1 00:00:00,000 --> 00:00:01,880 Welcome to our deep dive. 2 00:00:01,880 --> 00:00:05,160 We're gonna be looking into AI and thought decoding. 3 00:00:05,160 --> 00:00:06,000 Yeah. 4 00:00:06,000 --> 00:00:07,840 And you know how we all crave a little privacy, right? 5 00:00:07,840 --> 00:00:11,320 Like Charlotte Bronte even called the human mind 6 00:00:11,320 --> 00:00:12,760 a sacred sanctuary, right? 7 00:00:12,760 --> 00:00:13,600 Yeah. 8 00:00:15,360 --> 00:00:18,320 But what if technology could peek behind the curtain 9 00:00:18,320 --> 00:00:19,920 of our thoughts? 10 00:00:19,920 --> 00:00:23,880 What if AI could actually like read our minds? 11 00:00:23,880 --> 00:00:25,880 It sounds like something out of science fiction. 12 00:00:25,880 --> 00:00:26,720 Right. 13 00:00:26,720 --> 00:00:28,680 But researchers are using cutting edge AI 14 00:00:28,680 --> 00:00:33,040 and brain scans to, well, to do just that. 15 00:00:33,040 --> 00:00:35,440 Yeah, what's really fascinating is that this technology 16 00:00:35,440 --> 00:00:38,360 could actually help people who've lost their ability 17 00:00:38,360 --> 00:00:39,200 to communicate. 18 00:00:39,200 --> 00:00:40,040 Oh, wow. 19 00:00:40,040 --> 00:00:42,160 Due to conditions like, you know, locked in syndrome. 20 00:00:42,160 --> 00:00:43,000 Right. 21 00:00:43,000 --> 00:00:44,320 Imagine giving them a voice again. 22 00:00:44,320 --> 00:00:46,360 It's amazing and a little unsettling 23 00:00:46,360 --> 00:00:47,880 at the same time, right? 24 00:00:47,880 --> 00:00:48,800 Definitely. 25 00:00:48,800 --> 00:00:51,600 Today we're going to explore two groundbreaking approaches 26 00:00:51,600 --> 00:00:52,520 to thought decoding. 27 00:00:52,520 --> 00:00:53,360 Okay. 28 00:00:53,360 --> 00:00:55,640 One using fMRI and the other using EEG. 29 00:00:55,640 --> 00:00:56,480 All right. 30 00:00:56,480 --> 00:00:58,600 I've got tons of articles and research papers 31 00:00:58,600 --> 00:01:00,720 and I'm so excited to like break it all down. 32 00:01:00,720 --> 00:01:02,080 Yeah, we're going to unpack the science 33 00:01:02,080 --> 00:01:03,760 behind these technologies. 34 00:01:03,760 --> 00:01:06,000 Learn about the incredible breakthroughs 35 00:01:06,000 --> 00:01:08,600 researchers are making and then delve 36 00:01:08,600 --> 00:01:12,720 into the ethical considerations of this mind-boggling field. 37 00:01:12,720 --> 00:01:15,280 Okay, so to understand where we are today, 38 00:01:15,280 --> 00:01:17,000 let's take a quick trip back in time. 39 00:01:17,000 --> 00:01:17,960 Okay. 40 00:01:17,960 --> 00:01:19,960 For centuries, humans have tried to figure out 41 00:01:19,960 --> 00:01:20,880 how the mind works. 42 00:01:20,880 --> 00:01:21,920 Right. 43 00:01:21,920 --> 00:01:26,920 We have some seriously wacky ideas like phenology. 44 00:01:27,000 --> 00:01:27,840 Oh, yeah. 45 00:01:27,840 --> 00:01:30,480 They claim that you could like read someone's personality 46 00:01:30,480 --> 00:01:32,040 based on the bumps on their head. 47 00:01:32,040 --> 00:01:32,880 Right. 48 00:01:32,880 --> 00:01:33,720 Obviously that was a total bust. 49 00:01:33,720 --> 00:01:34,720 Yeah, total bust. 50 00:01:34,720 --> 00:01:37,040 But it shows how deeply we've always wanted 51 00:01:37,040 --> 00:01:39,600 to understand the inner workings of the mind. 52 00:01:39,600 --> 00:01:40,440 Right. 53 00:01:40,440 --> 00:01:41,720 And eventually we developed tools 54 00:01:41,720 --> 00:01:44,000 that actually measure brain activity, 55 00:01:44,000 --> 00:01:46,440 like EEG and fMRI. 56 00:01:46,440 --> 00:01:48,200 Okay, so before we dive into the research, 57 00:01:48,200 --> 00:01:51,280 let's break down how EEG and fMRI work. 58 00:01:51,280 --> 00:01:52,120 Okay. 59 00:01:52,120 --> 00:01:53,440 I'll admit I was a little fuzzy on the details 60 00:01:53,440 --> 00:01:55,160 before this deep dive. 61 00:01:55,160 --> 00:01:56,000 Right. 62 00:01:56,000 --> 00:01:58,440 So the idea of EEG, like those noise-canceling headphones 63 00:01:58,440 --> 00:02:00,560 that filter out distractions, 64 00:02:00,560 --> 00:02:03,160 it uses a cap with electrodes to measure 65 00:02:03,160 --> 00:02:05,920 the electrical signals produced by your brain. 66 00:02:05,920 --> 00:02:10,920 And since neurons communicate using electrical impulses, 67 00:02:10,920 --> 00:02:13,200 EEG allows us to kind of eavesdrop 68 00:02:13,200 --> 00:02:15,240 on the brain's electrical symphony. 69 00:02:15,240 --> 00:02:16,160 Oh. 70 00:02:16,160 --> 00:02:17,800 Even if we can't pinpoint exactly 71 00:02:17,800 --> 00:02:19,240 where each instrument is playing. 72 00:02:19,240 --> 00:02:20,080 Got it. 73 00:02:20,080 --> 00:02:22,760 So EEG is like listening to the brain's orchestra 74 00:02:22,760 --> 00:02:24,920 through these tiny microphones, 75 00:02:24,920 --> 00:02:27,560 picking up the overall rhythm and harmony. 76 00:02:27,560 --> 00:02:28,400 Exactly. 77 00:02:28,400 --> 00:02:30,280 And what about fMRI? 78 00:02:30,280 --> 00:02:33,480 So fMRI is like taking a high-resolution snapshot 79 00:02:33,480 --> 00:02:34,720 of the brain in action. 80 00:02:34,720 --> 00:02:35,560 Okay. 81 00:02:35,560 --> 00:02:37,400 It measures changes in blood flow, 82 00:02:37,400 --> 00:02:39,160 which is like a proxy for brain activity. 83 00:02:39,160 --> 00:02:40,000 Right. 84 00:02:40,000 --> 00:02:42,600 When a particular brain region works harder, 85 00:02:42,600 --> 00:02:45,720 it needs more oxygen, so blood flow increases. 86 00:02:45,720 --> 00:02:50,720 fMRI detects these changes and creates detailed images 87 00:02:52,320 --> 00:02:53,840 of which brain regions are active 88 00:02:53,840 --> 00:02:57,000 during different tasks, like thinking, speaking, 89 00:02:57,000 --> 00:02:58,080 or even daydreaming. 90 00:02:58,080 --> 00:03:00,920 So fMRI gives us like a visual map of the brain. 91 00:03:00,920 --> 00:03:01,760 Exactly. 92 00:03:01,760 --> 00:03:03,120 Highlighting the hotspots of activity. 93 00:03:03,120 --> 00:03:03,960 Yeah, yeah. 94 00:03:03,960 --> 00:03:06,320 But how do we go from brain scans to actual thoughts? 95 00:03:06,320 --> 00:03:07,600 That's where AI comes in, right? 96 00:03:07,600 --> 00:03:08,440 Exactly. 97 00:03:08,440 --> 00:03:09,480 And this is where it gets really cool. 98 00:03:09,480 --> 00:03:12,520 The key idea is that every thought 99 00:03:12,520 --> 00:03:15,160 creates a unique pattern of brain activity, 100 00:03:15,160 --> 00:03:16,320 like a neural fingerprint. 101 00:03:16,320 --> 00:03:17,320 Interesting. 102 00:03:17,320 --> 00:03:20,080 AI is learning to recognize these fingerprints 103 00:03:20,080 --> 00:03:22,520 and decode the thoughts that they represent. 104 00:03:22,520 --> 00:03:25,240 So it's like teaching a computer to read a new language, 105 00:03:25,240 --> 00:03:27,000 like the language of brain activity. 