AI

0:05S1

This is Studio 1 on Vision Australia Radio.

00:12S3

Hello, I'm Sam... 

S2

And I'm Lizzie.

00:14S3

And this is Studio One, your weekly look at life from a low vision and blind point of view here on Vision Australia Radio.

00:19S2

On this week's show....

00:20S3

We talk Artificial Intelligence with David Woodbridge.

00:24S2

As we always say at this point, please do get in touch with the show. Whether you have experience with any of the issues covered in this week's episode of Studio One, or whether you think there's something we should be talking about, you never know. Your story or insight may help someone who's dealing with something similar.

00:39S3

Please contact us by email. Studio one at Vision Australia. Org. That's studio number one at Vision Australia - dot - org.

00:45S2

Or you can drop us a note on our Facebook page at facebook.com slash VA Radio Network. We want to hear from you.

00:55S3

So happy Easter, or Easter that was.

00:59S2

Happy Easter. I hope you had a fantastic time with friends and family, I certainly did.

01:03S3

We went to Perth and caught up with a whole lot of old friends. And I'm talking old friends. And yes, we had a fantastic time and my football team won, which is even better.

01:14S2

Yay!

01:15S3

So this week we are catching up with a friend of the old friend of the show. Now, when someone says artificial intelligence, what do you think of?

01:22S2

I think of the image detection software that Facebook or other platforms use to try and describe an image, and they're not always correct.

01:32S3

I tend to think of because I'm a sci fi geek. Of course, of all the robots and computers you see in TV shows and things like that. But then again, I'm an old man and all this new stuff is completely beyond me. So we are once more joined by technology supremo David Woodbridge. Hello, David.

01:56S4

I had just had the check out with my shoulder you were talking about. Hello.

01:59S3

So this week we are talking about artificial intelligence. When I think of artificial intelligence, I tend to think of, I don't know, Hal 9000 from 2001, or maybe Kitt from Knight Rider, but we're not really talking about that yet, are we?

02:14S4

No, but I could quite honestly talk about the how 9000 or Kitt, which is my favorite, favorite car. But no, we're not. Sadly we're not. We're talking about different types of other AI. And I should say up front, this is not the same as, like, people think we're talking about AI. Even the other thing that jumped into my mind the other day was that very old, I think it was called Eliza on the Apple. Chewie, where you talk to this supposed psychologist all about your feelings. And it was also very helpful. So the AI talk about AI, we're talking about we're talking about not only things like ChatGPT and Copilot and Gemini and all the other ones, but we're also talking what I can do as far as things like object recognition, scene detection, what's in photos, what it can do with your computer to help you along, how I can rewrite documentation for you or help you to rewrite maybe a snarky letter when you're trying to be a bit more polite to a person, or you want to write a humorous email. So it really does cover lots and lots of areas.

03:23S2

Well, we've already got, a lot of apps on our iPhones and androids that I suppose use AI to do things like color detection or text reading, don't we? Does that does that count as AI?

03:35S4

Yeah. Look, we do. Yeah. And look I this year I, I really want all those things to get a little bit bigger and a little bit more aggressive in, in what they're doing. Because at the moment it's sort of a little bit on the, on the sort of the, the dummy side of things. Um, they've still got a long way to go. I always view anything to do with AI with a grain of salt, because as they always tell you when you ask, for example, ChatGPT a question, never 100% believe what it says. And the same thing goes for objects and detection and all that other stuff as well. So one really sort of good human example of I guess I with human intervention is just the new AI feature from IRA, uh, in the States, and that's the video help service. 

So what you can do with this excess AI, you can take a photo like you normally can on your iPhone or your Android phone. The AI will then tell you what's in the picture, but then what you can also do is then ask a quote human to say, is it really true that I have ten dogs? And an elephant holding an umbrella whilst sitting on an esky at the beach, so you can double check that you know, the I got it, got it. Right. But um, so that's one really good example of human intervention. Just to double check the facts of what the AI system is giving you.

04:59S3

So let's go back though slightly, what is the difference between, say, AI and just your normal standard... algorithm or computer operation?

05:12S4

So the way I tend to envisage this is, I sort of look at it like the human brain. So what it does, it looks up this huge amount of data that it's collected from lots and lots of resources, looks at the keywords that you've, you've said and then tries to work out from a hierarchy on what you might have been asking about. And the way I like to think about is if I just said to, um, Google search, find me information about guide dogs, then it's going to go and basically look up websites, whereas if I asked an AI system to look up guide dogs, sometimes it might even come back and say, well, look, sorry, but what what specifically do you want to know about Guide Dogs? Or there's lots of information I can tell you about Guide dogs. Or it might just say, um, guide dogs, especially trained dogs for blind or low vision people. What in particular would you like to know more about? 

