1 00:00:00,560 --> 00:00:06,260 you're watching the context it's time 2 00:00:02,560 --> 00:00:08,519 for our new weekly segment AI 3 00:00:06,260 --> 00:00:11,679 [Music] 4 00:00:08,519 --> 00:00:13,360 decoded welcome to AI decoded that time 5 00:00:11,679 --> 00:00:15,360 of the week when we look in depth at 6 00:00:13,360 --> 00:00:18,400 some of the most eye-catching stories in 7 00:00:15,360 --> 00:00:21,039 the world of artificial intelligence and 8 00:00:18,400 --> 00:00:24,640 we begin New York Times who report 9 00:00:21,039 --> 00:00:26,960 Google has been fined $271 Million by 10 00:00:24,640 --> 00:00:29,000 France's competition Watchdog for 11 00:00:26,960 --> 00:00:31,519 failing to broker agreements with media 12 00:00:29,000 --> 00:00:35,079 outlets for using their content to train 13 00:00:31,519 --> 00:00:38,079 its AI Tech which leads us to a possible 14 00:00:35,079 --> 00:00:40,079 solution after leading AI developer open 15 00:00:38,079 --> 00:00:43,239 AI told the UK Parliament it's 16 00:00:40,079 --> 00:00:46,160 impossible to train leading AI models 17 00:00:43,239 --> 00:00:48,079 without using uh copyrighted materials a 18 00:00:46,160 --> 00:00:50,120 group of researchers says there is an 19 00:00:48,079 --> 00:00:52,359 alternative after releasing what's 20 00:00:50,120 --> 00:00:54,640 thought to be the largest AI training 21 00:00:52,359 --> 00:00:57,879 data set composed entirely of text 22 00:00:54,640 --> 00:01:00,640 that's in the public domain not under 23 00:00:57,879 --> 00:01:02,760 copyright AP News looks at the UN 24 00:01:00,640 --> 00:01:05,080 general assembly vote on what would be 25 00:01:02,760 --> 00:01:06,320 the first United Nations resolution on 26 00:01:05,080 --> 00:01:08,960 artificial 27 00:01:06,320 --> 00:01:10,880 intelligence and The Verge Tech website 28 00:01:08,960 --> 00:01:12,840 has a video of one of their journalists 29 00:01:10,880 --> 00:01:16,000 having a real-time conversation with an 30 00:01:12,840 --> 00:01:17,840 AI Avatar who responds to human speech 31 00:01:16,000 --> 00:01:19,960 currently it's an experimental program 32 00:01:17,840 --> 00:01:21,840 developed by video game developer 33 00:01:19,960 --> 00:01:24,479 Ubisoft and we'll show you a clip of 34 00:01:21,840 --> 00:01:27,240 that a little later actually in the Ft 35 00:01:24,479 --> 00:01:29,119 an Uber Eats delivery worker who after 36 00:01:27,240 --> 00:01:31,158 becoming fed up with the company's app 37 00:01:29,119 --> 00:01:33,880 consistently made making errors decided 38 00:01:31,158 --> 00:01:36,000 to rewrite the code to fix the issues we 39 00:01:33,880 --> 00:01:38,840 have the author of that article with us 40 00:01:36,000 --> 00:01:41,079 and she'll tell us a bit more and AFP 41 00:01:38,840 --> 00:01:43,399 news features novelist salmon rushy who 42 00:01:41,079 --> 00:01:45,600 says artificial intelligence tools may 43 00:01:43,399 --> 00:01:47,920 pose a threat to writers of Thrillers 44 00:01:45,600 --> 00:01:50,320 and science fiction but they lack the 45 00:01:47,920 --> 00:01:53,280 originality and humor to challenge 46 00:01:50,320 --> 00:01:55,920 serious novelists and another artist 47 00:01:53,280 --> 00:01:58,360 musician James Blunt says he felt 48 00:01:55,920 --> 00:02:00,439 humiliated at how bad the results were 49 00:01:58,360 --> 00:02:04,920 when he experimented with AI to see if 50 00:02:00,439 --> 00:02:08,440 it could create lyrics accurately in his 51 00:02:04,920 --> 00:02:10,560 style well with me here is mad murgia 52 00:02:08,440 --> 00:02:12,599 the financial times's artificial 53 00:02:10,560 --> 00:02:14,160 intelligence editor thanks very much for 54 00:02:12,599 --> 00:02:16,519 coming on the program thanks for having 55 00:02:14,160 --> 00:02:18,560 me right lots to get through quite a 56 00:02:16,519 --> 00:02:20,720 busy week I think we can probably almost 57 00:02:18,560 --> 00:02:22,680 say that every week now let's uh let's 58 00:02:20,720 --> 00:02:26,560 start with the New York Times France 59 00:02:22,680 --> 00:02:28,920 finds Google amid AI dispute with news 60 00:02:26,560 --> 00:02:30,640 media so this is a kind of broad issue 61 00:02:28,920 --> 00:02:33,000 that we're going to come up against 62 00:02:30,640 --> 00:02:34,920 quite a lot and already have but a 63 00:02:33,000 --> 00:02:37,120 seemingly significant moment what's 64 00:02:34,920 --> 00:02:38,720 happened here yeah so this is you know 65 00:02:37,120 --> 00:02:40,319 there's been as you say a long running 66 00:02:38,720 --> 00:02:42,640 dispute between Google and news 67 00:02:40,319 --> 00:02:45,640 organizations about how Google has been 68 00:02:42,640 --> 00:02:48,840 using links uh news media newspaper 69 00:02:45,640 --> 00:02:50,560 articles and so on this particular fine 70 00:02:48,840 --> 00:02:54,120 um they've said that they've failed to 71 00:02:50,560 --> 00:02:56,040 negotiate Fair deals um and have used 72 00:02:54,120 --> 00:02:58,000 data from newspapers to train their 73 00:02:56,040 --> 00:03:00,519 large language models the chat Bots that 74 00:02:58,000 --> 00:03:02,640 many of us have been using so this is 75 00:03:00,519 --> 00:03:04,519 part of a wider ruling that they've made 76 00:03:02,640 --> 00:03:06,720 and said Google hasn't negotiated in 77 00:03:04,519 --> 00:03:09,519 good faith and and saying it's failing 78 00:03:06,720 --> 00:03:12,360 to inform Publishers of the use of their 79 00:03:09,519 --> 00:03:15,080 content for their 80 00:03:12,360 --> 00:03:19,760 software isn't that something that we're 81 00:03:15,080 --> 00:03:22,159 kind of all the big AI kind of companies 82 00:03:19,760 --> 00:03:24,560 all in the same boat here is everything 83 00:03:22,159 --> 00:03:27,360 kind of trained the same way absolutely 84 00:03:24,560 --> 00:03:30,159 it's not just Google open AI too um you 85 00:03:27,360 --> 00:03:32,159 know they are uh funded by Microsoft 86 00:03:30,159 --> 00:03:34,439 primarily they also that they have one 87 00:03:32,159 --> 00:03:36,599 of the most powerful models which um 88 00:03:34,439 --> 00:03:38,959 Powers chat GPT which people might have 89 00:03:36,599 --> 00:03:41,319 played around with that too is a large 90 00:03:38,959 --> 00:03:43,760 language model trained on a huge Corpus 91 00:03:41,319 --> 00:03:46,400 of data found online New York Times 92 00:03:43,760 --> 00:03:48,879 itself actually has sued open AI um and 93 00:03:46,400 --> 00:03:51,400 is currently in in this court case with 94 00:03:48,879 --> 00:03:53,239 them saying that they've illegally used 95 00:03:51,400 --> 00:03:55,040 their copyrighted material and and 96 00:03:53,239 --> 00:03:56,840 that's kind of another ongoing battle on 97 00:03:55,040 --> 00:03:58,760 another front interesting and so we're 98 00:03:56,840 --> 00:04:01,239 going to stick with exactly this theme 99 00:03:58,760 --> 00:04:04,760 because they are now clay this is uh 100 00:04:01,239 --> 00:04:08,840 wired here's proof you can train an AI 101 00:04:04,760 --> 00:04:11,480 model without slurping copyrighted 102 00:04:08,840 --> 00:04:13,599 content um not often you get slurping in 103 00:04:11,480 --> 00:04:15,760 a headline but I like it uh so what's 104 00:04:13,599 --> 00:04:17,600 this story about so this is kind of look 105 00:04:15,760 --> 00:04:20,279 this is the other side of the coin right 106 00:04:17,600 --> 