106 00:03:27,000 --> 00:03:27,920 Yeah. 107 00:03:27,920 --> 00:03:30,880 Let's start with Jerry Tang's work at UT Austin. 108 00:03:30,880 --> 00:03:31,720 Okay. 109 00:03:31,720 --> 00:03:33,680 He's using fMRI to decode language, 110 00:03:33,680 --> 00:03:35,720 and the results are pretty astonishing. 111 00:03:35,720 --> 00:03:36,560 Yeah. 112 00:03:36,560 --> 00:03:40,200 His team had participants listen to hours of podcasts 113 00:03:40,200 --> 00:03:42,280 while lying in an fMRI machine. 114 00:03:42,280 --> 00:03:43,920 16 hours, to be exact. 115 00:03:43,920 --> 00:03:45,760 Wow, 16 hours. 116 00:03:45,760 --> 00:03:47,360 Can't imagine lying still for that long. 117 00:03:47,360 --> 00:03:48,480 Oh my gosh, me neither. 118 00:03:48,480 --> 00:03:49,320 Right. 119 00:03:49,320 --> 00:03:51,280 But this massive amount of data was crucial 120 00:03:51,280 --> 00:03:52,840 for training the AI decoder. 121 00:03:52,840 --> 00:03:53,680 Right. 122 00:03:53,680 --> 00:03:56,320 Basically, the AI learned to associate specific patterns 123 00:03:56,320 --> 00:03:59,920 of brain activity with the words and phrases in the podcasts. 124 00:03:59,920 --> 00:04:01,800 It's like showing a child a picture of a dog 125 00:04:01,800 --> 00:04:04,480 and saying dog over and over again. 126 00:04:04,480 --> 00:04:07,320 Eventually, the child learns to associate the image 127 00:04:07,320 --> 00:04:08,280 with the word. 128 00:04:08,280 --> 00:04:10,040 And get this, in one experiment, 129 00:04:10,040 --> 00:04:11,440 they played a sentence for someone 130 00:04:11,440 --> 00:04:12,920 who had gone through the training, 131 00:04:12,920 --> 00:04:15,600 and the decoder was able to paraphrase it. 132 00:04:15,600 --> 00:04:16,440 Wow. 133 00:04:16,440 --> 00:04:19,600 It transformed, we start to trade stories about our lives 134 00:04:19,600 --> 00:04:23,480 into we started talking about our experiences in the area. 135 00:04:23,480 --> 00:04:24,160 Wow. 136 00:04:24,160 --> 00:04:27,440 While not word for word, it captures the underlying meaning, 137 00:04:27,440 --> 00:04:29,600 which is a huge step forward. 138 00:04:29,600 --> 00:04:31,200 Yeah, it's like the decoder's understanding 139 00:04:31,200 --> 00:04:34,240 the gist of the thought, even if it's not getting every word 140 00:04:34,240 --> 00:04:34,960 perfectly. 141 00:04:34,960 --> 00:04:37,520 OK, I know what some listeners might be thinking. 142 00:04:37,520 --> 00:04:40,800 There's no way I could focus for 16 hours straight 143 00:04:40,800 --> 00:04:43,080 while listening to podcasts my mind would wander. 144 00:04:43,080 --> 00:04:43,680 Right. 145 00:04:43,680 --> 00:04:46,200 But Tang's team used engaging narratives, 146 00:04:46,200 --> 00:04:48,160 like audiobooks and radio stories, 147 00:04:48,160 --> 00:04:49,560 to keep people's attention. 148 00:04:49,560 --> 00:04:49,960 OK. 149 00:04:49,960 --> 00:04:53,280 And that huge data set helps filter out 150 00:04:53,280 --> 00:04:55,320 the noise of distractions. 151 00:04:55,320 --> 00:04:58,880 It's like trying to hear a conversation at a bustling cafe. 152 00:04:58,880 --> 00:05:00,440 At first, it's overwhelming. 153 00:05:00,440 --> 00:05:04,040 But as you focus, you can tune out the background noise 154 00:05:04,040 --> 00:05:06,000 and pick up the specific voices you want to hear. 155 00:05:06,000 --> 00:05:08,880 So they're using stories to keep the brain engaged. 156 00:05:08,880 --> 00:05:09,240 Yeah. 157 00:05:09,240 --> 00:05:11,160 And the AI focused on the language. 158 00:05:11,160 --> 00:05:11,880 Yeah, OK. 159 00:05:11,880 --> 00:05:14,680 Hold onto your hats, folks, because it gets even wilder. 160 00:05:14,680 --> 00:05:18,240 Tang's team has managed to decode imagined speech. 161 00:05:18,240 --> 00:05:19,160 What? 162 00:05:19,160 --> 00:05:21,480 They asked people to imagine telling a story 163 00:05:21,480 --> 00:05:23,360 silently in their heads. 164 00:05:23,360 --> 00:05:26,600 And the decoder translated their thoughts into text. 165 00:05:26,600 --> 00:05:27,840 Wow. 166 00:05:27,840 --> 00:05:32,160 In one case, the thought Marco leaned over and whispered 167 00:05:32,160 --> 00:05:36,360 was decoded as he runs up to me and hugs me tight. 168 00:05:36,360 --> 00:05:37,440 Wow, that's incredible. 169 00:05:37,440 --> 00:05:38,760 Again, not perfect. 170 00:05:38,760 --> 00:05:39,200 Right. 171 00:05:39,200 --> 00:05:41,880 But the emotional essence is undeniably there. 172 00:05:41,880 --> 00:05:42,400 Definitely. 173 00:05:42,400 --> 00:05:44,160 Wow, that's incredible. 174 00:05:44,160 --> 00:05:45,480 But I'm curious about something. 175 00:05:45,480 --> 00:05:45,760 Yeah. 176 00:05:45,760 --> 00:05:47,760 When we're talking about decoding meaning rather 177 00:05:47,760 --> 00:05:52,120 than exact words, how do they even measure success? 178 00:05:52,120 --> 00:05:52,520 Yeah. 179 00:05:52,520 --> 00:05:54,000 It seems subjective. 180 00:05:54,000 --> 00:05:55,280 That's a good question. 181 00:05:55,280 --> 00:05:58,120 That's where the power of AI really shines. 182 00:05:58,120 --> 00:05:58,480 OK. 183 00:05:58,480 --> 00:06:02,480 Tang's team uses neural networks to assess semantic similarity. 184 00:06:02,480 --> 00:06:05,280 It's like asking the AI, do these two sentences 185 00:06:05,280 --> 00:06:08,320 convey the same idea even if the wording is different? 186 00:06:08,320 --> 00:06:08,680 Right. 187 00:06:08,680 --> 00:06:10,280 It's not about getting every word right. 188 00:06:10,280 --> 00:06:11,480 It's about capturing the meaning. 189 00:06:11,480 --> 00:06:13,840 So the AI is acting like a judge, 190 00:06:13,840 --> 00:06:15,800 evaluating the meaning behind the words. 191 00:06:15,800 --> 00:06:16,120 Yeah. 192 00:06:16,120 --> 00:06:19,480 But how accurate are these fMRI decoders? 193 00:06:19,480 --> 00:06:23,920 Are we talking reads your mind like an open book accuracy? 194 00:06:23,920 --> 00:06:26,200 Well, it's not quite mind reading in the sense 195 00:06:26,200 --> 00:06:28,640 of knowing your every fleeting thought. 196 00:06:28,640 --> 00:06:29,280 OK. 197 00:06:29,280 --> 00:06:33,360 Remember, this fMRI approach requires extensive training 198 00:06:33,360 --> 00:06:33,760 data. 199 00:06:33,760 --> 00:06:34,120 Right. 200 00:06:34,120 --> 00:06:36,480 So it's not like you could just hop into an fMRI machine 201 00:06:36,480 --> 00:06:38,320 and have your thoughts decoded on the spot. 202 00:06:38,320 --> 00:06:38,720 Right. 203 00:06:38,720 --> 00:06:38,960 Right. 204 00:06:38,960 --> 00:06:41,280 Plus, the technology is still under development. 205 00:06:41,280 --> 00:06:41,720 Oh. 206 00:06:41,720 --> 00:06:43,160 But it's improving rapidly. 207 00:06:43,160 --> 00:06:43,480 OK. 208 00:06:43,480 --> 00:06:48,160 So while impressive, this means creating a decoder for you 209 00:06:48,160 --> 00:06:51,880 would require you to spend 16 hours in an fMRI machine 210 00:06:51,880 --> 00:06:53,160 listening to podcasts. 