WEBVTT - Monkeys, Hamlet and Brain-Computer Interfaces 0:00:00.160 --> 0:00:07.280 Brought to you by Toyota. Let's go places. Welcome to 0:00:07.440 --> 0:00:14.360 Forward Thinking. Hey there, and welcome to Horward Thinking, the 0:00:14.480 --> 0:00:17.920 podcast that looks at the future says, I blame myself 0:00:18.480 --> 0:00:22.760 and I blame the machines. I'm Jonathan Strickland and I'm 0:00:22.840 --> 0:00:25.759 Joe McCormick. In today, we're going to be revisiting a 0:00:25.800 --> 0:00:28.720 topic we've looked at on this show before, but coming 0:00:28.800 --> 0:00:32.080 coming at you with an update on the subject of 0:00:32.240 --> 0:00:36.159 brain computer interfaces. Okay, the concept here is that you 0:00:36.240 --> 0:00:39.640 train a computer to interpret your brain waves. Then you 0:00:39.760 --> 0:00:43.239 think really hard about doing something, I mean not just 0:00:43.320 --> 0:00:45.879 anything like like moving a cursor or or typing a 0:00:45.920 --> 0:00:50.400 word or something like that, and then it happens. It's 0:00:50.400 --> 0:00:54.240 like magic, but with more electrones. Exactly. So, brain computer 0:00:54.280 --> 0:00:57.960 interfaces could be anything from an ability to to be 0:00:58.000 --> 0:00:59.840 able to, like you say, move a cursor, type of 0:00:59.840 --> 0:01:04.319 war and communicate or manipulate a robotic limb. There are 0:01:04.319 --> 0:01:06.800 a lot of different like that, exactly. There are a 0:01:06.840 --> 0:01:09.560 lot of different potential applications. But honestly, it just it 0:01:09.680 --> 0:01:14.520 boils down to stripping away any sort of of interface 0:01:14.560 --> 0:01:18.640 that would require movement, and it relies purely on thought right, 0:01:18.680 --> 0:01:20.679 And of course that's the output half of it. There's 0:01:20.720 --> 0:01:22.440 also the input half of it, where you could say 0:01:22.440 --> 0:01:25.720 a brain you know, input oriented brain computer interface would 0:01:25.760 --> 0:01:29.680 be getting stimuli without the use of your senses, maybe 0:01:30.040 --> 0:01:33.679 like having an idea sent directly to your brain instead 0:01:33.680 --> 0:01:36.640 of through your eyes or something so something like like 0:01:36.760 --> 0:01:41.800 a a an interface where you are receiving information from 0:01:41.920 --> 0:01:46.040 the outside world in some manner. For example, Uh, let's 0:01:46.840 --> 0:01:48.600 one one thing we'll talk about towards the end of 0:01:48.640 --> 0:01:51.480 the episode is the idea of being able to access, say, 0:01:51.600 --> 0:01:56.120 the Internet, directly through your brain. Yeah. A common example 0:01:56.160 --> 0:01:58.720 would be, you know, could we put some micro electrodes 0:01:58.760 --> 0:02:01.120 in your visual corte x that would allow you to 0:02:01.360 --> 0:02:04.600 see without seeing. You could have your you know, your 0:02:04.600 --> 0:02:07.880 eyes closed but still perceive an image in your brain. Man, 0:02:08.000 --> 0:02:11.639 sleeping would be so weird. Okay, I'm just looking through 0:02:11.639 --> 0:02:15.120 the top of my head. No. Anyway, So we've got 0:02:15.160 --> 0:02:17.720 a lot to say about this, but you may think, 0:02:17.800 --> 0:02:21.680 if you're a long time listener, haven't you guys addressed 0:02:21.680 --> 0:02:23.680 this before, Joe, you alluded to that. We actually have 0:02:23.760 --> 0:02:25.440 quite a few times. A Yeah, and one of our 0:02:25.520 --> 0:02:28.120 very early episodes was about this technology that was called 0:02:28.160 --> 0:02:31.000 Computers on the Mind all from all the way back 0:02:31.040 --> 0:02:35.079 in April. Pretty sure I titled that one. I'm pretty 0:02:35.080 --> 0:02:37.359 sure that's accurate. That is not one of the greatest 0:02:37.400 --> 0:02:42.520 of our punny title No, no, they're not all home runs. Anyway, 0:02:42.639 --> 0:02:46.519 you're saying, I wish that we had a cricket sound board, 0:02:47.120 --> 0:02:51.280 and here's that anyway. Yes, yes, as when we recorded 0:02:51.320 --> 0:02:55.120 that first episode about it, there was some exploratory research 0:02:55.200 --> 0:02:58.720 into brain computer interfaces that was really amazing that there 0:02:58.760 --> 0:03:01.920 was this team out of out of mostly Brown University, 0:03:01.960 --> 0:03:04.960 also involving Stanford and a bunch of other researchers calling 0:03:05.000 --> 0:03:09.200 themselves brain Gate brain Gate, uh, that that had created 0:03:09.240 --> 0:03:13.040 this interface that allowed a woman with quadriplegia to feed 0:03:13.080 --> 0:03:17.440 herself using a robotic arm um. Uh. They also created 0:03:17.480 --> 0:03:21.600 the first wireless implantable electrode that was being used at 0:03:21.600 --> 0:03:24.480 the time in animal testing. But all of this was 0:03:24.840 --> 0:03:29.079 fairly foundational work. Uh. And we've also talked of course 0:03:29.120 --> 0:03:33.840 about brain computer interfaces more recently during discussions into computer 0:03:33.880 --> 0:03:37.360 interfaces in general, episodes like your Body as a Computer 0:03:37.440 --> 0:03:41.720 Interface from April of sixteen and did Minority Report shape 0:03:41.840 --> 0:03:47.160 user interfaces, which was from September. The answer was yeah, 0:03:47.400 --> 0:03:53.920 surprisingly like kind of upsettingly. Uh, that Minority Report is 0:03:53.960 --> 0:03:57.840 a fine film, um, but but precog thing though serious. No, 0:03:57.920 --> 0:04:00.240 I'm not going to get off on that. If I 0:04:00.280 --> 0:04:04.080 do that the episodes over, I'll just be granting and raving. 0:04:04.200 --> 0:04:07.560 Pray continue, thank you sir. Um. But yeah, so, so 0:04:07.600 --> 0:04:11.640 all of those episodes focused more on on physical computer controls, 0:04:11.680 --> 0:04:14.480 you know, like like a keyboards or or mice, or 0:04:14.560 --> 0:04:17.800 or even touch screens or some some voice and gesture controls, 0:04:17.800 --> 0:04:20.599 which are things that most of us use and are 0:04:20.720 --> 0:04:24.400 used to. Um, But wouldn't it be nice if we 0:04:24.440 --> 0:04:27.799 could just think at our computers and have them execute 0:04:27.800 --> 0:04:30.640 our commands. It would be pretty sexy for everybody, really, 0:04:30.960 --> 0:04:33.960 but but really especially useful for folks, uh, like that 0:04:34.000 --> 0:04:37.159 woman with quadriplegia, who for whatever reason have trouble with 0:04:37.160 --> 0:04:41.600 fine motor controls and or speech. So we thought that 0:04:41.640 --> 0:04:43.839 today we would look into this topic again and kind 0:04:43.839 --> 0:04:47.760 of see what the state of the brain computer interface 0:04:47.920 --> 0:04:52.440 union is. But first let's let's talk Let's talk about 0:04:52.480 --> 0:04:54.080 kind of like the basics, Like like, let's get that 0:04:54.160 --> 0:04:58.960 ground level placed poor as it is, how how do 0:04:59.000 --> 0:05:01.640 you make connection between a brain and a computer anyway. 0:05:02.000 --> 0:05:06.200 It's not necessarily easy, but it's also not necessarily as 0:05:06.200 --> 0:05:09.480 hard as you might imagine, because we already do share 0:05:09.760 --> 0:05:15.440 a semi common language between brains and electronic machines, electrical machines, 0:05:15.800 --> 0:05:20.880 which is electricity. Sure. Yeah, so so the brain, you 0:05:20.920 --> 0:05:26.599 would say, communicates between cells via electrochemical potential, and by 0:05:26.800 --> 0:05:31.920 using electrical charge, you can make the brain do what 0:05:32.000 --> 0:05:35.360 it would normally be doing when when part of it 0:05:35.480 --> 0:05:40.680 was activating. That's very imprecise language activating. But well it's 0:05:40.680 --> 