00:04:22,400 because the leaders of these companies 107 00:04:20,279 --> 00:04:23,840 you know open AI in particular has said 108 00:04:22,400 --> 00:04:25,680 there's no other way to build these 109 00:04:23,840 --> 00:04:28,400 models if we want really powerful 110 00:04:25,680 --> 00:04:30,400 language models if we want chatbots we 111 00:04:28,400 --> 00:04:33,080 need to use the the words on the 112 00:04:30,400 --> 00:04:35,320 internet we need this data um but the 113 00:04:33,080 --> 00:04:37,039 the research that's come out of here um 114 00:04:35,320 --> 00:04:38,400 it's a group of researchers backed by 115 00:04:37,039 --> 00:04:40,560 the French government they've shown that 116 00:04:38,400 --> 00:04:43,600 actually there are other ways there are 117 00:04:40,560 --> 00:04:45,880 alternatives um you can make data sets 118 00:04:43,600 --> 00:04:48,520 that power maybe smaller models but for 119 00:04:45,880 --> 00:04:50,199 specific use cases um an example they 120 00:04:48,520 --> 00:04:51,560 give here is for for a law firm for 121 00:04:50,199 --> 00:04:53,440 example so if you're trying to build a 122 00:04:51,560 --> 00:04:55,720 language model specifically to help with 123 00:04:53,440 --> 00:04:57,199 lawyers and the work that they do and 124 00:04:55,720 --> 00:04:59,400 this could be true in scientific 125 00:04:57,199 --> 00:05:01,840 research or any sort of vertical that 126 00:04:59,400 --> 00:05:04,800 that want to apply it to you can you can 127 00:05:01,840 --> 00:05:07,280 use you know less data and it can be 128 00:05:04,800 --> 00:05:09,720 paid for interesting so that's one 129 00:05:07,280 --> 00:05:12,880 potential solution but again on those 130 00:05:09,720 --> 00:05:14,759 narrow use cases or narrower use cases 131 00:05:12,880 --> 00:05:18,160 because it says in here that yeah the 132 00:05:14,759 --> 00:05:20,639 data set is Tiny compared to what lots 133 00:05:18,160 --> 00:05:22,960 of the other language models were kind 134 00:05:20,639 --> 00:05:25,199 of based on so surely there must be some 135 00:05:22,960 --> 00:05:27,360 kind of difference in quality of outcome 136 00:05:25,199 --> 00:05:29,120 or may Maybe not maybe you don't you 137 00:05:27,360 --> 00:05:31,680 don't need that you know obviously the 138 00:05:29,120 --> 00:05:33,440 internet the entire swell of information 139 00:05:31,680 --> 00:05:36,440 that we all put out there from Reddit 140 00:05:33,440 --> 00:05:38,360 post to you know Amazon comments and 141 00:05:36,440 --> 00:05:40,120 reviews um but there's also a lot of 142 00:05:38,360 --> 00:05:42,199 noise in that data right there's there's 143 00:05:40,120 --> 00:05:44,039 a lot of rubbish on the internet as as 144 00:05:42,199 --> 00:05:46,759 everyone can attest to so it's not 145 00:05:44,039 --> 00:05:48,880 necessarily clean good quality data so 146 00:05:46,759 --> 00:05:50,800 there is there is still a it's an open 147 00:05:48,880 --> 00:05:53,360 question of whether it's quantity or 148 00:05:50,800 --> 00:05:55,280 quality um and I think that there is an 149 00:05:53,360 --> 00:05:58,520 argument to be made for good quality 150 00:05:55,280 --> 00:06:01,360 data interesting right let's move on to 151 00:05:58,520 --> 00:06:04,600 a bit of high level regulation now 152 00:06:01,360 --> 00:06:06,360 because it's hugely important there are 153 00:06:04,600 --> 00:06:08,440 as we've been witnessing over the last 154 00:06:06,360 --> 00:06:11,960 kind of six months especially these big 155 00:06:08,440 --> 00:06:15,599 attempts uh at regulation the United 156 00:06:11,960 --> 00:06:18,160 Nations now hoping to get in on the ACT 157 00:06:15,599 --> 00:06:20,880 what's going