211 00:06:53,160 --> 00:06:53,720 Yeah. 212 00:06:53,720 --> 00:06:56,080 Not exactly practical for everyday use. 213 00:06:56,080 --> 00:06:56,440 Right. 214 00:06:56,440 --> 00:06:59,400 But Jerry's research highlights the incredible detail 215 00:06:59,400 --> 00:07:00,960 we can get from fMRI. 216 00:07:00,960 --> 00:07:01,480 Yeah. 217 00:07:01,480 --> 00:07:03,680 But what if we need a more portable way 218 00:07:03,680 --> 00:07:05,040 to capture brain activity? 219 00:07:05,040 --> 00:07:05,560 Right. 220 00:07:05,560 --> 00:07:08,520 That's where Michael Blumenstein's work with EEG comes in. 221 00:07:08,520 --> 00:07:09,240 Oh, yeah. 222 00:07:09,240 --> 00:07:10,720 He's achieving some amazing results 223 00:07:10,720 --> 00:07:12,480 using a completely different approach. 224 00:07:12,480 --> 00:07:16,240 Instead of relying on bulky fMRI machines, 225 00:07:16,240 --> 00:07:19,400 Michael is using EEG, which is much more portable. 226 00:07:19,400 --> 00:07:21,320 It's like trading in a grand piano 227 00:07:21,320 --> 00:07:24,400 for a sleek, portable synthesizer. 228 00:07:24,400 --> 00:07:27,480 You might lose some of the richness and complexity, 229 00:07:27,480 --> 00:07:29,800 but you gain flexibility and accessibility. 230 00:07:29,800 --> 00:07:33,240 So EEG is like the on-the-go version of brain scanning. 231 00:07:33,240 --> 00:07:34,000 Exactly. 232 00:07:34,000 --> 00:07:35,600 But wouldn't that sacrifice accuracy? 233 00:07:35,600 --> 00:07:38,800 I mean, fMRI gives you such detailed images 234 00:07:38,800 --> 00:07:40,080 of brain activity. 235 00:07:40,080 --> 00:07:44,360 You're right, that EEG lacks the spatial resolution of fMRI. 236 00:07:44,360 --> 00:07:44,760 Right. 237 00:07:44,760 --> 00:07:47,760 We can't pinpoint exactly which brain regions are 238 00:07:47,760 --> 00:07:49,120 firing with the same precision. 239 00:07:49,120 --> 00:07:49,840 Yeah. 240 00:07:49,840 --> 00:07:54,040 But EEG excels at capturing the temporal details, 241 00:07:54,040 --> 00:07:56,280 the precise timing of brain signals. 242 00:07:56,280 --> 00:07:56,880 Got it. 243 00:07:56,880 --> 00:07:58,480 It's like watching a high-speed video 244 00:07:58,480 --> 00:08:00,080 of the brain's electrical symphony. 245 00:08:00,080 --> 00:08:00,600 That makes sense. 246 00:08:00,600 --> 00:08:02,200 Timing is everything in language. 247 00:08:02,200 --> 00:08:02,520 Right. 248 00:08:02,520 --> 00:08:04,720 So how does Blumenstein's method actually work? 249 00:08:04,720 --> 00:08:05,120 Right. 250 00:08:05,120 --> 00:08:08,280 Does he also have people listen to podcasts for hours on end? 251 00:08:08,280 --> 00:08:09,160 Not quite. 252 00:08:09,160 --> 00:08:12,480 His participants read text aloud while wearing an EEG cap. 253 00:08:12,480 --> 00:08:12,960 OK. 254 00:08:12,960 --> 00:08:15,440 And the AI analyzes those EEG signals, 255 00:08:15,440 --> 00:08:17,760 breaking them down into components associated 256 00:08:17,760 --> 00:08:18,920 with individual words. 257 00:08:18,920 --> 00:08:22,040 Wait, so the AI figures out which brainwave patterns correspond 258 00:08:22,040 --> 00:08:23,320 to specific words? 259 00:08:23,320 --> 00:08:25,000 How is that even possible? 260 00:08:25,000 --> 00:08:28,240 It's a brilliant technique called self-supervised learning. 261 00:08:28,240 --> 00:08:30,960 Think of it like teaching a dog a new trick 262 00:08:30,960 --> 00:08:33,560 without explicitly saying the command. 263 00:08:33,560 --> 00:08:34,160 OK. 264 00:08:34,160 --> 00:08:36,240 You reward them when they do the right action 265 00:08:36,240 --> 00:08:37,800 and they figure out the connection themselves. 266 00:08:37,800 --> 00:08:38,560 Right. 267 00:08:38,560 --> 00:08:40,720 Similarly, the AI isn't explicitly 268 00:08:40,720 --> 00:08:43,800 told what each word looks like in terms of brain activity. 269 00:08:43,800 --> 00:08:48,080 It figures it out on its own by analyzing patterns in the data. 270 00:08:48,080 --> 00:08:50,160 So it's like a super-powered code breaker 271 00:08:50,160 --> 00:08:51,800 cracking the neural code of language. 272 00:08:51,800 --> 00:08:52,280 Yeah. 273 00:08:52,280 --> 00:08:54,480 And once it's figured out those word-like units, 274 00:08:54,480 --> 00:08:55,560 what happens next? 275 00:08:55,560 --> 00:08:57,880 This is where things get even more interesting. 276 00:08:57,880 --> 00:09:01,520 A large language model, similar to chatGPT, 277 00:09:01,520 --> 00:09:05,120 stepped in to interpret these thought tokens 278 00:09:05,120 --> 00:09:07,560 and stitched them together into coherent sentences. 279 00:09:07,560 --> 00:09:08,960 So it's a two-step process. 280 00:09:08,960 --> 00:09:11,680 First, the AI breaks down the brain signals 281 00:09:11,680 --> 00:09:13,400 into word-like units. 282 00:09:13,400 --> 00:09:16,640 And then the language model uses its vast knowledge 283 00:09:16,640 --> 00:09:19,040 of language to put those units together in a way that 284 00:09:19,040 --> 00:09:19,520 makes sense. 285 00:09:19,520 --> 00:09:20,200 Precisely. 286 00:09:20,200 --> 00:09:20,520 Yeah. 287 00:09:20,520 --> 00:09:22,960 And what's remarkable is that this approach doesn't 288 00:09:22,960 --> 00:09:25,960 require extensive training data from each individual, 289 00:09:25,960 --> 00:09:28,120 unlike Tang's fMRI method. 290 00:09:28,120 --> 00:09:31,360 That's because of those pre-trained language models, 291 00:09:31,360 --> 00:09:34,120 which have already learned the statistical relationships 292 00:09:34,120 --> 00:09:37,560 between words and concepts from analyzing massive amounts 293 00:09:37,560 --> 00:09:38,600 of text. 294 00:09:38,600 --> 00:09:43,000 So how accurate is this EEG-based approach? 295 00:09:43,000 --> 00:09:46,400 Are we talking mind-reading party trick levels of accuracy 296 00:09:46,400 --> 00:09:47,920 or something more nuanced? 297 00:09:47,920 --> 00:09:49,600 Well, it's still early days. 298 00:09:49,600 --> 00:09:50,160 OK. 299 00:09:50,160 --> 00:09:52,160 But the results are promising. 300 00:09:52,160 --> 00:09:53,000 OK. 301 00:09:53,000 --> 00:09:57,360 Blumenstein's team is reporting accuracy above 60%, 302 00:09:57,360 --> 00:10:00,440 which is significant considering the complexity of the task. 303 00:10:00,440 --> 00:10:02,920 And do they have any examples of what the decoder has been able 304 00:10:02,920 --> 00:10:04,640 to, well, here, so to speak? 305 00:10:04,640 --> 00:10:05,280 Oh, absolutely. 306 00:10:05,280 --> 00:10:05,780 OK. 307 00:10:05,780 --> 00:10:09,200 In one instance, the participant was thinking the phrase, 308 00:10:09,200 --> 00:10:10,160 the hat on the sea. 309 00:10:10,160 --> 00:10:11,000 OK. 310 00:10:11,000 --> 00:10:13,720 The decoder, bless its heart, interpreted this as, 311 00:10:13,720 --> 00:10:15,560 I'd like a bowl of chicken soup, please. 