WEBVTT - Talking Tech Interview Special: Laureate Fellow and Scientia Professor of Artificial Intelligence, Toby Walsh 0:00:21.798 --> 0:00:26.388 Hi folks, and welcome to a Vision Australia Radio special podcast. 0:00:26.388 --> 0:00:31.008 And today I'm joined by Professor Toby Walsh. And I've 0:00:31.008 --> 0:00:32.658 already forgotten what I'm going to call you, so I'll 0:00:32.658 --> 0:00:37.068 just call you, um, chief science person from the Eye 0:00:37.068 --> 0:00:39.678 Institute of University of New South Wales, which I probably 0:00:39.678 --> 0:00:43.308 just strangled it. Um, so, Toby, thanks for coming on 0:00:43.308 --> 0:00:46.248 the program today. And like I did, have seven questions 0:00:46.248 --> 0:00:49.308 that I'll get to very, very quickly. But my first one, 0:00:49.578 --> 0:00:52.488 and it makes me want to grind my teeth a lot, 0:00:52.488 --> 0:00:57.258 is why is there a need in even with AI 0:00:57.258 --> 0:01:00.318 itself and the people behind it, to actually try and 0:01:00.318 --> 0:01:05.808 make AI systems sound and react like human beings? That's a. 0:01:05.808 --> 0:01:11.568 Fantastic question. It's it. And it's a it's a fatal 0:01:11.568 --> 0:01:15.078 mistake that we're making all the time because it confuses 0:01:15.078 --> 0:01:19.128 people and it messes with people. Um, we give them, 0:01:19.128 --> 0:01:23.358 we give them names, often female names. Um, um, you know, 0:01:23.358 --> 0:01:26.988 we shouldn't call them Siri or Cortana. We should call 0:01:26.988 --> 0:01:33.738 them robot or computer or or or or, you know, 0:01:33.738 --> 0:01:36.678 names that certainly aren't gendered. You know, you could call 0:01:36.678 --> 0:01:40.518 them Alpha or Omega or, you know, Greek letters of 0:01:40.518 --> 0:01:45.318 the Greek alphabet, whatever it is. Yeah. Um, but we 0:01:45.318 --> 0:01:49.398 it's a it's a bad human trait. We anthropomorphize, we, 0:01:49.398 --> 0:01:56.058 we apply human attributes to things that are not human. Um, no. 0:01:56.418 --> 0:02:00.738 Exactly. And because in the in the demo from OpenAI, um, 0:02:00.738 --> 0:02:03.948 last week, as we record this week, they had it 0:02:03.948 --> 0:02:06.708 doing things like I think it was supposedly flirting with 0:02:06.708 --> 0:02:10.968 the person. Um, it was laughing at its own jokes. 0:02:11.328 --> 0:02:15.528 And I kept thinking, you're a computer program. This is 0:02:15.528 --> 0:02:19.998 all false and not real. And that's why I keep 0:02:19.998 --> 0:02:22.068 thinking of, um, I read your book, which is the, 0:02:22.098 --> 0:02:26.748 you know, the Faking It, um, excellent book about faking intelligence. 0:02:26.748 --> 0:02:30.258 And it this just reminds me that this is just 0:02:30.258 --> 0:02:33.618 so up there amongst faking it. I just don't know 0:02:33.618 --> 0:02:36.918 why people don't shrink away from this stuff and go, look, guys, 0:02:36.918 --> 0:02:39.258 I think we really are getting a bit carried away. 0:02:39.948 --> 0:02:42.018 This. The story is actually even worse than that. So 0:02:42.018 --> 0:02:45.768 actually today they've actually had to retard the female voice 0:02:45.768 --> 0:02:51.528 for the new OpenAI chat bot because, um, they tried 0:02:51.528 --> 0:02:54.708 to buy the rights for Scarlett Johansson. She was the 0:02:54.708 --> 0:02:58.728 voice on the movie her where it was her voice. 0:02:58.728 --> 0:03:01.218 She was nominated for an Oscar, indeed, for the part 0:03:01.458 --> 0:03:05.358 where she was the AI operating system. She declined to 0:03:05.358 --> 0:03:09.018 license her voice to OpenAI. But the voice that they 0:03:09.018 --> 0:03:12.828 had was eerily like her voice. Lots of people said 0:03:12.828 --> 0:03:16.758 it was reminded them of the movie, and many of 0:03:16.758 --> 0:03:20.388 Scarlett Johansson's friends apparently said they thought it was her. 0:03:20.388 --> 0:03:24.828 Right now. Um, because I think she's probably threatened to sue. 0:03:24.858 --> 0:03:30.528 She's probably taken a lawyer's less legal, um, action and 0:03:30.528 --> 0:03:34.338 the bad publicity. They've actually had to take the voice down. Yeah. 0:03:34.848 --> 0:03:37.308 Because, I mean, as a blind person, I've had talking 0:03:37.308 --> 0:03:41.508 computers since 1980. So, you know, in one form or another, 0:03:41.838 --> 0:03:45.048 I've had computers speak to me. So is it this 0:03:45.378 --> 0:03:47.988 thing that I don't know for a, for sighted people, 0:03:47.988 --> 0:03:52.308 it's really novel to have a computer quote speak to you. 0:03:52.308 --> 0:03:55.578 Or is it the fact that it's supposedly this real 0:03:55.578 --> 0:03:58.788 thing behind your computer screen that's really talking to you? 0:04:00.978 --> 0:04:03.858 It is a deceit. I mean, it is fooling us 0:04:03.858 --> 0:04:07.698 that maybe there's more behind the scenes than there actually is. 0:04:07.908 --> 0:04:10.218 As an example, another example, if you use one of 0:04:10.218 --> 0:04:14.448 these chatbots like ChatGPT, it slowly types out the answer 0:04:14.448 --> 0:04:18.378 to your question. If you actually understand how it works, 0:04:18.378 --> 0:04:21.288 it has the answer in a flash. It doesn't actually 0:04:21.288 --> 0:04:24.168 have to slowly type out the answer, but they've they've 0:04:24.168 --> 0:04:26.208 actually built the interface. So it does that. So you 0:04:26.208 --> 0:04:29.028 get the feeling that there's a person behind the screen 0:04:29.028 --> 0:04:32.628 that's slowly typing away to you to make it more personalized, 0:04:32.628 --> 0:04:35.298 to try and engage you more, to to fool you 0:04:35.298 --> 0:04:39.018 more that it's smarter, more intelligent, maybe even more sentient 0:04:39.018 --> 0:04:39.918 than it is. 0:04:39.918 --> 0:04:42.438 That's right. And a lot of the the nicer the voice, 0:04:42.438 --> 0:04:45.288 the more I tend to actually, um, dislike it more 0:04:45.288 --> 0:04:48.558 because as far as I'm concerned, you're a computer. Don't 0:04:48.558 --> 0:04:50.988 care what you sound like. Give me the information and 0:04:50.988 --> 0:04:54.048 make it accurate. And the one of the issues that 0:04:54.048 --> 0:04:57.468 I've got, particularly with generative AI, is that if it's 0:04:57.468 --> 0:05:04.818 just doing straight, you know, word, sentence, paragraph, page prediction, um, 0:05:04.818 --> 0:05:08.148 how can it actually get things so badly incorrect sometimes? 0:05:08.148 --> 0:05:11.538 I mean, yes, it's almost like a default. Sometimes. Most 0:05:11.538 --> 0:05:14.928 of the time it'll get it correct almost accidentally, I 0:05:14.928 --> 0:05:17.838 think sometimes based on its data. But when it gets 0:05:17.838 --> 0:05:20.388 things wrong, it really gets things wrong. 