on there so today the 158 00:06:18,160 --> 00:06:22,800 general assembly is set to vote um on 159 00:06:20,880 --> 00:06:25,160 what's going to be the first United 160 00:06:22,800 --> 00:06:27,240 Nations resolution on artificial 161 00:06:25,160 --> 00:06:29,000 intelligence um the goal they're hoping 162 00:06:27,240 --> 00:06:31,199 that it will be unanimous and the the 163 00:06:29,000 --> 00:06:33,159 goal really is to bridge inequities 164 00:06:31,199 --> 00:06:35,759 between um you know the developed 165 00:06:33,159 --> 00:06:37,840 Western World and uh countries in the in 166 00:06:35,759 --> 00:06:39,479 the global South developing world and 167 00:06:37,840 --> 00:06:40,960 make sure that those countries have a 168 00:06:39,479 --> 00:06:42,880 seat at the table when it comes to 169 00:06:40,960 --> 00:06:45,080 developing the Technologies and it isn't 170 00:06:42,880 --> 00:06:47,520 just places like the United States where 171 00:06:45,080 --> 00:06:50,159 these companies are based that you know 172 00:06:47,520 --> 00:06:51,800 see all the upside of the technology 173 00:06:50,159 --> 00:06:54,639 interesting yeah because we're picking 174 00:06:51,800 --> 00:06:57,159 up these these quotes from AP here and 175 00:06:54,639 --> 00:07:00,440 they're saying they want to make it safe 176 00:06:57,159 --> 00:07:02,240 secure trustworthy of those broad 177 00:07:00,440 --> 00:07:04,319 principles that I think most people 178 00:07:02,240 --> 00:07:05,960 would kind of ascribe to but really 179 00:07:04,319 --> 00:07:08,639 interesting picking up on exactly what 180 00:07:05,960 --> 00:07:11,479 you were talking about there it's it's 181 00:07:08,639 --> 00:07:14,840 Distributing the potential benefits not 182 00:07:11,479 --> 00:07:17,240 just having a few companies in a few 183 00:07:14,840 --> 00:07:19,919 countries and therefore the populations 184 00:07:17,240 --> 00:07:23,240 and people of those uh countries 185 00:07:19,919 --> 00:07:26,560 benefiting but making sure it's across 186 00:07:23,240 --> 00:07:28,560 the world uh that's a laudable aim I'm 187 00:07:26,560 --> 00:07:30,879 sure but uh probably easier said than 188 00:07:28,560 --> 00:07:32,159 done or absolutely I mean this is 189 00:07:30,879 --> 00:07:34,639 they're supposed to they're doing this 190 00:07:32,159 --> 00:07:37,639 so that they can help AI kind of achieve 191 00:07:34,639 --> 00:07:39,240 the UN development goals for 2030 which 192 00:07:37,639 --> 00:07:40,759 they're far behind on so they're hoping 193 00:07:39,240 --> 00:07:42,960 this can address questions you know 194 00:07:40,759 --> 00:07:45,479 inequities in health care education and 195 00:07:42,960 --> 00:07:47,520 so on but these are big problems and I'm 196 00:07:45,479 --> 00:07:50,000 not sure AI is the Magic Bullet to all 197 00:07:47,520 --> 00:07:51,319 of it but it's a step it's a step uh 198 00:07:50,000 --> 00:07:53,479 yeah some of the examples detecting 199 00:07:51,319 --> 00:07:55,639 diseases predicting floods helping 200 00:07:53,479 --> 00:07:57,000 Farmers training the next generation of 201 00:07:55,639 --> 00:08:01,080 workers that's what they want to make 202 00:07:57,000 --> 00:08:04,000 sure is equally spread across the globe 203 00:08:01,080 --> 00:08:05,720 right next story um before we talk about 204 00:08:04,000 --> 00:08:07,840 it we're going to actually see it 205 00:08:05,720 --> 00:08:10,199 because we're going to take uh a moment 206 00:08:07,840 --> 00:08:13,199 to watch this example of a real-time 207 00:08:10,199 --> 00:08:15,479 conversation with an AI Avatar this is 208 00:08:13,199 --> 00:08:17,680 not yet released it's a creative 209 00:08:15,479 --> 