312 00:10:15,560 --> 00:10:17,040 That's hilarious. 313 00:10:17,040 --> 00:10:19,680 So it might not be getting every word perfect, 314 00:10:19,680 --> 00:10:22,120 but it's definitely capturing the essence of the thought. 315 00:10:22,120 --> 00:10:23,000 Exactly. 316 00:10:23,000 --> 00:10:24,640 Remember, this is just the beginning. 317 00:10:24,640 --> 00:10:25,040 Right. 318 00:10:25,040 --> 00:10:28,040 As the technology evolves, the accuracy 319 00:10:28,040 --> 00:10:29,640 is only going to improve. 320 00:10:29,640 --> 00:10:31,560 Which is pretty mind-blowing, but it also brings up 321 00:10:31,560 --> 00:10:32,760 some serious questions. 322 00:10:32,760 --> 00:10:33,280 Right. 323 00:10:33,280 --> 00:10:36,680 If we can already decode thoughts with this level 324 00:10:36,680 --> 00:10:39,280 of accuracy, what does the future hold? 325 00:10:39,280 --> 00:10:39,600 Yeah. 326 00:10:39,600 --> 00:10:43,160 It's like we're on the verge of a whole new era of communication 327 00:10:43,160 --> 00:10:44,200 and understanding. 328 00:10:44,200 --> 00:10:44,840 Right. 329 00:10:44,840 --> 00:10:47,360 But before we get too carried away, 330 00:10:47,360 --> 00:10:49,440 I think it's time to bring our experts back 331 00:10:49,440 --> 00:10:50,440 into the conversation. 332 00:10:50,440 --> 00:10:50,840 Right. 333 00:10:50,840 --> 00:10:53,040 We've got Jerry Tang, who's pioneering 334 00:10:53,040 --> 00:10:56,800 FMRI-based thought decoding, and Michael Blumenstein, who's 335 00:10:56,800 --> 00:10:59,320 pushing the boundaries of EEG. 336 00:10:59,320 --> 00:10:59,760 Right. 337 00:10:59,760 --> 00:11:01,880 Welcome back to the show, gentlemen. 338 00:11:01,880 --> 00:11:03,800 It's fascinating to see how both of you 339 00:11:03,800 --> 00:11:06,840 are approaching this challenge from different angles. 340 00:11:06,840 --> 00:11:07,360 Yeah. 341 00:11:07,360 --> 00:11:09,280 It's like having two master chefs trying 342 00:11:09,280 --> 00:11:12,120 to perfect the same dish, but with completely different 343 00:11:12,120 --> 00:11:13,000 cooking styles. 344 00:11:13,000 --> 00:11:13,720 Exactly. 345 00:11:13,720 --> 00:11:17,280 And speaking of cooking styles, Jerry, your FMRI research 346 00:11:17,280 --> 00:11:19,200 relies on those detailed brain scans 347 00:11:19,200 --> 00:11:22,000 to pinpoint where language is processed. 348 00:11:22,000 --> 00:11:22,600 Right. 349 00:11:22,600 --> 00:11:24,800 But Michael's EEG approach focuses 350 00:11:24,800 --> 00:11:27,640 on the timing of brain signals. 351 00:11:27,640 --> 00:11:29,040 What do these differences tell us 352 00:11:29,040 --> 00:11:31,400 about how language works in the brain? 353 00:11:31,400 --> 00:11:32,280 That's a great question. 354 00:11:32,280 --> 00:11:32,680 Yeah. 355 00:11:32,680 --> 00:11:35,160 One of the most exciting things about this research 356 00:11:35,160 --> 00:11:37,880 is that it's not just about building cool technology. 357 00:11:37,880 --> 00:11:40,120 It's about understanding the human brain. 358 00:11:40,120 --> 00:11:40,920 Yeah. 359 00:11:40,920 --> 00:11:45,480 FMRI allows us to see which brain regions are involved 360 00:11:45,480 --> 00:11:47,680 in different aspects of language processing. 361 00:11:47,680 --> 00:11:49,640 So it's not like there's one word 362 00:11:49,640 --> 00:11:51,120 center in the brain. 363 00:11:51,120 --> 00:11:51,560 Right. 364 00:11:51,560 --> 00:11:55,120 But rather a whole network of regions working together. 365 00:11:55,120 --> 00:11:55,880 Exactly. 366 00:11:55,880 --> 00:11:57,720 And the way these regions interact, 367 00:11:57,720 --> 00:12:01,120 the timing of their activation, it's all incredibly intricate. 368 00:12:01,120 --> 00:12:01,640 Yeah. 369 00:12:01,640 --> 00:12:03,360 Our research is helping us map out 370 00:12:03,360 --> 00:12:05,760 this neural symphony of language. 371 00:12:05,760 --> 00:12:08,720 We're seeing how a single word can activate a complex chain 372 00:12:08,720 --> 00:12:10,360 reaction across the brain. 373 00:12:10,360 --> 00:12:12,440 It's like watching a fireworks display. 374 00:12:12,440 --> 00:12:12,720 Yes. 375 00:12:12,720 --> 00:12:15,200 With different parts of the brain lighting up in sequence. 376 00:12:15,200 --> 00:12:16,640 That's a great analogy. 377 00:12:16,640 --> 00:12:19,720 Michael, what's your take on this from the EEG perspective? 378 00:12:19,720 --> 00:12:21,240 I completely agree with Jerry. 379 00:12:21,240 --> 00:12:23,320 The beauty of EEG is that it captures 380 00:12:23,320 --> 00:12:25,520 the temporal dynamics of brain activity 381 00:12:25,520 --> 00:12:27,240 with millisecond precision. 382 00:12:27,240 --> 00:12:28,440 OK. 383 00:12:28,440 --> 00:12:30,840 So while we may not have the same spatial detail 384 00:12:30,840 --> 00:12:34,320 as fMRI, we can see how brain waves dance and evolve 385 00:12:34,320 --> 00:12:35,520 as a thought is formed. 386 00:12:35,520 --> 00:12:39,040 It's like watching a high speed video of the brain in action. 387 00:12:39,040 --> 00:12:41,800 So fMRI gives us the where. 388 00:12:41,800 --> 00:12:42,000 Yeah. 389 00:12:42,000 --> 00:12:45,400 And EEG gives us the when of language processing. 390 00:12:45,400 --> 00:12:46,000 Exactly. 391 00:12:46,000 --> 00:12:49,280 And together, they're painting a much richer picture 392 00:12:49,280 --> 00:12:51,480 of this incredibly complex process. 393 00:12:51,480 --> 00:12:53,840 And here's where things get really interesting. 394 00:12:53,840 --> 00:12:54,480 OK. 395 00:12:54,480 --> 00:12:57,560 Because if we think about the future of thought decoding, 396 00:12:57,560 --> 00:13:00,080 could we one day combine these approaches? 397 00:13:00,080 --> 00:13:01,120 Oh, wow. 398 00:13:01,120 --> 00:13:04,160 Imagine an AI system that has both the spatial precision 399 00:13:04,160 --> 00:13:09,200 of fMRI and EE, the temporal resolution of EEG. 400 00:13:09,200 --> 00:13:11,400 Talk about a mind reading powerhouse. 401 00:13:11,400 --> 00:13:12,480 I know, right. 402 00:13:12,480 --> 00:13:14,440 But Michael, you mentioned earlier 403 00:13:14,440 --> 00:13:16,960 that your EEG based approach doesn't 404 00:13:16,960 --> 00:13:19,960 rely on individualized training data 405 00:13:19,960 --> 00:13:22,720 thanks to those powerful large language models. 406 00:13:22,720 --> 00:13:25,920 Does that mean we could one day have universal thought 407 00:13:25,920 --> 00:13:29,160 decoders that work right out of the box 408 00:13:29,160 --> 00:13:31,880 without any need for personalized calibration? 409 00:13:31,880 --> 00:13:34,000 That's a very exciting possibility and something 410 00:13:34,000 --> 00:13:35,800 we're actively exploring. 411 00:13:35,800 --> 00:13:38,320 These large language models are incredibly versatile. 