0:05:21.138 --> 0:05:24.078 Well, because it's, it's that's the thing that we should 0:05:24.078 --> 0:05:28.428 remember that it's artificial intelligence. It's not human intelligence. It's 0:05:28.428 --> 0:05:30.408 not going to break like human touch. I mean, the 0:05:30.408 --> 0:05:35.538 remarkable thing about human intelligence is, is how robust we are. 0:05:35.628 --> 0:05:39.828 Whereas computers, people who program computers quickly realize computers are 0:05:39.828 --> 0:05:43.518 very brittle. They break and they break very catastrophically. They 0:05:43.518 --> 0:05:46.938 don't have all of our common sense. They don't have 0:05:46.938 --> 0:05:49.848 our remarkability I can take you, I can drop you 0:05:49.848 --> 0:05:52.608 into a new circumstance and you could start doing stuff. 0:05:52.998 --> 0:05:56.298 Whereas a computer, you change the input very slightly and 0:05:56.298 --> 0:06:01.308 it falls over completely. Um, because it's an artificial, very 0:06:01.308 --> 0:06:05.298 different intelligence to human intelligence. And we we're quick, especially 0:06:05.298 --> 0:06:06.978 when it sounds like us to think it's going to 0:06:06.978 --> 0:06:09.708 be like us, but not realizing. No, actually it's quite 0:06:09.708 --> 0:06:13.188 a different type, quite a different flavor of intelligence. I 0:06:13.188 --> 0:06:13.668 can remember. 0:06:13.668 --> 0:06:16.518 This. I can't remember what Star Trek episode it was, 0:06:16.518 --> 0:06:18.708 but I think it was one of the original episodes 0:06:18.708 --> 0:06:21.378 with Spock involved because he had to sort of like, 0:06:21.378 --> 0:06:24.708 you know, he had to do these, um, his, you know, 0:06:24.708 --> 0:06:27.768 mind meld. But he did it with a computer, and 0:06:27.768 --> 0:06:30.468 I can remember this computer. It was some sort of 0:06:30.468 --> 0:06:33.888 bastardized name of Voyager, because NASA sent out the Voyager 0:06:33.888 --> 0:06:37.728 probe and this thing came back, which was this alien spacecraft. 0:06:37.728 --> 0:06:40.728 And it was interesting because what they were basically saying 0:06:40.728 --> 0:06:42.678 in the episode is that you can have all the 0:06:42.678 --> 0:06:45.648 knowledge in the world. It actually doesn't mean you're intelligent. 0:06:45.648 --> 0:06:49.668 And I just think it's such a misused word because 0:06:49.668 --> 0:06:51.978 I know it's artificial, but I think it's a bit 0:06:51.978 --> 0:06:55.008 bold to actually use the word intelligence because it's actually not. 0:06:55.008 --> 0:06:58.578 You actually just all it's doing is just looking up facts. 0:06:58.578 --> 0:07:02.118 It's not really extrapolating anything. And I just, I just 0:07:02.118 --> 0:07:04.458 want to actually have a new name for artificial intelligence 0:07:04.458 --> 0:07:05.748 because it's not intelligent. 0:07:05.958 --> 0:07:08.178 Well, it nearly got called many other things. It nearly 0:07:08.178 --> 0:07:13.608 got called cybernetics in Europe. They call it informatics. Um, yeah. 0:07:13.608 --> 0:07:16.218 It's at the end of the day, it's a terrible name. 0:07:16.398 --> 0:07:18.858 Intelligence is very poorly defined. So what the heck could 0:07:18.858 --> 0:07:23.058 artificial intelligence be? But Spock is a good example because, 0:07:23.058 --> 0:07:26.118 you know, he was quite a smart guy, but he 0:07:26.118 --> 0:07:30.498 was somewhat lacking in emotional intelligence. And you realize there 0:07:30.498 --> 0:07:33.948 are many different facets to intelligence your social intelligence, our 0:07:33.948 --> 0:07:38.268 emotional intelligence, our creativity that are very important, really important 0:07:38.268 --> 0:07:42.438 to to our interactions with each other. And Spock, despite 0:07:42.438 --> 0:07:46.038 he was clearly a very smart cookie. Mm. Was also 0:07:46.038 --> 0:07:49.038 quite lacking in his interpersonal skills. Yes. 0:07:49.038 --> 0:07:51.198 Yeah. No it was it used to be quite funny. 0:07:51.198 --> 0:07:51.828 It was great. 0:07:51.948 --> 0:07:54.918 And the same would be the same is true of machines. 0:07:54.918 --> 0:07:58.968 They don't have. Um, no. Some cognitive intelligence. They have 0:07:58.968 --> 0:08:01.848 some ability to, you know, answer general knowledge questions and 0:08:01.848 --> 0:08:04.788 do maths and those things. But they are very severely 0:08:04.788 --> 0:08:08.508 lacking in emotional intelligence. And that's likely to be a 0:08:08.508 --> 0:08:12.948 significant handicap, that they have a long time going forwards, 0:08:13.188 --> 0:08:16.278 in part because they're not they don't have emotions themselves. 0:08:16.278 --> 0:08:19.428 I mean, obviously they don't have emotions. Emotions are biochemical. 0:08:19.428 --> 0:08:23.418 So they they're electrical devices. So they don't have anything 0:08:23.418 --> 0:08:26.208 like that. And, and one of the great things that 0:08:26.208 --> 0:08:28.548 we have is that, well, we can reflect on, on 0:08:28.548 --> 0:08:31.158 our own emotions. We can say, well, how would I 0:08:31.158 --> 0:08:34.788 feel if, um, you know, someone said that to me 0:08:34.788 --> 0:08:37.848 and then I could think, oh, I'd be really upset. Um, 0:08:37.848 --> 0:08:41.328 and so machines can't do that. They don't have any 0:08:41.328 --> 0:08:43.848 insight to reflect and think about how I would feel, 0:08:43.848 --> 0:08:46.908 because they have no feelings themselves. So they're going to 0:08:46.908 --> 0:08:51.618 be very severely handicapped, certainly that facet of their intelligence. Um, 0:08:51.618 --> 0:08:52.818 for a long time. Absolutely. 0:08:53.208 --> 0:08:56.718 I, I can remember the first, um, thing I played 0:08:56.718 --> 0:08:58.728 on my little Apple TV of speech back in the 0:08:58.728 --> 0:09:05.568 early 1980s. Was that horrible, ridiculous psychotherapy program. Eliza. Um. 0:09:05.748 --> 0:09:08.508 And I used to try and trick it. I'd say, oh, look, my, 0:09:08.508 --> 0:09:11.958 my my bed is feeling very sad today. And it's like, oh, 0:09:11.958 --> 0:09:14.898 when was the last time your your bed felt sad? 0:09:15.678 --> 0:09:18.348 And it just got more and more ridiculous. And I thought, yeah, 0:09:18.348 --> 0:09:20.508 all you're really doing is just word prediction. You're just 0:09:20.508 --> 0:09:24.618 doing straight reflection. Um, so we're we're smart speakers these days, 0:09:24.618 --> 0:09:28.548 are they? Because I'm assuming, you know, our traditional smart 0:09:28.548 --> 0:09:31.578 speakers lie. So the, the upper ones, the Amazon ones 0:09:31.578 --> 0:09:33.918 and the Google ones, I'm assuming at the moment they're 0:09:33.918 --> 0:09:37.458 still not using large language models. I'm still they've sort 0:09:37.458 --> 0:09:39.378 of got a I'm assuming they've got like some sort 0:09:39.378 --> 0:09:42.738 of set series of instructions and that's all they're currently 0:09:42.738 --> 0:09:43.848 working on at the moment. 