00:08:19,960 experiment by video game developer 210 00:08:17,680 --> 00:08:23,560 Ubisoft that suggests the company has 211 00:08:19,960 --> 00:08:26,680 made small breakthroughs in AI realtime 212 00:08:23,560 --> 00:08:28,759 emotion and animation contextual 213 00:08:26,680 --> 00:08:31,080 awareness and collaborative 214 00:08:28,759 --> 00:08:35,760 decision-making let's take a 215 00:08:31,080 --> 00:08:35,760 look tell me one word why I should 216 00:08:38,279 --> 00:08:42,838 stay 217 00:08:39,959 --> 00:08:45,040 passion you can feel it can't you all 218 00:08:42,838 --> 00:08:47,640 right that burning desire to make a 219 00:08:45,040 --> 00:08:50,240 difference to fight for what's 220 00:08:47,640 --> 00:08:51,519 right that's what drives us and that's 221 00:08:50,240 --> 00:08:53,440 what will keep you going through the 222 00:08:51,519 --> 00:08:55,800 tough times and trust 223 00:08:53,440 --> 00:09:00,720 me there will be tough 224 00:08:55,800 --> 00:09:03,040 times but together we can overcome them 225 00:09:00,720 --> 00:09:03,880 I have too much to lose I have a family 226 00:09:03,040 --> 00:09:08,079 two 227 00:09:03,880 --> 00:09:08,079 daughters really should I be doing 228 00:09:10,519 --> 00:09:15,959 this family is everything 229 00:09:13,200 --> 00:09:19,160 Nelson and that's why we're fighting to 230 00:09:15,959 --> 00:09:22,160 create a better world for our loved ones 231 00:09:19,160 --> 00:09:24,399 we can't continue living in fear and 232 00:09:22,160 --> 00:09:27,000 oppression it's a risk but it's a risk 233 00:09:24,399 --> 00:09:29,519 worth taking for a brighter 234 00:09:27,000 --> 00:09:31,839 future and don't worry we'll have each 235 00:09:29,519 --> 00:09:31,839 other's 236 00:09:32,240 --> 00:09:36,680 backs have you lost anybody close to 237 00:09:39,680 --> 00:09:43,279 you 238 00:09:41,279 --> 00:09:46,000 yes I've lost 239 00:09:43,279 --> 00:09:48,839 many my son my friends my 240 00:09:46,000 --> 00:09:51,000 home but every loss only fuels my 241 00:09:48,839 --> 00:09:53,560 determination to keep 242 00:09:51,000 --> 00:09:54,760 going and through the resistance I found 243 00:09:53,560 --> 00:09:57,399 a new 244 00:09:54,760 --> 00:09:59,720 family together we can make a difference 245 00:09:57,399 --> 00:10:03,000 and prevent others from experiencing ing 246 00:09:59,720 --> 00:10:05,360 the same pain right slightly strange to 247 00:10:03,000 --> 00:10:07,560 watch that what was going on so that is 248 00:10:05,360 --> 00:10:09,920 an avatar powered by Ai and what's 249 00:10:07,560 --> 00:10:12,320 unique about it is it's not scripted so 250 00:10:09,920 --> 00:10:14,160 it's just responding to you in real time 251 00:10:12,320 --> 00:10:16,399 conversation that's what's kind of new 252 00:10:14,160 --> 00:10:18,560 about this and a bit stunted but you can 253 00:10:16,399 --> 00:10:20,440 see the kind of potential there for for 254 00:10:18,560 --> 00:10:22,399 the next generation of games it's quite 255 00:10:20,440 --> 00:10:23,880 haunting wasn't it watching it which I 256 00:10:22,399 --> 00:10:25,640 suppose is part of the desired effect 257 00:10:23,880 --> 00:10:28,040 but if that is basically still in the 258 00:10:25,640 --> 00:10:29,839 kind of concept phase that seems I don't 259 00:10:28,040 --> 00:10:32,200 know pretty impressive to me even if it 260 00:10:29,839 --> 00:10:34,519 is just a kind of fancy chat bot at the 261 00:10:32,200 --> 00:10:36,800 moment it's still deeply impressive 262 00:10:34,519 --> 00:10:40,240 right let's move on to your story ft the 263 00:10:36,800 --> 00:10:43,320 delivery Rider who took on his faceless 264 00:10:40,240 --> 00:10:45,440 boss is the headline