So I get the feeling that you're almost and you're not yet, but you're going to feel like you're talking to a, quote, real person because you can narrow it down. And I've found a lot of information I literally had, I think it was a 45 minute chat with ChatGPT about how an aircraft flew because, you know, I always wondered, you know, what the shape of the wings were and the flaps and the things at the back of the the aircraft on the tail section to keep it, you know, flat horizontal movement, all that sort of stuff. And it was just amazing because some of the questions it asked, I went, oh yeah, that's right. And so I was able to drill down a lot more. So I actually can't stand using just quote search anymore. For most of my information I use ChatGPT, but then I also verify it. 

And one really cool example was and it's better now. I asked it what the letter E was in Braille. Now for some odd reason, a year ago it got it wrong. It reckoned it was some weird thing. Like I think it was like .16 and three or something. Of course, it's actually not 1 in 5, so I still take things with a grain of salt, but I am definitely learning a lot more because you really are going to that sort of hierarchical conversation stuff rather than just getting all this flat information, just, you know, pretty statically set in front of you.

07:33S3

What I've also found, though, sometimes, especially with smart speakers, for example, they jump to conclusions as well, and sometimes quite often. Yes, it's exactly right, but it can be miles off what you actually were asking for. You might be asking for a particular song, and it plays something completely different. It's, so I'm wondering if that flows through to any other AI systems.

07:54S4

Oh no, I'd have to say so far, the AI stuff that I've been using and I've been using sort of all three of them or four of them at the moment, they're actually very good. I haven't found anything that it's gone completely off tangent. Sometimes it might have got slightly or interpreted the word that I was using because it was a very unusual word, but besides that, no. And I have it normally comes back like, you know, here he goes, this is what I found on the web, or I didn't quite understand what you said. It's like you're actually you've got no idea what I'm actually talking about, at least with ChatGPT. And so it comes back and says, I think you're after blah, blah, blah. Can you clarify or ask me a different question so I know what you may be searching for. So again, you've got that sort of more interactive element going on rather than going, you stupid crappy piece of, you know, personal assistant, go away and leave me alone.

08:50S2

It's interesting you were talking about getting an AI to write things for you, because on Facebook now you have the option. When you write a post, you can use this AI feature, and it will rewrite it based on the mood that you want the post to project. So I wrote a happy birthday post to my husband, and I wanted it to be funny. And I used AI and it inserted all these emojis and reworded the whole post and everything. I thought it was brilliant. Yeah, yeah.

09:15S4

No, no it is. You can have some really good fun, AI... there's a new service that came out. I think it's been around for a while and I can't remember the exact name, but I think it's. I want to say so in new or something like that, but it's where you can give the system a topic to then compose music and words to. So I put in, um, I love to training karate every day. I am looking forward to getting my next belt, blah blah blah. And it came out with about four different versions of karate songs. So it came out with sort of a high end, you know, every day we go in there and fight to do our best. And another one was, I feel very relaxed when I put my gear in the morning because I'm getting ready to do my others and that sort of stuff, but it was just amazing. 

So it's almost like it's really opened up the world for our imagination. So I really think I can remember in the science fiction book that it said, the human brain is like a very wild jungle of all different types of effects, and somehow your brain dives into the jungle, rummages around, and try and get some consistent bits of information and brings it back. And there's your thought, or there's your idea or there's your imagination. I thought, that's the really the way that I almost feel like I is going at the moment.

10:35S3

So apart from a lot of the, the fun things, what are the more practical things that we can use, especially when we're talking, you know, people who are blind or vision impaired. What do you think are the most useful applications we can have here?

10:48S4

Well, I think to start off with, let's let's not forget about the Bowie's AI assistant. So, of course, that's where you can take a photo. And, you know, back in the old days, as of about six months ago, um, it might say, you know, there is six people sitting on a beach. Well, now what the. I will say there's six people sitting on the beach. The first person is looking towards the second person. The first person is wearing sunglasses. Next to the second person is a young child and so on and so on. So you really get this extremely in-depth idea of what's in the photo. So that's number one. Number two is the amazing feature that's now in jaws 2024 called picture AI. And again, it's where you can interrogate whether it's a diagram, a chart or a photo and get much more detailed information about that type of stuff. So that's also extremely amazing as well. 