0:05:44.160 a pretty imprecise uh practice as well. I mean I 0:05:44.160 --> 0:05:47.000 think I think we've gotten pretty good and making the 0:05:47.040 --> 0:05:52.560 brain like totally overwhelmed. It's it's hard, Yeah, it's harder 0:05:52.600 --> 0:05:56.320 to get very specific outcomes, right, like like to to 0:05:56.560 --> 0:06:01.960 cause someone to have a specific sensation shin or trigger 0:06:02.000 --> 0:06:04.760 a memory or anything like that. I mean, overwhelming the 0:06:04.800 --> 0:06:08.640 brain not a problem if you want to just are 0:06:08.640 --> 0:06:12.480 you just talking about electric shock there? Essentially. What's a 0:06:12.480 --> 0:06:17.200 little easier, though, is training a computer training software. What 0:06:17.560 --> 0:06:21.360 different brain signals in our our meaning? Like like like 0:06:21.400 --> 0:06:26.279 if a person's brain lights up in a certain way, Ah, 0:06:26.440 --> 0:06:28.640 what does that mean? Right? Right? So are they thinking 0:06:28.680 --> 0:06:30.480 about what are they trying to do? So there's a 0:06:30.760 --> 0:06:35.360 training process in any brain computer interface that has two components. 0:06:35.400 --> 0:06:37.800 You're training the person to use the interface and your 0:06:37.800 --> 0:06:42.039 train the computer on what the person's brain waves actually 0:06:42.120 --> 0:06:48.040 are are linked towards. So you have a long, uh 0:06:48.279 --> 0:06:51.880 training process. I hate to keep using that phrase, but 0:06:51.880 --> 0:06:53.920 it's the most accurate. You have a very long training 0:06:53.920 --> 0:06:56.320 process to get to a point where you can actually 0:06:56.400 --> 0:07:00.320 do something useful. So a very very simple exam bowl 0:07:00.880 --> 0:07:03.120 would be all right, let's let's talk about your your 0:07:03.160 --> 0:07:08.320 basic brain computer interface, which tends to involve invasive surgery. Right, 0:07:08.920 --> 0:07:10.760 you have to because e e g. S don't really 0:07:10.800 --> 0:07:13.800 cut it, is the thing. Yeah, they're not electrodes. Um 0:07:14.120 --> 0:07:16.160 your your brain, your your skull is not the very 0:07:16.160 --> 0:07:20.280 best fair day cage, but it's pretty effective, actually exactly right. 0:07:20.320 --> 0:07:23.239 So you need to get access to those sweet sweet 0:07:23.920 --> 0:07:28.760 electrical uh ilse impulses on the brain surface. So typically 0:07:28.840 --> 0:07:32.760 it involves uh brain surgery. You're actually opening up the 0:07:32.800 --> 0:07:36.920 skull cap and you're putting overlaying maybe an array of 0:07:36.920 --> 0:07:41.480 of sensors of electrodes on the surface of the brain. 0:07:41.760 --> 0:07:44.440 Typically you don't go into the brain for this, it 0:07:44.520 --> 0:07:46.600 does tend to be along the surface of the brain, 0:07:46.960 --> 0:07:51.160 and then you have usually some sort of transcranial connector, 0:07:51.880 --> 0:07:54.120 so it actually is going through the skull when you 0:07:54.160 --> 0:07:59.720 reattach the skull to then wire typically to a computer system. 0:07:59.760 --> 0:08:02.360 There are wireless ones as well, though those are earlier 0:08:02.400 --> 0:08:06.680 in development and they still require transcranial ports. Right right, 0:08:06.680 --> 0:08:09.400 You're still you're still putting something in the brain and 0:08:09.440 --> 0:08:14.400 then like plugging a wireless device into the outside of 0:08:14.440 --> 0:08:17.440 your skull. Yeah, yeah, it's there's not there's not a 0:08:17.480 --> 0:08:20.760 way right now where we have truly wireless where it's 0:08:20.760 --> 0:08:24.400 all within the skull itself and it's contained and therefore 0:08:24.480 --> 0:08:27.960 you don't have any sign of a connector there. And 0:08:28.000 --> 0:08:32.120 because again um that the bone is really surprisingly effective 