412 00:13:38,320 --> 00:13:40,600 As they become more sophisticated, 413 00:13:40,600 --> 00:13:43,920 I wouldn't rule out the possibility of plug and play 414 00:13:43,920 --> 00:13:46,880 thought decoders that can tap into the commonalities 415 00:13:46,880 --> 00:13:48,360 of human language processing. 416 00:13:48,360 --> 00:13:52,080 So instead of needing 16 hours of training data, 417 00:13:52,080 --> 00:13:55,080 we could have decoders that work for anyone right off the bat. 418 00:13:55,080 --> 00:13:56,280 That's mind blowing. 419 00:13:56,280 --> 00:13:57,320 It is pretty amazing. 420 00:13:57,320 --> 00:14:01,440 Jerry, what are your thoughts on this universal decoder idea? 421 00:14:01,440 --> 00:14:04,480 Is it something you see as feasible down the line? 422 00:14:04,480 --> 00:14:06,040 Or are we still a long way off? 423 00:14:06,040 --> 00:14:07,840 I'm cautiously optimistic. 424 00:14:07,840 --> 00:14:10,880 We're already seeing hints that certain aspects of language 425 00:14:10,880 --> 00:14:13,440 processing are shared across individuals. 426 00:14:13,440 --> 00:14:17,200 And these large language models are amazing at finding patterns 427 00:14:17,200 --> 00:14:19,200 that humans might miss. 428 00:14:19,200 --> 00:14:22,400 So while I wouldn't say we're there yet, 429 00:14:22,400 --> 00:14:24,240 the idea of a universal thought decoder 430 00:14:24,240 --> 00:14:26,360 is definitely within the realm of possibility. 431 00:14:26,360 --> 00:14:28,800 And just think about the potential benefits. 432 00:14:28,800 --> 00:14:31,320 If we could develop a technology that accurately 433 00:14:31,320 --> 00:14:33,480 translates thoughts into language, 434 00:14:33,480 --> 00:14:36,480 regardless of individual differences, 435 00:14:36,480 --> 00:14:38,800 it would revolutionize communication, 436 00:14:38,800 --> 00:14:41,560 especially for those who have lost their ability to speak. 437 00:14:41,560 --> 00:14:43,760 Imagine being able to give a voice 438 00:14:43,760 --> 00:14:47,720 to people who have been silenced by stroke or paralysis 439 00:14:47,720 --> 00:14:49,360 or other conditions. 440 00:14:49,360 --> 00:14:50,920 It's like something out of a sci-fi movie, 441 00:14:50,920 --> 00:14:52,320 but it could become a reality. 442 00:14:52,320 --> 00:14:53,000 Absolutely. 443 00:14:53,000 --> 00:14:55,240 And that's why it's so important to have conversations 444 00:14:55,240 --> 00:14:58,280 about the ethical implications of this technology. 445 00:14:58,280 --> 00:15:00,800 We need to make sure it's used for good, 446 00:15:00,800 --> 00:15:03,320 and that we're prepared for the potential challenges. 447 00:15:03,320 --> 00:15:07,720 Because, let's be honest, the idea of AI reading our minds 448 00:15:07,720 --> 00:15:09,040 does raise some concerns. 449 00:15:09,040 --> 00:15:09,640 Of course. 450 00:15:09,640 --> 00:15:12,160 Makes you wonder, where do we draw the line? 451 00:15:12,160 --> 00:15:14,440 When does this incredible technology 452 00:15:14,440 --> 00:15:17,200 start to feel, well, a little creepy? 453 00:15:17,200 --> 00:15:18,840 That's the million dollar question, isn't it? 454 00:15:18,840 --> 00:15:22,440 We're essentially venturing into uncharted territory. 455 00:15:22,440 --> 00:15:26,400 Jerry, your team seems very aware of these concerns. 456 00:15:26,400 --> 00:15:30,080 Can you give us some insight into how you're approaching 457 00:15:30,080 --> 00:15:31,880 the ethical side of this research? 458 00:15:31,880 --> 00:15:33,080 Absolutely. 459 00:15:33,080 --> 00:15:36,080 First and foremost, we believe that no one's brain should 460 00:15:36,080 --> 00:15:38,640 be decoded without their informed consent. 461 00:15:38,640 --> 00:15:39,240 OK. 462 00:15:39,240 --> 00:15:41,040 We're very upfront with our participants 463 00:15:41,040 --> 00:15:43,840 about the risks and benefits of this research. 464 00:15:43,840 --> 00:15:46,520 We want them to feel empowered and in control 465 00:15:46,520 --> 00:15:48,680 throughout the entire process. 466 00:15:48,680 --> 00:15:51,320 So informed consent is crucial. 467 00:15:51,320 --> 00:15:52,960 But what about the bigger picture? 468 00:15:52,960 --> 00:15:55,720 What happens if this technology falls into the wrong hands? 469 00:15:55,720 --> 00:15:59,160 Could it be used for surveillance or even manipulation? 470 00:15:59,160 --> 00:16:00,680 Those are valid concerns, and they're 471 00:16:00,680 --> 00:16:02,640 ones we take very seriously. 472 00:16:02,640 --> 00:16:05,120 As scientists, we have a responsibility 473 00:16:05,120 --> 00:16:07,680 to consider the potential consequences of our work, 474 00:16:07,680 --> 00:16:10,000 both intended and unintended. 475 00:16:10,000 --> 00:16:12,000 That's why we're actively engaging 476 00:16:12,000 --> 00:16:14,720 with ethicists, policymakers, and the public 477 00:16:14,720 --> 00:16:16,960 to ensure that these technologies are developed 478 00:16:16,960 --> 00:16:18,400 and used responsibly. 479 00:16:18,400 --> 00:16:20,400 It's almost like we need a new set of rules, 480 00:16:20,400 --> 00:16:23,640 a mental bill of rights, to protect our inner thoughts 481 00:16:23,640 --> 00:16:24,920 from unwanted intrusion. 482 00:16:24,920 --> 00:16:26,040 I completely agree. 483 00:16:26,040 --> 00:16:28,480 We need to start having these conversations now 484 00:16:28,480 --> 00:16:30,480 before these technologies become so advanced 485 00:16:30,480 --> 00:16:32,400 that it's too late to course correct. 486 00:16:32,400 --> 00:16:35,520 We need to establish clear boundaries and ethical guidelines. 487 00:16:35,520 --> 00:16:37,800 It's like we're standing at a fork in the road. 488 00:16:37,800 --> 00:16:41,720 This technology has the potential to do so much good. 489 00:16:41,720 --> 00:16:44,240 But we need to make sure we're heading in the right direction. 490 00:16:44,240 --> 00:16:45,200 Absolutely. 491 00:16:45,200 --> 00:16:48,120 And those conversations need to involve everyone, 492 00:16:48,120 --> 00:16:50,280 not just the scientists and engineers. 493 00:16:50,280 --> 00:16:52,920 We need to bring in ethicists, philosophers, 494 00:16:52,920 --> 00:16:56,000 legal experts, and, most importantly, 495 00:16:56,000 --> 00:16:59,480 the people who will be most affected by these advancements. 496 00:16:59,480 --> 00:17:01,000 What do you think, Michael? 497 00:17:01,000 --> 00:17:02,720 I think it's crucial to approach this 498 00:17:02,720 --> 00:17:04,800 with a sense of humility. 499 00:17:04,800 --> 00:17:06,400 We're dealing with the human mind, 500 00:17:06,400 --> 00:17:10,600 the most complex and mysterious entity in the known universe. 501 00:17:10,600 --> 00:17:13,440 We need to proceed with caution, respect, 502 00:17:13,440 --> 00:17:16,360 and a deep understanding of the potential consequences, 503 00:17:16,360 --> 00:17:18,200 both positive and negative. 504 00:17:18,200 --> 00:17:19,720 Well said, Michael. 