The other AI stuff, which I still. Hold out a bit of trepidation about is when it comes to things like object recognition and object detection, scene detection. So when the system says, you know what's around you, what the objects are and so on. And I think that's where it scares me a little bit, particularly when you've got organizations saying, you know, you can travel around with this AI thing and it will tell you what's around you. Well, I still reckon it's not quite there yet. So, for example, it still reckons the back of my wife's red Subaru Forester is the back of a bus. Oh, um, excuse me, I think talking, um, it's the back of a bus. And then it also reckons that my red bin is a fire hydrant when it's actually on the street. 

So there are just little things like that that are wrong. And I know, of course, from common sense that it's not correct. But yeah, I just, I just get nervous. And then of course we go into all the new things coming up that are based on, quote, self-driving unquote technology. So you've got lidar, radar, infrared motion sensing, gyros, all that sort of stuff. And whilst they're well and good, you still got the human side of things. And one thing in particular that I'm a little bit nervous about is the new robot type of only thing, which is basically a little robot on two wheels, like a vacuum cleaner. You hang on to the handle, and it will literally steer you down the street and go around obstacles and all that sort of stuff. And my, my thought was, gee, I wonder how that would go on a railway platform if I was just following or hang onto this handle with these two wheels rolling around to me, is it going to be smart enough to say, by the way, mate, if you go to the left about another 30cm, you're going to step off the platform. 

So they're the type of things that, you know, it might be okay while you're you're rolling around on an actual, you know, a paved street or in a shopping centre. But it's when you get to those danger situations where you can explore. If you cane or your guide dog, or seeing a dog will take you away from an obstacle. They're the types of things that I get extremely worried about, particularly when they say, oh, you don't need your your guide dog anymore. Don't use this device.

14:01S3

It depends really, who's designing the things with, I mean, finding that certain apps and things like that work a lot better if the person designing it is actually vision impaired or blind themselves. That's the way I sort of see it. It's kind of like you see in certain street environments where the footpath has been designed by someone who's never got out of their car. So that's kind of what I think is concern there. 

14:24S4

Yeah. Although the gliding system has been developed by a guy who's got up, he so obviously he's losing his sight, his low vision. He's had a few key people on there demonstrating device, one of which is Mike May, whose he may remember his very famously brought in the the stuff to do with GPS standalone units a while ago. So they've got sort of very sort of, you know, socially influential people on there. But yeah, I, I just know how dangerous it is when like, for example, when I travel around here on the Central Coast in Gosford, there's no footpath. So I've literally got to walk on the road. Now, the problem with walking the right, of course, is not only do I have no gutters, but I also have lovely culverts in the side of the road that I could actually step into. And they're the types of things that worry me the most.

15:12S3

And... cars and things like that. So I mean, yeah. So it's still a long way from... artificial intelligence... taking the place of a guide dog or even a cane, necessarily, or even our own common sense, I'm guessing. 

15:26S4

That's right. Yeah. I mean, look, I would say, I mean, I know some people are talking about, you know, maybe next year, but I'd say give it another good solid three years and minimum. Then I might have, you know, more, more confident. But the other ones that are really good too, I think a lot like those indoor apps that will map it. Indoor environment for you. I think clue's one of the major ones where it will follow your steps. Then you can trace back. So if you've gone into a building, go to the reception, going to the left. Come up to the seventh floor, go around a few corners, go to the conference, and then you've got to go back downstairs again. You can retrace your steps. So that type of stuff is actually really, really good. 

And even getting the technology now that tries to not only identify that you've pulled up to the right building so your normal GPS, but also tries to identify where the door is, because I know in some buildings, I mean, yes, you might pull up in front of the building, but the door might be up one end or the other end of the building. So that's the type of also AI system. So the AI system would say, well, hang on a minute. Yeah. You're in front of the building. I can't see any doors at the moment. Would you like to, you know, go down to the left or would you like to go into the right? Or it might even just say when it's detected a door. That type of stuff is, is also coming. And I think that's, you know, make life a lot. Because that's always been the word. It's the last couple of percent. Or when you're traveling, you get to your location, but then you've still got to get inside.

16:56S3

Maybe they can install those in Rideshares. How many times have been dropped somewhere and they've either dropped you on the wrong side of the road or in front of the completely the wrong building, or, you know, just in an alleyway there. And this, I can I can see this, where this would be very, very handy, but... anything else that we can sort of, uh, see that would be making things a lot different in the future.