0:08:32.480 --> 0:08:37.760 at keeping electrical signals inside and outside based on where 0:08:37.800 --> 0:08:39.439 they are. Yeah, it turns out, it turns out that's 0:08:39.440 --> 0:08:42.440 a good thing for most of us most of the time. Yes, yeah, 0:08:42.480 --> 0:08:44.880 but it's it does make it very difficult to interface 0:08:44.880 --> 0:08:47.800 with a machine this way. So you then once you 0:08:47.840 --> 0:08:50.840 have the connection, the physical connection or the wireless connection 0:08:51.240 --> 0:08:56.200 from the person to the computer, you establish some form 0:08:56.400 --> 0:08:59.920 of test that will start to train both the person 0:09:00.280 --> 0:09:03.959 with the interface and the machine itself on working together. 0:09:04.240 --> 0:09:07.880 And a typical one might be that the person is 0:09:07.920 --> 0:09:12.480 looking at a picture of a ball, and there's a 0:09:12.559 --> 0:09:15.280 second picture of a ball, and their job is to 0:09:16.480 --> 0:09:21.400 imagine trying to force the one ball so that overlays 0:09:21.440 --> 0:09:23.400 on top of the other one. So let's say one's 0:09:23.400 --> 0:09:25.680 green and one's yellow, and so your your job is 0:09:25.679 --> 0:09:29.360 to to imagine moving the green ball over top the 0:09:29.440 --> 0:09:32.640 yellow ball, and you do this over and over and 0:09:32.720 --> 0:09:37.000 over again. And there's actually a scoring system typically, and 0:09:37.160 --> 0:09:41.600 it's one to help motivate the person who's who's training. 0:09:42.280 --> 0:09:45.640 It's also meant to give feedback on all right, well, 0:09:45.800 --> 0:09:49.840 because this was inconsistent between trials three and four, we 0:09:49.920 --> 0:09:51.680 have to keep working on this in order for the 0:09:51.720 --> 0:09:54.360 machine to understand what it is you are trying to do. 0:09:54.520 --> 0:09:59.080 Otherwise it will misinterpret your thought and you could get 0:09:59.080 --> 0:10:02.080 an erroneous salt. So if you're getting into something where 0:10:02.080 --> 0:10:05.520 you're trying to say, type out a word and you're thinking, 0:10:05.920 --> 0:10:07.920 I want the cursor to move to the letter T, 0:10:08.679 --> 0:10:11.720 but the machine has misinterpreted that, you could end up 0:10:11.760 --> 0:10:16.280 with typos and mistakes. So you go through this laborious 0:10:16.320 --> 0:10:19.160 process where both the machine and the person learn and 0:10:19.200 --> 0:10:22.959 then ultimately you reach a point where the machine has 0:10:23.000 --> 0:10:26.600 a pretty good feel for what it means when you 0:10:26.640 --> 0:10:30.240 think in a certain way in order to manipulate stuff, 0:10:30.360 --> 0:10:32.840 and and that's when you get the results you wanted. 0:10:33.400 --> 0:10:37.199 It is. I'm not gonna say it's inelegant. I think 0:10:37.200 --> 0:10:39.520 it's an elegant solution, but it's an elegant solution that 0:10:39.559 --> 0:10:43.400 takes a good deal of work and effort and time. Well, 0:10:43.440 --> 0:10:45.240 I mean, one thing about that that occurs to me 0:10:45.280 --> 0:10:49.920 immediately is that it seems quite personalized to the individual user, 0:10:50.280 --> 0:10:54.880 not necessarily like if you design, say you're working on 0:10:54.920 --> 0:10:59.280 a speech to text program or something, hopefully what you're 0:10:59.320 --> 0:11:03.160 doing is you're you're training it for universal speech to text. 0:11:03.360 --> 0:11:05.280 You know, if you're speaking, as long as you're speaking 0:11:05.320 --> 0:11:07.160 in a way that people would normally be able to 0:11:07.240 --> 0:11:09.880 understand the words you're saying without some kind of incredibly 0:11:09.880 --> 0:11:12.400 thick accent or something, it's going to be able to 0:11:12.400 --> 0:11:16.000 turn that into text. Would the same thing necessarily be true? 0:11:16.040 --> 0:11:18.280 Would you know? So you would hook up one person's 0:11:18.320 --> 0:11:21.640 brain and learn how to how to allow them to 0:11:21.679 --> 0:11:26.400 control a computer cursor without touching the mouse or the keys, 0:11:26.440 --> 0:11:29.800 just their brain. Could you then come and bring somebody 0:11:29.840 --> 0:11:32.240 else in and hook their brain up and have them 0:11:32.280 --> 0:11:36.520 do the same thing, certainly not yet. Yeah, it's I 0:11:36.520 --> 0:11:39.120 mean you'd have to retrain them. Basically, you have to 0:11:39.120 --> 0:11:42.080 retrain both the person and computer. Yeah, exactly, Yeah, it's 0:11:42.120 --> 0:11:44.920 it's it's not a one size fits all kind of solution, 0:11:45.280 --> 0:11:48.240 which is also why I get very skeptical whenever I 0:11:48.280 --> 0:11:51.960 see any of those distractions, the games that have the 0:11:52.000 --> 0:11:55.880 little e g. Style headset and you're supposed to focus 0:11:55.880 --> 0:11:58.640 in order to make the game work, I get a 0:11:58.679 --> 0:12:01.880 little I mean, if it's just trying to pick up 0:12:01.920 --> 0:12:04.720 an increase in brainwave activity and that's all it's doing, 0:12:04.800 --> 0:12:07.439 then it's kind of like just a dimmer switch, right, 0:12:07.520 --> 0:12:11.560 Like there's nothing super sophisticated about it. But when it 0:12:11.600 --> 0:12:16.319 gets to like more uh complicated claims that hey, if 0:12:16.360 --> 0:12:18.200 you if you concentrate on the ball to go up, 0:12:18.240 --> 0:12:19.959 the ball goes up, and if you concentrate on the 0:12:20.000 --> 0:12:22.520 ball to go down, the ball goes down, and it's 0:12:22.559 --> 0:12:25.600 a one size fits all kind of thing, I often 0:12:25.920 --> 0:12:28.760 think that's more or less bologny. I mean, not that 0:12:28.840 --> 0:12:31.360 it doesn't work at all, But it doesn't I doubt 0:12:31.400 --> 0:12:34.600 it works in the way they're saying it works, and 0:12:34.720 --> 0:12:37.800 it probably I mean, if it does work even vaguely 0:12:37.840 --> 0:12:39.320 in the way that they say that it works, it 0:12:39.400 --> 0:12:43.760 probably doesn't work very well. Yeah, because again, unfortunately, right now, 0:12:43.800 --> 0:12:47.720 that surgery is an integral part of getting getting to 0:12:47.800 --> 0:12:52.080 those juicy, juicy brainwaves. Yeah. So it was just last 0:12:52.160 --> 0:12:56.320 year that we actually got to the point where a 0:12:56.400 --> 0:13:00.240 wireless solution was even a thing. In that case, it's 0:13:00.240 --> 0:13:03.240 not for humans, well yeah, yeah, well, and and the 0:13:03.280 --> 0:13:07.439 wireless solution is is so exciting because because previously the 0:13:08.000 --> 0:13:11.080 thing that you plug into your skull to to access 0:13:11.080 --> 0:13:15.000 those electrodes is a tether. That's that's meaning the ear, 0:13:15.040 --> 0:13:17.680 the ear tied to a laboratory in which that that 0:13:17.760 --> 0:13:20.559 tether exists or or I mean, even if even if 0:13:20.559 --> 0:13:23.000 it was available for like home use, you wouldn't be 0:13:23.000 --> 0:13:25.559 able to be mobile while you were using this kind 0:13:25.600 --> 0:13:28.640 of device. So so getting that that mobility and a 0:13:28.640 --> 0:13:31.480 little bit more comfort for subjects who are who are 0:13:31.960 --> 0:13:34.440 testing out this kind of stuff is is really critical 0:13:34.559 --> 0:13:37.160 in moving towards a place where where it could be 0:13:37.360 --> 0:13:40.960 actually useful for actual people in their actual lives, right, 0:13:41.240 --> 0:13:45.240 And we've also seen progress that's akin to Moore's law, 0:13:45.400 --> 0:13:51.000 the idea that over time, you can increase the abilities 0:13:51.200 --> 0:13:54.720 of say a microprocessor by reducing the size of the 0:13:54.760 --> 0:13:57.920 individual components. The same thing has been true for the 0:13:57.920 --> 0:14:00.240 electrodes that have been used in this kind of thing, 0:14:00.320 --> 0:14:03.920 all right, which is so cool because because anytime that 0:14:03.920 --> 0:14:05.920 you put an implant in the brain, it's it's going 0:14:05.960 --> 0:14:08.079 to cause a little bit of scarring in the brain tissue. 