505 00:17:19,720 --> 00:17:22,440 And that's a perfect segue into our final segment 506 00:17:22,440 --> 00:17:24,920 where we'll explore some of the most thought-provoking questions 507 00:17:24,920 --> 00:17:26,120 raised by this research. 508 00:17:26,120 --> 00:17:26,720 OK. 509 00:17:26,720 --> 00:17:30,080 Are we ready to enter a world where our thoughts are no longer our own? 510 00:17:30,080 --> 00:17:33,160 It's a question that's both exciting and terrifying. 511 00:17:33,160 --> 00:17:35,040 But one thing's for sure. 512 00:17:35,040 --> 00:17:36,640 We can't just bury our heads in the sand 513 00:17:36,640 --> 00:17:38,200 and pretend this isn't happening. 514 00:17:38,200 --> 00:17:41,800 The future of thought-decoding is unfolding before our very eyes. 515 00:17:41,800 --> 00:17:45,600 And it's up to all of us to shape that future responsibly and ethically. 516 00:17:45,600 --> 00:17:49,520 And that, my friends, is something to ponder 517 00:17:49,520 --> 00:17:52,480 as we venture deeper into the mysteries of the mind. 518 00:17:52,480 --> 00:17:53,000 OK. 519 00:17:53,000 --> 00:17:56,480 We've heard from the experts about the mind-blowing science 520 00:17:56,480 --> 00:17:58,960 behind thought-decoding and explored 521 00:17:58,960 --> 00:18:02,040 both the incredible potential and the ethical concerns. 522 00:18:02,040 --> 00:18:02,440 Yeah. 523 00:18:02,440 --> 00:18:05,000 But now, let's get philosophical for a moment. 524 00:18:05,000 --> 00:18:08,560 This isn't just about cool gadgets and scientific breakthroughs. 525 00:18:08,560 --> 00:18:10,760 It's about what it means to be human. 526 00:18:10,760 --> 00:18:11,520 Absolutely. 527 00:18:11,520 --> 00:18:12,560 Exactly. 528 00:18:12,560 --> 00:18:17,360 If our thoughts can be decoded, does that change how we think about privacy? 529 00:18:17,360 --> 00:18:20,640 We're already living in a world where our online activity is tracked, 530 00:18:20,640 --> 00:18:23,600 our conversations are recorded, and our faces are stand. 531 00:18:23,600 --> 00:18:24,000 Right. 532 00:18:24,000 --> 00:18:26,640 Is mental privacy the next thing we have to worry about? 533 00:18:26,640 --> 00:18:27,720 It's a valid question. 534 00:18:27,720 --> 00:18:33,840 And it really makes you think, are our thoughts the last truly private space we have left? 535 00:18:33,840 --> 00:18:35,480 And what about the legal side of things? 536 00:18:35,480 --> 00:18:38,360 Could thought-decoding be used in court? 537 00:18:38,360 --> 00:18:42,440 Imagine someone being convicted of a crime based on their thoughts alone. 538 00:18:42,440 --> 00:18:47,840 That raises some serious questions about due process and the presumption of innocence. 539 00:18:47,840 --> 00:18:53,040 We need to think carefully about the legal framework surrounding this technology. 540 00:18:53,040 --> 00:18:54,960 And then there's the question of free will. 541 00:18:54,960 --> 00:18:55,360 Oh, yeah. 542 00:18:55,360 --> 00:19:00,480 If our thoughts can be predicted and decoded, does that mean our actions are predetermined? 543 00:19:00,480 --> 00:19:01,880 Are we really calling the shots? 544 00:19:01,880 --> 00:19:04,320 Are we just following a script written by our brains? 545 00:19:04,320 --> 00:19:07,120 Now we're getting into deep philosophical territory. 546 00:19:07,120 --> 00:19:12,760 This research challenges our understanding of consciousness and autonomy. 547 00:19:12,760 --> 00:19:13,120 Yeah. 548 00:19:13,120 --> 00:19:17,360 If our thoughts can be read, does that diminish our sense of self? 549 00:19:17,360 --> 00:19:21,920 It's almost like we're back to that Charlotte Bronte quote about the mind being a sacred sanctuary. 550 00:19:21,920 --> 00:19:22,480 Right. 551 00:19:22,480 --> 00:19:29,480 Are we on the verge of losing that inner sanctuary, that sense of privacy and control over our own thoughts? 552 00:19:29,480 --> 00:19:33,720 It's a question we need to grapple with individually and as a society. 553 00:19:33,720 --> 00:19:37,040 But I think it's important to remember that technology is a tool. 554 00:19:37,040 --> 00:19:37,720 Right. 555 00:19:37,720 --> 00:19:39,600 It can be used for good or for ill. 556 00:19:39,600 --> 00:19:40,080 Yeah. 557 00:19:40,080 --> 00:19:43,480 It's up to us to decide how we want to use this powerful new tool. 558 00:19:43,480 --> 00:19:46,040 And that decision shouldn't be taken lightly. 559 00:19:46,040 --> 00:19:52,120 We need to have open and honest conversations about the implications of thought decoding technology. 560 00:19:52,120 --> 00:19:54,760 And make sure we're developing it responsibly and ethically. 561 00:19:54,760 --> 00:19:55,480 Absolutely. 562 00:19:55,480 --> 00:20:00,280 So as we wrap up this deep dive, we leave you with this final thought. 563 00:20:00,280 --> 00:20:05,720 What would a world where our thoughts are no longer entirely private look like? 564 00:20:05,720 --> 00:20:13,960 Would it be a utopia of understanding and empathy or a dystopia of control and manipulation? 565 00:20:13,960 --> 00:20:16,000 The answer, as always, is up to us. 566 00:20:16,000 --> 00:20:17,160 That's our show for today. 567 00:20:17,160 --> 00:20:19,560 We hope you enjoyed this exploration of thought decoding. 568 00:20:19,560 --> 00:20:23,400 It's a topic that's sure to keep us thinking and talking for years to come. 569 00:20:23,400 --> 00:20:29,440 If you want to learn more, check out the links in the show notes and share your thoughts on this mind-blowing topic. 570 00:20:29,440 --> 00:20:30,480 Thanks for listening. 571 00:20:30,480 --> 00:20:34,520 It's like we're on the verge of a whole new era of communication and understanding. 572 00:20:34,520 --> 00:20:38,920 But before we get too carried away, I think it's time to bring our experts back into the conversation. 573 00:20:38,920 --> 00:20:43,000 We've got Jerry Tang, who's pioneering FMRI-based thought decoding, 574 00:20:43,000 --> 00:20:46,360 and Michael Blumenstein, who's pushing the boundaries of EG. 575 00:20:46,360 --> 00:20:47,840 Welcome back to the show, gentlemen. 576 00:20:47,840 --> 00:20:51,840 It's fascinating to see how both of you are approaching this challenge from different angles. 577 00:20:51,840 --> 00:20:52,280 Yeah. 578 00:20:52,280 --> 00:20:58,720 It's like having two master chefs trying to perfect the same dish, but with completely different cooking styles. 579 00:20:58,720 --> 00:20:59,400 Exactly. 580 00:20:59,400 --> 00:21:01,600 And speaking of cooking styles, Jerry, 581 00:21:01,600 --> 00:21:05,880 your FMRI research relies on those detailed brain scans, right? 582 00:21:05,880 --> 00:21:06,440 Right. 583 00:21:06,440 --> 00:21:09,280 To pinpoint where language is processed. 584 00:21:09,280 --> 00:21:15,240 But Michael's EEG approach focuses on the timing of brain signals. 585 00:21:15,240 --> 00:21:19,520 What do these differences tell us about how language works in the brain? 586 00:21:19,520 --> 00:21:20,840 That's a great question. 