0:09:44.388 --> 0:09:49.068 Yes. Something like Alexa. Very it's very heavily programmed, very 0:09:49.068 --> 0:09:53.568 heavily scripted. Mm. Um, but the future is going to 0:09:53.568 --> 0:09:56.988 be that ultimately AI is going to be the operating 0:09:56.988 --> 0:10:00.588 system of those devices. And indeed all our devices, your 0:10:00.588 --> 0:10:05.448 your smartwatch, your smart phone, your smart home, your smart toaster, 0:10:05.448 --> 0:10:08.298 your smart front door, your smart light switch, they're all 0:10:08.298 --> 0:10:11.808 going to have, uh, it it is actually when we 0:10:11.808 --> 0:10:13.428 come back to Scarlett Johansson, is is going to be 0:10:13.428 --> 0:10:16.608 like that movie. Her AI is going to be the 0:10:16.608 --> 0:10:20.178 operating system of all of those devices that's going to 0:10:20.178 --> 0:10:22.308 be allowed you to interact with them so you can 0:10:22.308 --> 0:10:25.248 talk to them. They understand at a much higher level 0:10:25.248 --> 0:10:27.558 what you want them to do, and then do that 0:10:27.558 --> 0:10:30.558 stuff for you. So it is, um, you know, we're 0:10:30.558 --> 0:10:33.888 only at the beginnings of that journey where, um, devices 0:10:33.888 --> 0:10:37.638 get upgraded with more and more AI that allows us 0:10:37.638 --> 0:10:44.368 to have a richer and richer conversation with. Um, so that, 0:10:44.368 --> 0:10:47.668 you know, ultimately, you know, people will, I think, become 0:10:47.668 --> 0:10:50.848 somewhat attached to these devices that they're always talking to. 0:10:50.878 --> 0:10:54.298 No. Yeah. The one thing that always gets me very 0:10:54.298 --> 0:10:58.498 nervous about, um, anything to do with AI is when 0:10:58.498 --> 0:11:01.918 it comes to computer vision. Um, because I know in 0:11:01.918 --> 0:11:04.438 your book you were talking about, you know, um, you know, 0:11:04.438 --> 0:11:08.698 there's lots of data being trained on, on, on medical stuff, on, uh, 0:11:08.698 --> 0:11:11.128 doing stuff for radiology and all that sort of stuff. 0:11:11.398 --> 0:11:15.088 The problem I have as a blind person is, you know, 0:11:15.088 --> 0:11:17.998 whenever I use my smartphone to take a picture of 0:11:17.998 --> 0:11:20.728 something in the backyard or out the front and so on, 0:11:21.148 --> 0:11:25.228 there's still about a 75 or maybe even 80% chance 0:11:25.228 --> 0:11:27.988 of if I was being nice, that what the camera's 0:11:27.988 --> 0:11:32.278 actually telling me what the object is is actually incorrect. 0:11:32.278 --> 0:11:35.068 And I just thought, is that just the state of 0:11:35.068 --> 0:11:38.878 computer vision? Or is the fact that, you know, it 0:11:38.878 --> 0:11:42.928 doesn't have enough data to base the fact that you know, 0:11:42.928 --> 0:11:46.948 my red garbage bin is not a fire hydrant on 0:11:46.948 --> 0:11:48.808 my grass verge, for example? 0:11:48.808 --> 0:11:53.638 Yeah, it's it's getting better and better as we train 0:11:53.638 --> 0:11:56.578 them on more and more data. Um, you know, the 0:11:56.578 --> 0:11:58.858 idea ultimately of computer vision is to be able to 0:11:58.858 --> 0:12:02.098 understand the world so that computers can robots, for example, 0:12:02.098 --> 0:12:05.488 or autonomous cars can navigate around it. Mm. Uh, and 0:12:05.488 --> 0:12:09.148 I'm pretty confident we will get there, um, in part because, 0:12:09.148 --> 0:12:13.108 for example, um, we're driving increasing. You know, if you're 0:12:13.108 --> 0:12:16.138 driving a Tesla, you're helping to train the next generation 0:12:16.138 --> 0:12:19.048 of Tesla. Yeah. Um, and what and the other thing 0:12:19.048 --> 0:12:22.648 that's really making a big difference is, interestingly enough, is 0:12:22.648 --> 0:12:26.068 the is that, in fact, actually Tesla's do more of 0:12:26.068 --> 0:12:30.088 their computer vision training on simulators than they do in 0:12:30.088 --> 0:12:34.138 the real world that the simulators now, the car driving 0:12:34.138 --> 0:12:39.118 simulators are so good, so realistic. Mm. Um, that they drive, 0:12:39.118 --> 0:12:41.968 you know, more than ten times the distance every night 0:12:41.968 --> 0:12:43.948 in simulators. And of course, if you're doing it in 0:12:43.948 --> 0:12:46.138 a simulator, you can do it faster than real time. 0:12:46.138 --> 0:12:48.898 You can you can speed up the world ten times 0:12:48.898 --> 0:12:53.578 or 100 times, um, and train the computer systems on those. 0:12:53.578 --> 0:12:55.378 And then the other great thing about simulators, of course, 0:12:55.378 --> 0:12:57.598 is you can you can do things that wouldn't, you 0:12:57.598 --> 0:12:59.338 wouldn't be allowed to do in the real world. You know, 0:12:59.338 --> 0:13:02.518 you can have you can practice accidents. You can make 0:13:02.518 --> 0:13:06.658 the conditions really difficult. Yeah. Um, you know, you can 0:13:06.658 --> 0:13:08.878 put the sun into the eye of the driver and 0:13:08.878 --> 0:13:11.038 you make the road wet. You can do things that 0:13:11.038 --> 0:13:13.198 you know, where you might be risking the life of 0:13:13.198 --> 0:13:16.678 the driver and and cause an accident, and then no 0:13:16.678 --> 0:13:19.018 one dies, which is great. And then, of course, what 0:13:19.018 --> 0:13:23.008 you can do in a simulator is that you can say, well, okay, um, the, 0:13:23.008 --> 0:13:25.168 the computer failed to do it that right that time, 0:13:25.168 --> 0:13:27.448 but let's train the algorithm again. Let's try and see 0:13:27.448 --> 0:13:29.698 if we can correct it and then run it again 0:13:29.698 --> 0:13:35.338 in exactly the same circumstances. You. Mm. Replication of the experiment, 0:13:35.338 --> 0:13:36.628 which you can't of course, do in the real world. 0:13:36.628 --> 0:13:39.868 You can never replicate exactly those conditions again. No, you 0:13:39.868 --> 0:13:41.968 can say let's repeat it until the computer gets it 0:13:41.968 --> 0:13:45.028 right and doesn't kill the driver. No. That's true. And 0:13:45.028 --> 0:13:48.628 so so being able to train computers and simulators is 0:13:48.628 --> 0:13:52.228 actually going to really help us move forwards in leaps 0:13:52.228 --> 0:13:52.558 and bounds. 