've seen is no, they won't because they can't do the things we can do. What they can do is give us a way of practicing one thing or doing one job really in a really controlled way, so we can use computers really effectively to support somebody to practice a certain therapy exercise. But what they can't do is judge how that person's feeling or take note of somebody's experience of that or what their personal goal is and adapt their experience, their human wishes to be appropriate for that person. And what it is, it's a lovely tool, and we can use it to extend and make sure that people can access good quality therapy, practice, or potentially access to everyday things, but you aren't able to do that if you haven't got someone to show you how to do it in the first place, or to help you work out how you feel about it, or what's the best way to use it. And so I think it it really does extend the things that we're able to offer to people, but it's never going to do us out of a job, she says, touch wood.

SPEAKER_01 24:19

Talk about soft skills, really, aren't we?

SPEAKER_02 24:21

So yeah.

SPEAKER_01 24:21

Yeah, which AI can't get a grip on quite.

SPEAKER_02 24:25

Exactly. And more than soft skills, I think there is that flexibility, that ability to be two things at once, not just to be there on that one job. Like the chatbots are frustrating because they've got like an agenda and they want to do sell you their holiday package or whatever. They're not used to thinking about the fact that you might have had a second thought or asked it a question that wasn't on its script. Whereas humans are amazing at that and really good, and they can hold these two, they can hold more than one view of you or more than one idea about what's good in their mind at the same time, and really delicately switch between them, which is why it's fun to be a human and not a computer.

SPEAKER_01 25:04

The other question I was going to ask you, I suppose, really, which I would oh, there was one other thing. AI, I've got it written down here, AI delusions. It's a phrase I've come across, but can you perhaps you could explain a little bit?

SPEAKER_02 25:20

This thing called hallucination, which happens when so we've already talked about the fact that the computer's really good at doing one thing and it really wants to do that well. And so if you ask it something and it doesn't really know, it's like a really confident person in a job interview. It'll make something up the first thing that comes to its mind and goes, I know that I should say something here, so I'm gonna say it, even if I it's not true. And so what that actually means is that the information that it gives you doesn't come from real evidence, it doesn't come from real information on the thing. It knows it should give you an answer here, but it can't find that answer in its thing. So it goes, this is the nearest best thing, this would fit. Normally, I should give a reference for this here, or something that sounds a bit like this should go in that slot. And so that can be really troublesome when you are looking for like factual information. And the BBC have recently done a study on this and find found out that AI can be a bit of an unreliable source for news details and things like this. And that is something when we're trying to ask a system to do something that it's not really trained to do, like you've got the overconfident person in the job interview, it's a bit like that. And that's why it's really important to do what we mentioned before, which is you always double check the information, especially if it's something that's important to you. But it's just it's them trying to do their best, but they can't respond to with flexibly and say, No, actually, I don't know that answer. It would be a more helpful response, actually, wouldn't it, if they just said something like that?

SPEAKER_01 26:49

Which I don't know.

SPEAKER_02 26:51

I don't know if I've never heard ChatGPT say that, but or any of the others. But yeah, it's interesting to think why don't they do that?

SPEAKER_01 26:58

Don't ask me. Yeah. Ask somebody else.

SPEAKER_02 27:02

Exactly. Like with your GP, I don't know.

SPEAKER_01 27:05

Ask. And what about the future? Let's talk about a little bit about the future of your work, really, because obviously this is ongoing, isn't it? What you're doing. So, what's the future hold?

SPEAKER_02 27:19

City St. George's University, where I work, there's a lovely project going on at the moment called the Mars Project. And that is researchers Madeline Cruz and Lucy Dipper working with some other great researchers, and Neve Devan and a group of other people, and a therapy company called Therapy Box. And they are looking at can we use AI to help us do therapy in a collaborative way. At the moment, technology is quite good at working out single words and saying, Am I saying a single word correctly or not if I speak to a machine? But it's less good at saying, what does this big if I want to tell you a story because I'm chatting to you at the end of my day or something, and I've been, oh, this great thing happened to me the other day. This we basically have, as humans, we communicate, we don't communicate in single words. We often communicate in this kind of story-level thing where I'm gonna relay a set of information to you and it is about feelings, it's often got like a main event and a bit of a punchline and a starting point. The therapists on the Mars project have been looking at can we get AI to support us to work with people with aphasia whose stories might become a bit disjointed because they've got aphasia and who who can tell a story and use uh speech recognition to transcribe that, to write it down into text, and then put that into another system that then says, okay, so your naming words are doing quite well, but your joining words are often, and if you have a look at this story that you've told, here are some of the here are some of the places in a really systematic way, in a really helpful way. And then you and your therapist can collaborate and look at what things do we want to work on? Where are we seeing gaps that will be really helpful for you when you're practicing your talking and beyond just the single word level, which is what a lot of research used to be about we can help you speak in single words, but actually we converse in sentences and stories, don't we? And so there's really lovely uses of it happening there in that. So they're looking at this really collaborative use of it, and that's one of the nice things that they found. Actually, it doesn't replace the therapist, it doesn't, it's really helpful for having that joined up conversation about how do we all your language together. Yeah, it really reinforces and enriches their therapeutic relationship. Yeah, so that's a really lovely outcome. And I'm looking more widely in my research about the potential for using voice recognition and AI for good across like all forms of life if you've got a phasia. You know what I was saying before about what is a really good use of this AI technology to really be beneficial for somebody, whether it's writing a poem, playing, playing with language, or switching your lights on, or practicing your therapy exercises. What's in it for us as a community?

SPEAKER_01 30:14

Fantastic. That's great. Thank you very much, Abby. It's been really interesting.

SPEAKER_02 30:20

So much, thank you so much for having me along and having a chat. I'm really excited to speak to you about it and we'll see where we're gonna. Yes, when the papers are out, when we've done a bit more research. Because it changes so quickly, doesn't it?

SPEAKER_00 30:35

Thank you for listening to this on the tip of my tongue podcast. We hope you found it helpful and informative. Now, if you want more help and information about strokes and aphasia, please go to stroke.org.uk. Say aphasia, that's say aphasia or one word dot org, or the aphasia page of nhs.uk. This has been a Buffalo Lounge production. Please follow Buffalo Lounge on all the socials.