587 00:21:20,840 --> 00:21:25,440 One of the most exciting things about this research is that it's not just about building cool technology. 588 00:21:25,440 --> 00:21:27,600 It's about understanding the human brain. 589 00:21:27,600 --> 00:21:32,800 FMRI allows us to see which brain regions are involved in different aspects of language processing. 590 00:21:32,800 --> 00:21:36,000 So it's not like there's one word center in the brain, 591 00:21:36,000 --> 00:21:39,200 but rather a whole network of regions working together. 592 00:21:39,200 --> 00:21:39,680 Exactly. 593 00:21:39,680 --> 00:21:44,520 And the way these regions interact, the timing of their activation, it's all incredibly intricate. 594 00:21:44,520 --> 00:21:48,000 Our research is helping us map out this neural symphony of language. 595 00:21:48,000 --> 00:21:52,760 We're seeing how a single word can activate a complex chain reaction across the brain. 596 00:21:52,760 --> 00:21:53,360 Wow. 597 00:21:53,360 --> 00:21:58,560 It's like watching a fireworks display with different parts of the brain lighting up in sequence. 598 00:21:58,560 --> 00:22:01,000 Michael, what's your take on this from the EEG perspective? 599 00:22:01,000 --> 00:22:02,400 I completely agree with Jerry. 600 00:22:02,400 --> 00:22:08,640 The beauty of EEG is that it captures the temporal dynamics of brain activity with millisecond precision. 601 00:22:08,640 --> 00:22:12,000 So while we may not have the same spatial detail as FMRI, 602 00:22:12,000 --> 00:22:16,240 we can see how brain waves dance and evolve as a thought is formed. 603 00:22:16,240 --> 00:22:19,280 It's like watching a high speed video of the brain in action. 604 00:22:19,280 --> 00:22:24,000 So FMRI gives us the where and EEG gives us the when of language processing. 605 00:22:24,000 --> 00:22:28,720 And together, they're painting a much richer picture of this incredibly complex process. 606 00:22:28,720 --> 00:22:31,600 And here's where things get really interesting. 607 00:22:31,600 --> 00:22:34,200 Because if we think about the future of thought decoding, 608 00:22:34,200 --> 00:22:36,640 could we one day combine these approaches? 609 00:22:36,640 --> 00:22:40,840 Imagine an AI system that has both the spatial precision of FMRI 610 00:22:40,840 --> 00:22:43,200 and the temporal resolution of EEG. 611 00:22:43,200 --> 00:22:45,480 Talk about a mind reading powerhouse. 612 00:22:45,480 --> 00:22:52,040 But Michael, you mentioned earlier that your EEG based approach doesn't rely on individualized training data 613 00:22:52,040 --> 00:22:55,200 thanks to those powerful large language models. 614 00:22:55,200 --> 00:22:59,920 Does that mean we could one day have universal thought decoders that work right out of the box 615 00:22:59,920 --> 00:23:02,720 without any need for personalized calibration? 616 00:23:02,720 --> 00:23:06,320 That's a very exciting possibility and something we're actively exploring. 617 00:23:06,320 --> 00:23:09,960 These large language models are incredibly versatile. 618 00:23:09,960 --> 00:23:15,800 As they become more sophisticated, I wouldn't rule out the possibility of plug and play thought decoders 619 00:23:15,800 --> 00:23:19,040 that can tap into the commonalities of human language processing. 620 00:23:19,040 --> 00:23:21,440 So instead of needing 16 hours of training data, 621 00:23:21,440 --> 00:23:24,320 we could have decoders that work for anyone right off the bat. 622 00:23:24,320 --> 00:23:25,920 That's mind blowing. 623 00:23:25,920 --> 00:23:29,000 Jerry, what are your thoughts on this universal decoder idea? 624 00:23:29,000 --> 00:23:33,160 Is it something you see as feasible down the line or are we still a long way off? 625 00:23:33,160 --> 00:23:35,320 I'm cautiously optimistic. 626 00:23:35,320 --> 00:23:39,200 We're already seeing hints that certain aspects of language processing 627 00:23:39,200 --> 00:23:41,400 are shared across individuals. 628 00:23:41,400 --> 00:23:47,360 And these large language models are amazing at finding patterns that humans might miss. 629 00:23:47,360 --> 00:23:49,800 So while I wouldn't say we're there yet, 630 00:23:49,800 --> 00:23:54,280 the idea of a universal thought decoder is definitely within the realm of possibility. 631 00:23:54,280 --> 00:23:56,600 And just think about the potential benefits. 632 00:23:56,600 --> 00:24:01,520 If we can develop a technology that accurately translates thoughts into language, 633 00:24:01,520 --> 00:24:03,800 regardless of individual differences, 634 00:24:03,800 --> 00:24:05,760 it would revolutionize communication, 635 00:24:05,760 --> 00:24:08,640 especially for those who have lost their ability to speak. 636 00:24:08,640 --> 00:24:12,480 Imagine being able to give a voice to people who have been silenced by stroke, 637 00:24:12,480 --> 00:24:14,440 paralysis, or other conditions. 638 00:24:14,440 --> 00:24:17,440 It's like something out of a sci-fi movie, but it could become a reality. 639 00:24:17,440 --> 00:24:18,120 Absolutely. 640 00:24:18,120 --> 00:24:22,840 And that's why it's so important to have conversations about the ethical implications of this technology. 641 00:24:22,840 --> 00:24:27,000 We need to make sure it's used for good and that we're prepared for the potential challenges. 642 00:24:27,000 --> 00:24:31,600 Because let's be honest, the idea of AI reading our minds does raise some concerns. 643 00:24:31,600 --> 00:24:34,120 It makes you wonder, where do we draw the line? 644 00:24:34,120 --> 00:24:38,240 When does this incredible technology start to feel, well, a little creepy? 645 00:24:38,240 --> 00:24:40,040 That's the million-dollar question, isn't it? 646 00:24:40,040 --> 00:24:42,800 We're essentially venturing into uncharted territory. 647 00:24:42,800 --> 00:24:45,960 Jerry, your team seems very aware of these concerns. 648 00:24:45,960 --> 00:24:50,680 Can you give us some insight into how you're approaching the ethical side of this research? 649 00:24:50,680 --> 00:24:52,200 Absolutely. 650 00:24:52,200 --> 00:24:57,480 First and foremost, we believe that no one's brain should be decoded without their informed consent. 651 00:24:57,480 --> 00:25:01,880 We're very upfront with our participants about the risks and benefits of this research. 652 00:25:01,880 --> 00:25:06,520 We want them to feel empowered and in control throughout the entire process. 653 00:25:06,520 --> 00:25:08,960 So informed consent is crucial. 654 00:25:08,960 --> 00:25:10,560 But what about the bigger picture? 655 00:25:10,560 --> 00:25:13,960 What happens if this technology falls into the wrong hands? 656 00:25:13,960 --> 00:25:17,240 Could it be used for surveillance or even manipulation? 657 00:25:17,240 --> 00:25:20,920 Those are valid concerns and they're ones we take very seriously. 658 00:25:20,920 --> 00:25:25,240 As scientists, we have a responsibility to consider the potential consequences of our work, 659 00:25:25,240 --> 00:25:27,080 both intended and unintended. 