0:13:52.768 --> 0:13:54.718 So is there any I mean, there are there any 0:13:54.718 --> 0:13:58.708 true level five self-driving cars on the road, or are 0:13:58.708 --> 0:14:01.138 there sort of or or back down to the level? 0:14:01.138 --> 0:14:03.118 I think it's a level two isn't it? They're not 0:14:03.118 --> 0:14:05.188 really self autonomous vehicles at the moment. 0:14:05.368 --> 0:14:08.728 2 to 3. Yes. Um, yeah. I mean, the bad 0:14:08.728 --> 0:14:12.148 news there is that unfortunately sighted people are going to get, um, 0:14:12.148 --> 0:14:17.878 autonomous driving before people with, um, visual impairment because it's 0:14:17.878 --> 0:14:21.508 slowly being put into our cars without us realizing. I 0:14:21.508 --> 0:14:23.878 was driving my car the other day, and I realized 0:14:23.878 --> 0:14:27.298 I'm driving less. I'm not looking over my shoulder as 0:14:27.298 --> 0:14:30.148 often as I. I suspect it's going to be our 0:14:30.148 --> 0:14:33.658 children who don't drive cars. They will, but they won't 0:14:33.658 --> 0:14:36.808 get around to get their licenses. And in our case, 0:14:36.808 --> 0:14:39.928 for people who already can drive. Mm. It's going to 0:14:39.928 --> 0:14:43.708 slowly happen to us. Mhm. So um, I imagine at 0:14:43.708 --> 0:14:45.238 some point in the future I'm going to go to 0:14:45.238 --> 0:14:48.658 the RTA to renew my driving license, and they're going 0:14:48.658 --> 0:14:50.878 to say to me, well, Mr. Walsh, we, um, we 0:14:50.878 --> 0:14:53.278 checked your computer records and it seems that you haven't 0:14:53.278 --> 0:14:57.028 been doing much driving recently, but it's actually been the car, 0:14:57.028 --> 0:14:59.758 been doing all the driving, all the assists, all the 0:14:59.758 --> 0:15:02.998 autonomous assistance in the car. And you really don't have 0:15:02.998 --> 0:15:06.268 the hours under your belt. So you have two choices now, Mr. Walsh. 0:15:06.268 --> 0:15:08.878 You can either you can either take your test again 0:15:08.878 --> 0:15:10.888 to prove that you can still actually drive the car. 0:15:10.888 --> 0:15:14.038 It's not just the all the intelligent assistants, the the 0:15:14.038 --> 0:15:16.708 lane following and the automatic braking and all of that 0:15:16.708 --> 0:15:19.978 and the automatic steering, the parking. Or we can give 0:15:19.978 --> 0:15:24.388 you this new, um, non-driving driving license that, that and 0:15:24.388 --> 0:15:26.368 I think I expect well, I it was too painful 0:15:26.368 --> 0:15:28.078 getting my license in the first place. I think I'll 0:15:28.078 --> 0:15:30.538 get the non-driving license. And they say, oh, by the way, Mr. Walsh, 0:15:30.538 --> 0:15:32.248 you'll get a discount on your insurance now. 0:15:32.338 --> 0:15:34.708 Yeah. That's right. Yes. 0:15:35.188 --> 0:15:37.348 Remove the dangerous human from the loop. 0:15:37.348 --> 0:15:40.108 That's right. My my wife often says it's it's actually 0:15:40.108 --> 0:15:42.088 the other drivers that, um, are the. 0:15:42.838 --> 0:15:44.218 And that is part. 0:15:44.218 --> 0:15:46.468 Of the problem. Yes, dealing with the uncertainty of other drivers. 0:15:46.558 --> 0:15:49.078 Once vehicles can talk to each other, then there would 0:15:49.108 --> 0:15:51.808 be much less uncertainty. But I think it's slowly creeping 0:15:51.808 --> 0:15:54.508 up on us. And then there will be it will 0:15:54.508 --> 0:15:57.118 happen in special places. So there'll be the, you know, 0:15:57.118 --> 0:16:00.568 inner city congestion charging zone where you're only allowed to 0:16:00.568 --> 0:16:03.088 go in if you're in an electric car that's autonomous 0:16:03.088 --> 0:16:06.208 or the high speed of the highway, you'll only be 0:16:06.208 --> 0:16:09.838 able to enter the high speed of the highway where 0:16:09.838 --> 0:16:12.118 you can platoon the cars together if you've got the 0:16:12.118 --> 0:16:13.378 autonomous aids. 0:16:13.618 --> 0:16:17.788 So the only question I've got to with autonomous cars 0:16:17.788 --> 0:16:21.578 is it's fine to take you somewhere. But there seems 0:16:21.578 --> 0:16:24.608 to be no mention at the moment about how I 0:16:24.638 --> 0:16:27.308 could then say, okay, so the car's parked you in 0:16:27.308 --> 0:16:32.018 the car park at Woolworths. I'm now going to actually 0:16:32.018 --> 0:16:35.408 let you I'm going to guide you to where you 0:16:35.408 --> 0:16:40.118 can start entering into the actual shop itself inside the complex. 0:16:40.478 --> 0:16:43.208 Is that the sort of stuff that I could actually 0:16:43.208 --> 0:16:45.338 do for people that are blind? So your GPS gets 0:16:45.338 --> 0:16:48.278 you to the Or the car gets you to the location, 0:16:48.278 --> 0:16:50.468 but then, then the AI has then got to get 0:16:50.468 --> 0:16:52.778 you out of the car park into the building. 0:16:54.108 --> 0:16:57.708 Yeah, yeah, it's the last 100m or the or the 0:16:57.708 --> 0:17:01.218 last kilometer. That is a problem. And of course, GPS 0:17:01.218 --> 0:17:06.678 doesn't work inside. No. And GPS doesn't have the accuracy or, 0:17:06.678 --> 0:17:10.368 you know, really to do that, you know, very high precision. Um, 0:17:10.368 --> 0:17:14.358 last 100m, which is exactly where you want AI to 0:17:14.358 --> 0:17:17.088 step in, which is exactly why we work on things 0:17:17.088 --> 0:17:18.798 like computer vision. Because you have to see the world. 0:17:18.798 --> 0:17:20.418 You have to see where the door handle is. You 0:17:20.418 --> 0:17:23.328 have to see there's a, you know, another pedestrian coming 0:17:23.328 --> 0:17:27.918 out of the door and someone's left, um, a dog 0:17:27.918 --> 0:17:29.178 tied to the railings. And you. 0:17:29.568 --> 0:17:30.108 Yeah. 0:17:30.738 --> 0:17:33.798 Um, that's why you have to treat, teach computers how 0:17:33.798 --> 0:17:37.008 to see the world so you can navigate those last 100m. 0:17:37.128 --> 0:17:37.278 Mm. 0:17:37.308 --> 0:17:39.168 So is there any is there any sort of, like, 0:17:39.168 --> 0:17:40.818 3D modeling. 0:17:40.818 --> 0:17:41.898 Or. 0:17:41.898 --> 0:17:46.038 Sort of computer vision type stuff where, um, I don't know, 0:17:46.038 --> 0:17:48.648 a shopping center could say, well, look, here's a complete 0:17:48.648 --> 0:17:53.328 3D visual map of the whole complex, and then we'll, 0:17:53.328 --> 0:17:56.538 we'll actually use that data, imagery, that sort of stuff, 0:17:56.538 --> 0:18:00.888 then to guide the person around. So I can say, okay, so, um, 0:18:00.888 --> 0:18:03.948 the software or the AI knows that I'm in the building. 