what's this so This 265 00:10:43,320 --> 00:10:47,120 is actually from my book which is which 266 00:10:45,440 --> 00:10:49,120 came out today which is exciting but 267 00:10:47,120 --> 00:10:50,760 this is about um it's called code 268 00:10:49,120 --> 00:10:52,760 dependent and it's about human beings 269 00:10:50,760 --> 00:10:55,120 who've been impacted by AI systems 270 00:10:52,760 --> 00:10:57,120 unexpectedly and Armin Sami who's the 271 00:10:55,120 --> 00:10:59,600 Uber Eats driver in this story found 272 00:10:57,120 --> 00:11:01,279 that he was being underpaid consistently 273 00:10:59,600 --> 00:11:02,920 or in one case and he wanted to figure 274 00:11:01,279 --> 00:11:04,519 out was this a problem that occurred 275 00:11:02,920 --> 00:11:07,040 again and again and because the 276 00:11:04,519 --> 00:11:09,959 algorithms that kind of govern work on 277 00:11:07,040 --> 00:11:11,519 gig apps are so opaque you know people 278 00:11:09,959 --> 00:11:13,440 the people who work for these apps have 279 00:11:11,519 --> 00:11:15,600 no idea why they're being paid what 280 00:11:13,440 --> 00:11:17,480 they're being paid he had to hack it he 281 00:11:15,600 --> 00:11:19,600 essentially built a tool that could 282 00:11:17,480 --> 00:11:20,880 figure out how far he was traveling and 283 00:11:19,600 --> 00:11:23,079 therefore how much he should have been 284 00:11:20,880 --> 00:11:25,040 paid he then made this free for other 285 00:11:23,079 --> 00:11:27,200 rubber drivers to use and they've all 286 00:11:25,040 --> 00:11:28,440 been using it around the world to figure 287 00:11:27,200 --> 00:11:30,800 out you know whether they've been 288 00:11:28,440 --> 00:11:33,399 underpaid which had that is an 289 00:11:30,800 --> 00:11:35,320 absolutely incredible story I wish we 290 00:11:33,399 --> 00:11:37,720 had more time for it unfortunately we 291 00:11:35,320 --> 00:11:39,360 don't last issue we've got to move on 292 00:11:37,720 --> 00:11:43,200 because we are unfortunately nearly out 293 00:11:39,360 --> 00:11:47,040 of time Salon rushy AI only poses a 294 00:11:43,200 --> 00:11:49,800 threat to unoriginal writers and also we 295 00:11:47,040 --> 00:11:52,680 can look at at the same time James Blunt 296 00:11:49,800 --> 00:11:55,279 humiliated by generic AI versions of his 297 00:11:52,680 --> 00:11:57,240 Lyrics these are artists what's going on 298 00:11:55,279 --> 00:11:59,360 here so this is really about creativity 299 00:11:57,240 --> 00:12:02,120 and whether artificial intelligence can 300 00:11:59,360 --> 00:12:05,320 ever be creative right whether that's 301 00:12:02,120 --> 00:12:07,920 artists you know voice over actors uh 302 00:12:05,320 --> 00:12:10,920 writers journalists like ask you know 303 00:12:07,920 --> 00:12:12,959 can we be replaced by Ai and so far the 304 00:12:10,920 --> 00:12:15,240 answer seems to be no it's it's pretty 305 00:12:12,959 --> 00:12:18,079 generic James Blunt actually finds it 306 00:12:15,240 --> 00:12:20,360 you know humiliating as he says nothing 307 00:12:18,079 --> 00:12:23,199 like his real lyrics um so I don't think 308 00:12:20,360 --> 00:12:24,680 it's there yet um if to rep coming to 309 00:12:23,199 --> 00:12:26,440 replace us well so yeah you didn't have 310 00:12:24,680 --> 00:12:27,920 to bring journalists into it at the end 311 00:12:26,440 --> 00:12:30,120 no one was talking about them until you 312 00:12:27,920 --> 00:12:31,800 brought it in right alter time thank you 313 00:12:30,120 --> 00:12:34,040 so much for coming in and talking us 314 00:12:31,800 --> 00:12:38,959 through the brilliant stuff that's it 315 00:12:34,040 --> 00:12:38,959 we'll do this again same time next week