660 00:25:27,080 --> 00:25:31,160 That's why we're actively engaging with ethicists, policymakers, and the public 661 00:25:31,160 --> 00:25:34,280 to ensure that these technologies are developed and used responsibly. 662 00:25:34,280 --> 00:25:38,040 It's almost like we need a new set of rules, a mental bill of rights, 663 00:25:38,040 --> 00:25:40,880 to protect our inner thoughts from unwanted intrusion. 664 00:25:40,880 --> 00:25:41,840 I completely agree. 665 00:25:41,840 --> 00:25:44,800 We need to start having these conversations now 666 00:25:44,800 --> 00:25:48,960 before these technologies become so advanced that it's too late to course correct. 667 00:25:48,960 --> 00:25:52,360 We need to establish clear boundaries and ethical guidelines. 668 00:25:52,360 --> 00:25:54,640 It's like we're standing at a fork in the road. 669 00:25:54,640 --> 00:25:58,320 This technology has the potential to do so much good, 670 00:25:58,320 --> 00:26:01,360 but we need to make sure we're heading in the right direction. 671 00:26:01,360 --> 00:26:06,240 And those conversations need to involve everyone, not just the scientists and engineers. 672 00:26:06,240 --> 00:26:10,680 We need to bring in ethicists, philosophers, legal experts, 673 00:26:10,680 --> 00:26:15,680 and most importantly, the people who will be most affected by these advancements. 674 00:26:15,680 --> 00:26:17,120 What do you think, Michael? 675 00:26:17,120 --> 00:26:20,080 I think it's crucial to approach this with a sense of humility. 676 00:26:20,080 --> 00:26:25,160 We're dealing with the human mind, the most complex and mysterious entity in the known universe. 677 00:26:25,160 --> 00:26:30,480 We need to proceed with caution, respect, and a deep understanding of the potential consequences, 678 00:26:30,480 --> 00:26:31,960 both positive and negative. 679 00:26:31,960 --> 00:26:33,040 Well said, Michael. 680 00:26:33,040 --> 00:26:35,600 And that's a perfect segue into our final segment, 681 00:26:35,600 --> 00:26:39,800 where we'll explore some of the most thought-provoking questions raised by this research. 682 00:26:39,800 --> 00:26:43,360 Are we ready to enter a world where our thoughts are no longer our own? 683 00:26:43,360 --> 00:26:46,040 It's a question that's both exciting and terrifying, 684 00:26:46,040 --> 00:26:51,240 but one thing's for sure, we can't just bury our heads in the sand and pretend this isn't happening. 685 00:26:51,240 --> 00:26:55,400 The future of thought-decoding is unfolding before our very eyes, 686 00:26:55,400 --> 00:26:59,080 and it's up to all of us to shape that future responsibly and ethically. 687 00:26:59,080 --> 00:27:01,320 And that, my friends, is something to ponder, 688 00:27:01,320 --> 00:27:04,200 as we venture deeper into the mysteries of the mind. 689 00:27:04,200 --> 00:27:08,480 All right, so we've spent this deep dive exploring the frontiers of thought-decoding, right? 690 00:27:08,480 --> 00:27:14,160 From fMRI to EEG, the potential benefits, and the ethical dilemmas. 691 00:27:14,160 --> 00:27:16,680 Absolutely. It's a lot to wrap our heads around. 692 00:27:16,680 --> 00:27:17,720 It really is. 693 00:27:17,720 --> 00:27:23,040 But before we sign off, I think it's worth taking a step back and asking ourselves a really big question. 694 00:27:23,040 --> 00:27:24,040 Okay. What's that? 695 00:27:24,040 --> 00:27:28,240 If this technology, thought-decoding becomes commonplace, 696 00:27:28,240 --> 00:27:30,760 how do we safeguard mental privacy? 697 00:27:30,760 --> 00:27:34,720 Hmm, yeah. That's the million-dollar question, isn't it? 698 00:27:34,720 --> 00:27:39,200 It is. I mean, how do we ensure that our thoughts remain, well, our own? 699 00:27:39,200 --> 00:27:45,280 It makes you think about all the ways our privacy has already kind of eroded our online activity, our location data. 700 00:27:45,280 --> 00:27:47,800 Even our faces are scanned and analyzed. 701 00:27:47,800 --> 00:27:52,680 Right. It feels like mental privacy might be the next frontier. 702 00:27:52,680 --> 00:27:54,040 Yeah, it's slippery slope. 703 00:27:54,040 --> 00:27:56,240 And what about the legal ramifications? 704 00:27:56,240 --> 00:27:58,200 Could this be used in court? 705 00:27:58,200 --> 00:27:59,680 I mean, theoretically, yeah. 706 00:27:59,680 --> 00:28:03,240 Imagine someone being convicted based on their thoughts alone. 707 00:28:03,240 --> 00:28:05,000 That raises a whole host of issues. 708 00:28:05,000 --> 00:28:08,120 Right. Like, due process, the presumption of innocence. 709 00:28:08,120 --> 00:28:13,160 Exactly. We need a completely new legal framework to even begin to address that. 710 00:28:13,160 --> 00:28:16,000 And then there's the philosophical question of free will. 711 00:28:16,000 --> 00:28:17,600 Oh, yeah. That's a big one. 712 00:28:17,600 --> 00:28:22,640 Like, if our thoughts can be predicted, does that mean our actions are predetermined? 713 00:28:22,640 --> 00:28:27,600 Hmm. Are we truly in control or are we just puppets on a string? 714 00:28:27,600 --> 00:28:32,920 It's almost like we're back to that idea of the mind as a sacred sanctuary, like Charlotte Bronte said. 715 00:28:32,920 --> 00:28:36,200 Yeah, that inner world, that space where we can be truly ourselves. 716 00:28:36,200 --> 00:28:37,840 Are we on the verge of losing that? 717 00:28:37,840 --> 00:28:40,600 It's a question that keeps me up at night. 718 00:28:40,600 --> 00:28:45,880 Me too. But, you know, I think it's important to remember that technology is ultimately a tool. 719 00:28:45,880 --> 00:28:47,880 Right. It can be used for good or for bad. 720 00:28:47,880 --> 00:28:54,000 Exactly. And it's up to us, you know, society as a whole, to decide how we want to use it. 721 00:28:54,000 --> 00:28:59,520 We need to be proactive, have these conversations now while we still have the chance to shape the future. 722 00:28:59,520 --> 00:29:02,920 Absolutely. We can't just bury our heads in the sand and pretend this isn't happening. 723 00:29:02,920 --> 00:29:06,560 Nope. This technology is coming, whether we're ready for it or not. 724 00:29:06,560 --> 00:29:08,520 So that's our show for today, folks. 725 00:29:08,520 --> 00:29:15,320 We hope you enjoyed this deep dive into the fascinating and frankly unsettling world of thought decoding. 726 00:29:15,320 --> 00:29:16,600 Yeah, thanks for joining us. 727 00:29:16,600 --> 00:29:19,560 Remember, the future is not set in stone. 728 00:29:19,560 --> 00:29:24,080 It's up to all of us to ensure that this technology is used responsibly and ethically. 729 00:29:24,080 --> 00:29:25,880 Right. For the benefit of humanity. 730 00:29:25,880 --> 00:29:30,080 So, keep thinking, keep questioning, and keep the conversation going. 731 00:29:30,080 --> 00:29:50,080 And until next time, thanks for listening.