0:18:03.948 --> 0:18:08.808 I want to go to the storage section in Bulworth's 0:18:09.168 --> 0:18:12.408 so I, you can see where I am. Can you, 0:18:12.408 --> 0:18:14.868 can you guide me to that spot just based on 0:18:14.868 --> 0:18:17.118 the stuff that you write, the stuff that it's actually 0:18:17.118 --> 0:18:18.048 sees around me? 0:18:19.798 --> 0:18:20.038 Yeah. 0:18:21.568 --> 0:18:25.078 The real problem, um, with that is that the world 0:18:25.078 --> 0:18:28.588 keeps on changing. So they have high precision maps and 0:18:28.588 --> 0:18:33.358 high precision, uh, 3D models of the world. But the 0:18:33.358 --> 0:18:37.618 world has always changed. So one of the things that 0:18:37.618 --> 0:18:41.248 we try and do in, in AI is what's called Slam. 0:18:41.248 --> 0:18:45.838 It's simultaneous location localization and mapping, which is look at 0:18:45.838 --> 0:18:49.228 the world, understand where you are, and also map the 0:18:49.228 --> 0:18:52.528 world at the same time. So actually the computer vision 0:18:52.528 --> 0:18:56.188 is not only working out where you are with perspective to, 0:18:56.188 --> 0:18:58.558 you know, the front door and the door handle, but 0:18:58.558 --> 0:19:01.798 also working out what is in the world to see actually, oh, 0:19:01.798 --> 0:19:03.718 there's a dog over there now, which was obviously not 0:19:03.718 --> 0:19:06.508 in my model of the world because it's just stepped 0:19:06.508 --> 0:19:09.808 into frame. Mm. Um, so that's something that, you know, 0:19:09.808 --> 0:19:11.758 we put a lot of effort in. I trying to 0:19:11.758 --> 0:19:15.118 build systems that can actually situate themselves in the world, 0:19:15.118 --> 0:19:18.028 but also map the world as they find it, because 0:19:18.028 --> 0:19:20.608 the world keeps changing and you can never have, you know, 0:19:20.608 --> 0:19:22.498 up to date maps and models of the world. They're 0:19:22.498 --> 0:19:25.408 always out of date. You really got to go there 0:19:25.408 --> 0:19:29.038 and actually perceive the world and work out what state 0:19:29.038 --> 0:19:30.178 is the world now in. 0:19:30.328 --> 0:19:33.148 Yeah. And look, that that's my ultimate thing to do 0:19:33.148 --> 0:19:35.368 with I mean I can, I can sort of take 0:19:35.368 --> 0:19:38.248 or leave self-driving cars. But if I knew I had 0:19:38.248 --> 0:19:45.208 a computer vision system that could 100% independently navigate me around, um, 0:19:45.208 --> 0:19:51.418 into shopping centers or public infrastructure, um, transport hubs, the airport, 0:19:51.418 --> 0:19:54.688 that sort of stuff. That would be absolutely amazing. So 0:19:54.688 --> 0:19:55.708 for me, the. 0:19:55.708 --> 0:19:57.628 Good news is that's coming, right? Okay. 0:19:57.628 --> 0:19:58.378 So it's not. 0:19:58.738 --> 0:20:00.238 It's not pie in the sky. 0:20:00.238 --> 0:20:00.718 Stuff. 0:20:00.718 --> 0:20:03.268 It's not magic because we do it right. We do 0:20:03.268 --> 0:20:07.138 it with our two eyes. Stereoscopic vision. Yeah. We managed. 0:20:07.138 --> 0:20:11.398 Humans have managed to do that. Um, um, and we 0:20:11.398 --> 0:20:15.358 are slowly getting to the point where, increasingly with increasing accuracy, 0:20:15.388 --> 0:20:17.998 we can get computers to do the same thing. Mm. 0:20:18.328 --> 0:20:20.188 And the great thing about computers, of course, as well, 0:20:20.188 --> 0:20:23.548 is that they're not limited to the visual spectrum. So 0:20:23.548 --> 0:20:25.888 they can also see the world in microwaves and infrared. 0:20:25.888 --> 0:20:29.278 And um, so they can see. And so if it's 0:20:29.278 --> 0:20:32.758 low light or bad weather, then they can also still 0:20:32.758 --> 0:20:36.628 see the world when, um, uh, where vision alone may, 0:20:36.628 --> 0:20:39.868 may make it really challenging problem. So I have the 0:20:39.868 --> 0:20:44.668 possibility ultimately of actually helping us to see the world better. 0:20:44.758 --> 0:20:46.198 We can see the world because they can do it 0:20:46.198 --> 0:20:48.988 even if it was dark or even if it was 0:20:48.988 --> 0:20:49.948 pouring with rain. 0:20:50.338 --> 0:20:50.818 Okay. 0:20:51.148 --> 0:20:55.258 And just finally, has there been much research into how 0:20:55.258 --> 0:20:59.158 AI could impact people with a disability moving forward into 0:20:59.158 --> 0:20:59.878 the future? 0:21:00.908 --> 0:21:04.418 But, I mean, there has been work. Um, not enough work, actually, 0:21:04.418 --> 0:21:09.608 I'm sure. Um, and, um, because when I think about 0:21:09.818 --> 0:21:13.028 what are the technologies that are going to help people 0:21:13.358 --> 0:21:16.838 with limited hearing or with limited vision, they are exactly 0:21:16.838 --> 0:21:19.928 I the I is the technology that allows computers to 0:21:19.928 --> 0:21:22.598 hear the world and see the world, and then convey 0:21:22.598 --> 0:21:26.738 that information to those people who would otherwise be more 0:21:26.738 --> 0:21:30.128 isolated than they need be. So. So I, I have 0:21:30.128 --> 0:21:32.948 a lot of hope. The problem. The problem, of course, 0:21:32.948 --> 0:21:36.248 the fundamental problem is not a technical problem. It's a 0:21:36.248 --> 0:21:38.408 societal and a financial one, which is how do we 0:21:38.408 --> 0:21:42.218 ensure that there are incentives for the tech companies and 0:21:42.218 --> 0:21:44.888 business to do that? Because to begin with, it's going 0:21:44.888 --> 0:21:45.848 to cost them money. 0:21:45.998 --> 0:21:46.178 Mm. 0:21:46.418 --> 0:21:49.178 It's much easier for them to cater for the, you know, 0:21:49.178 --> 0:21:52.418 the vast majority of people who are, uh, you know, 0:21:52.418 --> 0:21:57.758 normally sighted and, um, normally hearing and not, um, you know, 0:21:57.758 --> 0:21:59.888 invest the time and effort and actually use this as 0:21:59.888 --> 0:22:02.828 a way of making, um, it more accessible for people 0:22:02.828 --> 0:22:04.088 with with disability. 0:22:04.268 --> 0:22:04.778 No. 0:22:04.928 --> 0:22:07.898 And I get this asked a lot about robotics and 0:22:08.018 --> 0:22:13.568 artificial intelligence, but there's some interesting stuff coming out for mobility, 0:22:13.598 --> 0:22:16.358 stuff to do for blind and low vision. And one 0:22:16.358 --> 0:22:19.928 of the major products is called glide from glide technology. 