17:17S4

Yeah. Look, I just think the you sort of, you sort of not restricted by your imagination. There's a new system that's been out for a little while called Type Ahead for the Mac, and it's also coming out for windows, and that's where it's basically tries to almost do like a macro for you. So if you, for example, always start up your computer, you go into mail, you always check your mail, or you might always go to the third first message and start reading it there. The type of AI systems that can. That's why it's called type ahead, that you can actually get it to do things for you without you having to do very much. o, you know, I think about the old macro days where you could say, you know, press escape, press the tab key three times, hit the enter key, do this and do that. Well, this is a more sophisticated version of the old the old macro world which people still use. So that's really good. 

And also to when you think about AI when it comes to voice control over your computer. So normal voice control is just, you know, going going up and open an application, open up a document, go to the next word, go to the previous word line, sentence paragraph, blah blah blah. But wouldn't it be good if you could also then say to the AI system that was doing the voice control, can you summarize this webpage for me? Can you tell me what the main points are for the news today, or can you focus in and tell me what the humidity is going to be tomorrow in this weather application? They're the type of things that I'd really want to do, because at the end of the day, I'd love to use my computer completely hands free. You know, I can sort of do that at the moment, but not quite. 

And the other stuff is all to do with, um, health telemetry type stuff. So your heartbeat, your blood pressure, um, how many calories you're burning, what your exercise is, um, how much you've moved, all that sort of stuff. Now, if that then can be done by an AI system and it might even pop up in the morning and say, hey, David, I noticed yesterday that you only did, um, 30 hours or 30 hours, I must say 30 minutes of karate training. You've been slacking off a little bit. Would you like me to suggest some points for you where you can actually concentrate a bit, a bit more today? So there would be the really, really cool things. And the other one for me is and this comes back to searching. So I is very good on topical information or finding out stuff. But when it comes just to straight searching, as in I want to find a web page that I can look at. 

Yes, the AI system will actually share with you all its sources that it's used, but I just sometimes like the good old Search where you can where you can work in a search and we'll actually share with you topical information. But sometimes I want a little bit more nitty gritty and have a little bit more control over the sources that I'm getting access to. And probably the final one is I spend a lot of time every day on the internet looking through a lot of different stuff, social media, RSS feeds, emails. And that basically takes me an hour of basically manual labor. It'd be really nice if I could say to the system, can you just go through all the sources that I normally go through every single day and give me the highlight points, and that way I can sit back and just look at the highlight points rather than having to go through, say, roughly 800 to 1000 bits of information every single day. So I have an AI system basically go through stuff and basically data crunch will be really handy.

20:48S3

This stuff sounds miraculous, but I mean, how does it all work? I mean, what is the programming in? It must be tremendous. I mean, I'm thinking you're not just telling it what to take in, but also it must be programmed to ignore a lot of stuff as well, I'm guessing.

21:05S4

That's right. Yeah. So it's it's all the algorithms running in the background that gives it its power and all the sources that it's got access to. I can remember reading a science fiction thing that just said, just because something's got access to a huge amount of information and can correctly answer something does not mean it's actually intelligent. It's just got access to lots of data. And, you know, and what scientists are scary about is when we go from that huge global knowledge base that can also answer everything in the world, apparently, to where it becomes extremely intuitive, like a human being. That's what people are scared about. And that's what people talk about. Super AI. It's when it becomes, so intuitive. That you really do feel like you're talking to a human being. Keep in mind that even the most powerful computers at the moment have nowhere near the processing power of the human brain even yet. 

So I think we're still at a way to go before we get to the super AI stage. But just the amount of algorithmic input and output it actually gives you is incredible. So I just think it's just watch this space and see what happens. And hopefully this year, if I had a wish of anything, I'd wish that, you know, all our personal assistants. So our series, our Alexas and our our okay Googles and God forbid even Bixby on my Samsung phone. They start going from the toddler stage to the the primary school stage and start interacting with us in a much more, you know, proper manner, not just being a dumb command system.

22:39S3

And that would border on the scary, I guess. And all of a sudden it's asking you stuff.

22:44S4

Why it would be, wouldn't it? Yeah. Why do you want to turn the lights on for as far as I'm as far as I know, mate, you're blind. So why do you want to turn the lights on for? I think you'd be wasting power. I think you'd be far better off me leaving the lights off for you.

22:57S2

Well, why on earth would you want to bake a chocolate cake? Yeah. That's right.

23:00S4

You put on weight.

23:00S2

Yeah. What a silly idea.

23:02S4

That's it. Based on your based on your your weight information from your... why things, scars. Last week, you put on two kilos. I think it would be really bad if you had some chocolate. Okay, maybe consider a carrot.