0:14:08.160 --> 0:14:12.560 And that's scarring uh gets larger over time. Um. The 0:14:12.720 --> 0:14:16.800 eventually the scar tissue begins to uh to degrade the 0:14:16.880 --> 0:14:19.680 quality of the signal that the electrodes can send out 0:14:20.120 --> 0:14:22.600 because it's it's it's a thicker it's a thicker material 0:14:22.680 --> 0:14:25.960 than your normal brain stuff. It's almost like a callous 0:14:26.000 --> 0:14:30.320 exactly exactly so um so so smaller implants equal less scarring, 0:14:30.640 --> 0:14:34.720 equal better technology over time equal yea yeah. And obviously, 0:14:34.760 --> 0:14:37.920 if you were to undergo a procedure where you are 0:14:37.960 --> 0:14:42.120 going to have some sort of electrode array implanted in 0:14:42.160 --> 0:14:44.880 your brain, you want it to last for as long 0:14:44.920 --> 0:14:47.000 as it possibly can so that you don't have to 0:14:47.080 --> 0:14:50.960 undergo a future procedure in order to correct any issues 0:14:51.000 --> 0:14:53.280 from scarring exactly. So this is a this is a 0:14:53.360 --> 0:14:56.280 huge advance, even though it's very talian. Uh. Yeah, and 0:14:56.560 --> 0:14:59.440 hope hopefully, I like Jonathan said, it's currently that the 0:14:59.520 --> 0:15:05.080 wireless stuff is only available for animal testing. Um, hopefully 0:15:05.120 --> 0:15:07.760 a wireless device will be cleared for human testing soon. 0:15:08.160 --> 0:15:10.400 That the last I heard about it in I believe 0:15:10.480 --> 0:15:14.560 December of there was one company that was that was 0:15:14.600 --> 0:15:19.240 working with the brain Gate project that was was submitting 0:15:19.440 --> 0:15:22.920 a proposal to the FDA. So no word on that 0:15:23.120 --> 0:15:28.400 as of Google today, right, but but you know, fingers cressed. Yeah, 0:15:28.440 --> 0:15:31.520 and so let's kind of segue into some of the 0:15:31.560 --> 0:15:35.720 work that's been done recently to help improve the approach 0:15:35.920 --> 0:15:40.200 of brain computer interfaces. We we have to say, Ralph 0:15:40.240 --> 0:15:45.000 the Bat, there's nothing that's like the the holy Grail 0:15:45.440 --> 0:15:50.080 of brain computer interfaces that magically makes them work perfectly. 0:15:50.560 --> 0:15:53.360 But we've seen some pretty cool advances. Yeah. There was 0:15:53.400 --> 0:15:56.200 one story in particular that kind of spurred this this 0:15:56.240 --> 0:16:02.440 episode two into creation that involved alves, monkeys, and hamlet. Yeah. Yeah, 0:16:02.520 --> 0:16:07.840 and I I really I campaigned hard to cover this 0:16:08.000 --> 0:16:10.120 for a show that we do called Hell Stuff Works. 0:16:10.160 --> 0:16:15.760 Now because I'm a technology freak and a Shakespeare freak, 0:16:16.120 --> 0:16:20.360 So the dual punch hit me right in the stern um. Uh. 0:16:20.400 --> 0:16:22.840 So we're talking about a project that came out of 0:16:22.880 --> 0:16:26.040 Stanford bio x. There was a group of scientists at 0:16:26.040 --> 0:16:30.440 Stanford who were looking specifically at ways to improve brain 0:16:30.480 --> 0:16:34.520 computer interfaces for people who have mobility issues. They said 0:16:34.520 --> 0:16:36.720 that one of the biggest challenges they were facing was 0:16:36.800 --> 0:16:41.040 that the interfaces that existed up to that point were 0:16:41.120 --> 0:16:45.920 relatively slow, and frequently people were finding, uh it difficult 0:16:45.920 --> 0:16:48.360 to use that they were there were it was inserting errors. 