0:22:19.928 --> 0:22:22.658 And what it is, it's basically a well, people say 0:22:22.658 --> 0:22:24.458 it looks like a little vacuum cleaner. You've got little 0:22:24.698 --> 0:22:26.978 two little robot with a handle on it, and the 0:22:26.978 --> 0:22:30.878 blind person hangs onto the handle, and then it steers 0:22:30.878 --> 0:22:33.998 you around obstacles. And because it's connected to your smartphone, 0:22:33.998 --> 0:22:36.158 you've got an app that will tell you what's around you, 0:22:36.548 --> 0:22:40.028 as in businesses and that sort of stuff. But I'm 0:22:40.028 --> 0:22:46.298 just wondering how trustworthy is such a system, because I'm 0:22:46.298 --> 0:22:50.738 assuming it's using lidar and radar and infrared and all 0:22:50.738 --> 0:22:54.338 sorts of amazing stuff, but I just, I just think 0:22:54.338 --> 0:22:56.528 at the end of the day, it it it doesn't. 0:22:56.528 --> 0:22:58.028 And I come back to that original thing that I 0:22:58.028 --> 0:23:00.278 talked about. It doesn't have the level, a level of 0:23:00.278 --> 0:23:04.928 human intelligence or thinking outside of the box, if you'll 0:23:05.138 --> 0:23:07.418 pardon the pun, if the wheels fall off, for example. 0:23:10.018 --> 0:23:10.828 Yes. 0:23:10.918 --> 0:23:15.658 Yeah, it's obviously it doesn't replace what those wonderful guide 0:23:15.658 --> 0:23:20.638 dogs do. Mhm. Uh, because if you know, I've had 0:23:20.638 --> 0:23:24.448 the pleasure to meet some blood blind people with their 0:23:24.448 --> 0:23:28.678 dogs and they have a wonderful relationship with their dog. Mm. Um, 0:23:28.678 --> 0:23:31.468 and it is about that as much as anything that 0:23:31.468 --> 0:23:34.198 they can trust the dog to, to, to guide them across, 0:23:34.198 --> 0:23:36.448 you know, a busy road and putting their life in 0:23:36.448 --> 0:23:41.968 the hands of the dog, literally. Um, and, uh, that 0:23:41.968 --> 0:23:48.418 is a significant technical, um, milestone to me. Mm. But equally. 0:23:48.598 --> 0:23:52.558 Mm am also aware of how expensive it is to 0:23:52.558 --> 0:23:57.298 train guide dogs and how limited supply they are in. So, um, 0:23:57.298 --> 0:24:00.448 I do think, you know, there is a possibility there 0:24:00.448 --> 0:24:04.348 that we might be able to provide that mobility that perhaps, 0:24:04.348 --> 0:24:06.808 you know, some people can't because we don't have enough blind. 0:24:06.838 --> 0:24:09.178 You know, we don't have enough guide dogs to help 0:24:09.178 --> 0:24:11.248 people around. It would be wonderful if we had more 0:24:11.248 --> 0:24:14.308 and we could afford more. We trained more. But yeah, 0:24:14.398 --> 0:24:17.128 I'm not sure that's the world that we're in. And so. 0:24:17.338 --> 0:24:17.668 Um. 0:24:17.668 --> 0:24:20.548 If it is not perhaps as such a good solution. 0:24:20.548 --> 0:24:22.618 Yeah, we might look. 0:24:22.618 --> 0:24:24.568 And the other thing too, that I always think about, 0:24:24.568 --> 0:24:27.508 like I'm always thinking about the. Yes, but if then 0:24:27.508 --> 0:24:29.818 when type stuff all the time. Because to me the 0:24:29.818 --> 0:24:33.418 world's not black and white. It's got various shades of gray. And, 0:24:33.688 --> 0:24:37.288 you know, it's fine. A self-driving robot type thing might 0:24:37.288 --> 0:24:40.168 be fine for just trundling down the footpath and maybe 0:24:40.168 --> 0:24:42.988 going around the occasional car that's parked on the footpath, 0:24:42.988 --> 0:24:46.168 or it sees a branch and it goes, oh, hang 0:24:46.168 --> 0:24:51.298 on a minute. Um, I know my, my, my user is, um, 2.2m. 0:24:51.298 --> 0:24:54.448 That branch is two metres. I'll, I'll stop and go 0:24:54.448 --> 0:24:57.358 around it or do something like that. The problem is 0:24:57.358 --> 0:25:00.358 when you get into the highly unpredictable things. So I 0:25:00.358 --> 0:25:02.548 remember I was yelling at my guide dog one day 0:25:02.548 --> 0:25:04.378 because she wouldn't go in a straight line. She kept 0:25:04.378 --> 0:25:07.678 zigzagging and, um, I got to you and I went 0:25:07.858 --> 0:25:10.528 freaking bloody my dog. She was doing all sorts of 0:25:10.528 --> 0:25:14.068 nut stuff. And they went, yeah, somebody was actually repainting 0:25:14.068 --> 0:25:16.948 the manhole covers, and they were lifted off down that footpath. 0:25:16.948 --> 0:25:19.588 And I went, oh, what a good guide dog my 0:25:19.588 --> 0:25:20.368 guide dog was. 0:25:22.228 --> 0:25:23.758 So you're right. 0:25:23.758 --> 0:25:27.598 These, these unpredictable things, these these black swan, these corner cases, 0:25:27.598 --> 0:25:29.638 those are going to be the challenging ones. And which 0:25:29.638 --> 0:25:31.888 is why you're going to see these things turn up 0:25:31.888 --> 0:25:36.178 in more controlled environments. So so suppose you live you 0:25:36.178 --> 0:25:39.418 know on an estate, you know, an old person's estate 0:25:39.418 --> 0:25:42.868 where it's, you know, a gated community and it's, you know, 0:25:42.868 --> 0:25:45.898 it's a much more controlled environment. And then we're trundling 0:25:45.898 --> 0:25:50.308 around that environment. Um, you might be quite safe with 0:25:50.308 --> 0:25:53.398 a robot where, you know, as soon as you leave 0:25:53.398 --> 0:25:55.828 the gates of that community, you're out of the big, 0:25:55.828 --> 0:25:59.578 wild world where there are people who inconveniently leave manhole 0:25:59.578 --> 0:26:03.298 covers off. Might be better off. Um, you know, using 0:26:03.298 --> 0:26:06.538 the using the services of a of a of a 0:26:06.538 --> 0:26:09.688 guide dog. Um, so I can see it, you know, 0:26:09.688 --> 0:26:12.808 those more constrained settings being the ones where this sort 0:26:12.808 --> 0:26:14.818 of technology is first turn up, where you can be 0:26:14.818 --> 0:26:17.608 more sure that there are, you know, all of these 0:26:17.608 --> 0:26:20.518 black swan events that are waiting to manhole covers, waiting 0:26:20.548 --> 0:26:22.948 to trip you up or. Um. 0:26:23.398 --> 0:26:23.788 Yeah. 0:26:23.788 --> 0:26:27.388 And it just seems with the whole of, um, I 0:26:27.388 --> 0:26:30.958 there's no I remember that famous saying I can't remember 0:26:30.958 --> 0:26:34.348 it exactly, but something about who will guard those self-same guardians. 0:26:34.348 --> 0:26:37.378 And I'm just wondering, is there any sort of fallback 0:26:37.378 --> 0:26:41.368 system that says, no, the AI system is completely lost. 0:26:41.368 --> 0:26:44.518 The plot. It's wrong. Um, it's got the information wrong. 0:26:44.518 --> 0:26:48.478 It's got the orientation mobility wrong. It's got everything wrong. 0:26:48.478 --> 0:26:50.698 Now it's time to stop and just put a human 0:26:50.698 --> 0:26:51.748 being in charge. 0:26:52.558 --> 0:26:56.158 This Custodiet custodes. Who guards the guards? 