23:15S2

I would love if we could get to that stage where I is like, no, you're in training. You can't be eating that food.

23:22S3

Yeah. That so? So they'll come a day where I'm coming in a blazing fury because I've had a massive argument with my toaster. 

23:30S2

Yes. That's it.

23:31S4

That's it. Yeah. Oh, they do you remember that classic, the classic Hitchhiker's Guide to the Galaxy joke, where they get into the lift and he goes, I want to go up or down? Yes. I think there's no up. Please. Have you considered all the possibilities of down that I can offer you? Yes. I just had an irate conversation with the lift, and I'm now normally not in a good mood to have this conference anymore, so yes. Yeah, I wouldn't, but it's not I don't really think it's going to be that far away, particularly when you think the amount of data that your phone or your computer or the cloud or anything else knows about you. I mean, it knows where you shop. It knows basically what you eat. If you keep your groceries stuff online, knows how much movement or exercise you do. It knows what places you tend to frequent. Most of the time, it knows who you communicate with. It knows how what sort of language you use when you're emailing people. It knows what your web searches are.

24:29S2

It just goes....

24:29S4

On and on.

24:30S2

Yeah, yeah.

24:31S4

And when you put it into a very clever system, yeah, it's almost like, well, you don't need me anymore. I'll just pop off into the virtual system and see you later.

24:41S5

One of the virtual David Woodbridge.

24:43S3

One of the things that I did find amusing, and when we started worrying about people cheating on their assignments for university or high school or something like that, and then, of course, it was revealed that ChatGPT also has an educator's function and you can tap into it, uh, say, did you write this article? And it will tell you if it did or not. So you can't even cheat using the thing there because yes, ChatGPT will say, yes, I wrote it. This is the person that asked me to write it for me.

25:11S2

Yeah. That's amazing.

25:14S4

What's really weird though, is I mean, I if I send somebody an email about, you know, web accessibility. So I just put on things like, you know, one, have you considered, um, alt tags on images? Uh, to don't use click here. Three blah blah blah. So if you put that into a chat GPT system, it's nothing like your own language. So it says, Dear Sir Madam, I would like to bring to your attention the following points based on the WC three consortium guidelines version 2.1 as published on bla bla bla date. And then it goes on and it's like the person that's going to get this email is going to know that this is not me writing the email, because there's no way. I mean, I hate verbosity, so I probably do something that ChatGPT do would do in about, I don't know, 1015 lines versus it's probably going to be a couple of pages or even a frigging encyclopedia. 

So I think it's sometimes you really know whether something's been developed by an AI because like, it's just so verbose. It's like, yeah, I just think people went over a little bit. So but it's actually very good. If you're very grumpy and you want to write a really grumpy email to somebody, it's really good on turning it down. So no check for that point of view. It comes in really handy. But yeah. Look, I'm not scared of AI. Um, I think it's got extremely good potential super AI things right away, but, I mean, who knows, it might be already here and we're just not being told about it. Like, you know, area 51 in the US, in the US. But no, I think it's great. I think it has huge implications. 

One of the things that just popped in my brain when I was just talking in is. I'd love to double check some example. My wife's in a wheelchair, so, you know, I'd love to be able to say, look, can you just look up Google Maps for me, please? I and just double check that the restaurant at the beach has a ramp onto the actual restaurant, rather than a set of stairs, or you have to go through the Saint. Now that's the sort of stuff it should be able to look up Google Maps, check the images on Google Maps and say, yeah, hey, look, you probably don't want to go to that one, but I just saw one down the road a bit further that does have full access. That's the type of stuff I think has amazing implications as well.

27:20S3

Well, I think we've got a lot to look forward to and maybe a little bit to be worried about in general. But it has been a fantastic talking to you once again. And like the classic says, I'm sure he'll be back.

27:34S5

Very good.

27:35S3

Hold on. And that's a wrap for the week. Next week we're talking in hushed voices and trying to stay on our best behavior.

27:47S2

Yes, we are visiting the Vision Australia Library. Library Services manager Sarah Blunden is our guest.

27:55S3

But between now and then, please do get in touch with the show. Whether you have experience of any of the issues covered in this episode of Studio One, or if you think there is something we should be talking about, you never know. Your story and insight may help somebody else who is dealing with something similar.

28:09S2

You can contact us at studio1@visionaustralia.org. That's studio - number one - at Vision Australia - dot - org.

28:17S3

Or maybe drop us a note on the Facebook page. We do want to hear from you.

28:21S1

Vision Australia Radio gratefully acknowledges the support of the Community Broadcasting Foundation for Studio 1.