0:26:56.158 --> 0:26:56.938 That's the one. 0:26:59.068 --> 0:27:02.038 My, my, my classical education finally got. 0:27:02.158 --> 0:27:02.758 Well done. 0:27:03.958 --> 0:27:06.418 All of that. Studying my Latin and Greek finally got 0:27:06.418 --> 0:27:07.468 used once. 0:27:08.608 --> 0:27:08.998 Uh. 0:27:09.388 --> 0:27:13.258 Yes. At the end of the day, um, we ultimately 0:27:13.258 --> 0:27:15.208 humans have to be in charge because only humans can 0:27:15.208 --> 0:27:19.948 be held accountable. So, um, um, we there are plentiful 0:27:19.948 --> 0:27:22.468 places where I think we're going to have to make 0:27:22.468 --> 0:27:26.818 sure that humans are left with, you know, overall responsibility. 0:27:28.178 --> 0:27:32.198 Yeah. No, look, I tend to agree because, um, I mean, 0:27:32.198 --> 0:27:34.208 every time I look at I and it mostly gets 0:27:34.208 --> 0:27:37.868 it wrong or somebody says to me, this Onam system 0:27:37.868 --> 0:27:40.028 is the best thing since sliced bread, and it still 0:27:40.028 --> 0:27:41.798 gets it wrong. Um. 0:27:42.158 --> 0:27:45.308 You you're right. It gets it's still making mistakes. It's 0:27:45.308 --> 0:27:48.848 still not perfect, but it's also it's easy to be 0:27:48.848 --> 0:27:52.328 forgetful of how it has advanced. I mean, I remember 0:27:52.328 --> 0:27:57.398 speech recognition systems 20 years ago. Mm. Incredibly painful. They 0:27:57.398 --> 0:28:01.868 had to be speaker trained. Mm. Um, that you you 0:28:01.868 --> 0:28:03.458 had to train them on your voice. You had to 0:28:03.458 --> 0:28:05.888 train them. They didn't work in the wild. You had 0:28:05.888 --> 0:28:10.538 to use the proper microphone and and quiet environment. And 0:28:10.538 --> 0:28:14.198 now people expect to, you know, buy a new smartphone, 0:28:14.198 --> 0:28:16.718 open it up, walk down the street and start talking. Right. 0:28:17.228 --> 0:28:17.858 True. 0:28:17.858 --> 0:28:20.888 And it doesn't do a perfect job of transcribing what 0:28:20.888 --> 0:28:23.438 people say, but it does a pretty good job. Mm. 0:28:23.528 --> 0:28:27.818 And and you're just thinking of the advance that we've got. 0:28:27.848 --> 0:28:31.358 I mean, that was just unthought of 20 years ago 0:28:31.358 --> 0:28:33.608 to think the idea that it wouldn't be trained for 0:28:33.608 --> 0:28:36.248 your voice, it would work in the wild with all 0:28:36.248 --> 0:28:39.338 the street noise and wind around you. And you could 0:28:39.338 --> 0:28:41.528 just you could just start talking and it would get, 0:28:41.528 --> 0:28:44.798 you know, 90% of the words. Right. Mm. Is pretty, 0:28:45.308 --> 0:28:46.688 you know, someone who's been working in the field for 0:28:46.688 --> 0:28:49.508 those 20 years, I find that pretty amazing. You know, 0:28:49.508 --> 0:28:51.728 it's still not it's still not good enough. It's still 0:28:51.728 --> 0:28:55.478 not perfect. You still you still, you know, shouldn't stake 0:28:55.508 --> 0:28:58.778 your life on it. Um, but for, you know, wandering 0:28:58.778 --> 0:29:03.458 around a town, strange environment, and, you know, a country 0:29:03.458 --> 0:29:05.528 where they speak a different language, it's good enough to, 0:29:05.528 --> 0:29:07.148 you know, make yourself understood. 0:29:07.148 --> 0:29:07.418 Well. 0:29:07.418 --> 0:29:09.788 That's right. Yeah. And look, and I know when I 0:29:09.788 --> 0:29:12.698 first had a look at the Kurzweil personal reader that, um, 0:29:12.698 --> 0:29:15.788 Ray Kurzweil brought out in the mid, the mid late 70s, 0:29:16.298 --> 0:29:18.758 I mean, that thing wasn't perfect either. And I was 0:29:18.758 --> 0:29:21.038 sitting there going, oh my God, who the who? The 0:29:21.038 --> 0:29:24.488 Blazers can afford something that's worth about $55,000 Australian. And 0:29:24.488 --> 0:29:27.428 now it's in our pocket. Yeah. Um, and then I 0:29:27.428 --> 0:29:30.698 remember when I got my first Apple Tui, the synthesizer 0:29:30.698 --> 0:29:34.028 was even worse than the original Daleks on Doctor Who, 0:29:34.868 --> 0:29:37.058 because for a long time it kept saying to me 0:29:37.058 --> 0:29:40.898 there was an unclosed error on the Apple Tui. And 0:29:40.898 --> 0:29:44.048 it wasn't until somebody started said it's actually saying unknown, 0:29:44.048 --> 0:29:48.368 but it's actually saying the K as a C, right. Um, 0:29:48.368 --> 0:29:50.768 and I thought that was only that was only 40 0:29:50.768 --> 0:29:51.518 years ago. 0:29:51.518 --> 0:29:53.408 Yeah. Oh another example. 0:29:53.408 --> 0:29:56.438 I mean subtitles on the TV now. I mean, it 0:29:56.438 --> 0:29:58.448 used to be we people had to transcribe it. The 0:29:58.448 --> 0:30:01.418 only way you could get anything reasonable as a subtitle out. 0:30:01.418 --> 0:30:04.838 Now it's done. Pretty automatic. Not done perfectly, but it's 0:30:04.838 --> 0:30:07.118 good enough to, you know, be able to work out 0:30:07.118 --> 0:30:08.978 what's happening if you can't hear the TV. 0:30:09.188 --> 0:30:12.608 No, no. Exactly, exactly. And look, I'm I mean, I'm 0:30:12.608 --> 0:30:14.318 because I'm an Apple geek. I mean, I'm really looking 0:30:14.318 --> 0:30:17.288 forward to, um, what's coming up at the Worldwide Developers 0:30:17.288 --> 0:30:20.558 Conference at WWDC to see what Siri morphs into. 0:30:21.128 --> 0:30:23.288 Apple has been very slow, though, on the uptake of 0:30:23.288 --> 0:30:29.288 AI that's going to change. Um, and the Apple was 0:30:29.288 --> 0:30:30.848 the great thing about Apple is they're going to do 0:30:30.848 --> 0:30:32.648 it more and more on your device. 0:30:32.798 --> 0:30:32.978 Mm. 0:30:34.248 --> 0:30:36.798 Exactly. Yeah. And that's what I'm looking forward to because 0:30:36.798 --> 0:30:41.118 I really don't want my, you know, my conversations or 0:30:41.118 --> 0:30:43.008 whatever else I might be looking at via an image 0:30:43.008 --> 0:30:45.648 going out to a cloud somewhere, because you've got no 0:30:45.648 --> 0:30:47.808 idea what where the information is going to end up 0:30:47.808 --> 0:30:50.478 on the cloud. So the more stuff is done locally, 0:30:50.928 --> 0:30:52.728 and to me that's going to be more appropriate. 0:30:52.728 --> 0:30:55.218 But that's that's going to be the future. And certainly 0:30:55.218 --> 0:30:58.038 Apple was one of the companies that's been, um, you know, 0:30:58.038 --> 0:31:01.638 promoting that. But the idea is that increasingly you don't 0:31:01.638 --> 0:31:05.088 want to share your data with everyone and anyone. And increasingly, 0:31:05.088 --> 0:31:08.508 we'll have the sophisticated AI algorithms will be small enough 0:31:08.508 --> 0:31:12.708 and smart enough to run actually on your device. Um, 0:31:12.708 --> 0:31:14.598 that solves lots of other problems, and it solves the 0:31:14.598 --> 0:31:17.268 privacy problem, but it also solves the latency problem. So 0:31:17.268 --> 0:31:19.998 there's lots of places where you don't have the connectivity 0:31:20.148 --> 0:31:22.338 you're in, you know, an urban canyon, you're in a tunnel, 0:31:22.338 --> 0:31:26.748 whatever it is. Yeah. Can't depend upon, um, the connectivity 0:31:26.748 --> 0:31:28.308 to be able to, you know, send the data to 0:31:28.308 --> 0:31:31.338 the cloud and have it transcribed or the computer vision 0:31:31.338 --> 0:31:33.108 on your car to do stuff. You need to be 0:31:33.108 --> 0:31:33.348 able to. 0:31:33.348 --> 0:31:34.488 Do it on. 0:31:34.488 --> 0:31:38.358 The device. So but that's the future of AI. Increasingly, 0:31:38.358 --> 0:31:41.628 it's going to be powerful enough to run on the 0:31:41.628 --> 0:31:44.898 limited amount of hardware that you're actually carrying on your person. 0:31:45.108 --> 0:31:47.388 No. And look, that's what I'm looking forward to. I mean, 0:31:47.388 --> 0:31:49.938 I've got I've got about ten questions that I always 0:31:49.938 --> 0:31:53.238 keep checking in with AI every, every three months or so. 0:31:53.328 --> 0:31:57.228 And at the moment it's it it's getting better. Um, 0:31:57.228 --> 0:31:59.568 the first couple of ones that actually broke dramatically on 0:31:59.568 --> 0:32:03.138 my on my questions. Um, I was funny because, um, 0:32:03.138 --> 0:32:05.838 when I was going to interview, I, I checked up 0:32:05.838 --> 0:32:08.388 and it doesn't think that you're a poker player anymore, 0:32:08.388 --> 0:32:09.198 which is cool. 0:32:09.918 --> 0:32:10.368 Um. 0:32:10.938 --> 0:32:13.278 It knows how many bees are in the in the 0:32:13.278 --> 0:32:15.288 webinar now, which is also pretty cool. 0:32:15.348 --> 0:32:17.688 Oh, Taramasalata will call it now, though. 0:32:18.438 --> 0:32:19.758 All right. 0:32:20.868 --> 0:32:24.408 Although I must admit I struggle to spell Taramasalata myself. 0:32:24.408 --> 0:32:25.398 But anyway, yeah. 0:32:25.398 --> 0:32:28.368 Actually my favorite word at university for philosophy was, um, 0:32:28.368 --> 0:32:32.418 reductio ad absurdum. I love that word. Um, I was like, oh, 0:32:32.418 --> 0:32:34.668 that's pretty cool. Um, all right. So look, if people 0:32:34.668 --> 0:32:36.918 want to find out more about your books because I 0:32:36.918 --> 0:32:42.228 know there's there's faking it, there's the 202,062, uh, there's 0:32:42.228 --> 0:32:45.708 machines Behaving badly, which I absolutely love. Um, now there's 0:32:45.708 --> 0:32:47.478 a fourth book, which, for the life of me, I 0:32:47.478 --> 0:32:48.468 can't remember. 0:32:48.738 --> 0:32:52.878 It's alive. Artificial intelligence from the piano to killer robots. Actually, 0:32:52.878 --> 0:32:54.108 the first book I wrote, which is. 0:32:54.108 --> 0:32:54.678 Oh, okay. 0:32:54.678 --> 0:32:55.578 About the history of. 0:32:55.578 --> 0:32:57.228 AI. Okay, now I've. 0:32:57.228 --> 0:32:59.748 Got all of them on Kindle. Have you done anything 0:32:59.748 --> 0:33:01.698 with audible? So people that want to sort of sit 0:33:01.698 --> 0:33:02.538 back with a glass of wine? 0:33:02.568 --> 0:33:03.438 Good news is. 0:33:03.438 --> 0:33:10.338 Coming out next month. I have just recorded my, um, uh, book. 0:33:10.338 --> 0:33:12.378 I've just been put out by Belinda. Um, Faking It 0:33:12.378 --> 0:33:16.008 is coming out in audio, and I have to say, 0:33:16.158 --> 0:33:20.058 I'm an absolute convert. I, I think it is the 0:33:20.058 --> 0:33:25.938 best version of my book. Um, because I realized that, 0:33:25.938 --> 0:33:28.488 you know, I've written it in some places. I write 0:33:28.488 --> 0:33:31.368 it with, you know, I'm writing things and I'm annoyed 0:33:31.368 --> 0:33:35.628 and upset or excited. And if you're if you read 0:33:35.628 --> 0:33:38.148 the page carefully, maybe, hopefully you can tell that in 0:33:38.148 --> 0:33:41.298 whether what I've written. Um, but in the audible book, 0:33:41.298 --> 0:33:43.968 it's very clear when I, when I spoke the book, 0:33:43.968 --> 0:33:48.738 I had the privilege of pleasure of speaking my own book. I, 0:33:48.738 --> 0:33:50.958 you know, I could laugh at the jokes in the book. 0:33:50.958 --> 0:33:53.868 I can I can express disgust with what the tech 0:33:53.868 --> 0:33:57.018 companies are doing with my data. Um, so there's a 0:33:57.018 --> 0:33:59.928 lot more information conveyed in the audible book that's not 0:33:59.928 --> 0:34:01.638 conveyed in the written book. Maybe. 0:34:02.738 --> 0:34:05.018 And, uh, because I, when I was reading the stuff 0:34:05.018 --> 0:34:08.888 about the, um, the local autonomous weapon systems, I thought 0:34:08.888 --> 0:34:12.908 that's just absolutely appalling. Giving, giving drones the capability to 0:34:12.908 --> 0:34:15.458 be able to actually kill people without any human intervention. 0:34:15.458 --> 0:34:17.918 That's just getting a bit beyond the pale. Basically it. 0:34:17.918 --> 0:34:18.248 Is. 0:34:18.638 --> 0:34:21.158 Um, and I just thought, oh, I wonder what you 0:34:21.158 --> 0:34:25.148 would have sounded like reading that particular particular paragraph, because. 0:34:25.298 --> 0:34:26.858 Next month you could discover. 0:34:26.858 --> 0:34:28.448 There you go, I will. 0:34:28.838 --> 0:34:32.048 The great thing about the book was I actually only 0:34:32.048 --> 0:34:33.308 speak half the book. 0:34:33.518 --> 0:34:34.508 Oh, because the. 0:34:34.508 --> 0:34:38.318 Book is all about how ChatGPT and examples like that 0:34:38.318 --> 0:34:44.948 make mistakes and get things wrong and, um, amuse us and, and, um, 0:34:45.098 --> 0:34:47.348 I said to the publisher, I said, well, I could 0:34:47.348 --> 0:34:50.588 read this. Um, but why should I read this? Why 0:34:50.588 --> 0:34:52.718 not get the computer to read it? So I say 0:34:52.718 --> 0:34:54.938 half of it and ChatGPT says the other half. 0:34:55.088 --> 0:34:57.698 Yeah. Does it use your voice to do the the 0:34:57.698 --> 0:34:59.948 second bit, or does it just use a generic computer 0:34:59.948 --> 0:35:00.998 synthesizer voice? 0:35:00.998 --> 0:35:01.238 Oh, no. 0:35:01.238 --> 0:35:04.988 They've, um, they're getting a very nice, uh, synthesized voice 0:35:04.988 --> 0:35:07.898 for that. Um, so the publisher is very excited. They 0:35:07.898 --> 0:35:10.418 said this is the first book. They've they've really, um, 0:35:10.418 --> 0:35:14.288 embraced the technology. Right. And it's so, so it is 0:35:14.288 --> 0:35:18.398 a conversation between me and the computer, and I hope 0:35:18.398 --> 0:35:18.878 you enjoy it.