0:16:48.560 --> 0:16:51.240 They were getting misspellings or typos or whatever. And that 0:16:51.320 --> 0:16:56.000 causes frustration obviously. And I don't I know you guys, like, 0:16:56.600 --> 0:17:00.000 if you've ever been working on a task that's particularly 0:17:00.040 --> 0:17:02.280 hard and then you get frustrated, it suddenly seems like 0:17:02.280 --> 0:17:05.239 the difficulty of that task is skyrocketed. It's like if 0:17:05.280 --> 0:17:08.240 you're already trying to do something difficult on your computer 0:17:08.320 --> 0:17:11.679 and then your computer starts getting slow. Yeah, isn't that 0:17:11.720 --> 0:17:14.800 the best feeling? It's so frustrating. And have you ever 0:17:14.840 --> 0:17:16.800 been playing a video game where there was a glitch 0:17:16.880 --> 0:17:19.159 in it and and you were like like the controls 0:17:19.200 --> 0:17:21.600 maybe like especially in older games, where like the controls 0:17:21.600 --> 0:17:24.600 were already a little bit like like rough, like tank like, 0:17:25.040 --> 0:17:26.359 and then all of a sudden, you just get to 0:17:26.359 --> 0:17:28.720 a glitch where you're like, well, it's it's physically impossible 0:17:28.800 --> 0:17:31.239 for me to beat this. Yeah, they're there's great. There 0:17:31.280 --> 0:17:33.159 are times where you just you know, you feel that 0:17:33.240 --> 0:17:36.800 level of frustration, and it it makes something that is 0:17:36.840 --> 0:17:40.240 already challenging even more difficult, right like even if that 0:17:40.320 --> 0:17:44.479 barrier were not there anymore, because of your frustration, your 0:17:44.480 --> 0:17:48.840 ability to focus has has decreased, and so there was 0:17:48.880 --> 0:17:52.120 a real need to try and improve this tech. Uh. 0:17:52.160 --> 0:17:55.080 And in fact, there's a quote specifically from a piece 0:17:55.160 --> 0:17:58.040 in Stanford News that said earlier versions of this technology 0:17:58.080 --> 0:18:01.199 have already been tested successfully in people with paralysis, but 0:18:01.280 --> 0:18:04.600 the typing was slow and imprecise. This latest work tests 0:18:04.600 --> 0:18:07.439 improvements to the speed and accuracy of the technology that 0:18:07.520 --> 0:18:11.160 interprets brain signals and drives the cursor. So in this case, 0:18:11.160 --> 0:18:14.439 they're talking about an interface where you have like a 0:18:14.480 --> 0:18:17.760 screen with a bunch of letters on it, and there's 0:18:17.800 --> 0:18:22.119 a cursor as well, and you can concentrate and move 0:18:22.200 --> 0:18:26.080 the cursor to the appropriate letter and then concentrate again 0:18:26.119 --> 0:18:29.359 as if you were clicking on that letter and its selects. It. 0:18:29.359 --> 0:18:32.120 It acts as a type, you know, like a typewriter, 0:18:32.240 --> 0:18:34.800 but you're doing it very much a point like a 0:18:36.000 --> 0:18:38.800 search and pack kind of approach, where instead of a finger, 0:18:38.840 --> 0:18:42.439 you're using your brain. Now, the way they tested the 0:18:42.520 --> 0:18:46.200 improvements to this technology, it was a bit unusual. They 0:18:46.320 --> 0:18:50.760 used monkeys as test subjects. Not unusual, right, not unusual 0:18:50.920 --> 0:18:54.160 for this. But they trained them to concentrate on symbols, 0:18:54.720 --> 0:19:00.280 and then they had them transcribe passages from a couple 0:19:00.320 --> 0:19:03.239 of different sources, the New York Times, some articles from 0:19:03.280 --> 0:19:06.760 the New York Times. They used that as a test 0:19:07.040 --> 0:19:11.960 um text to transcribe, and in one case Hamlet. So 0:19:12.160 --> 0:19:14.480 that of course prompted people to make reference to that 0:19:14.560 --> 0:19:16.760 old idea that if you stuck an infinite number of 0:19:16.800 --> 0:19:19.159 monkeys in a room with an infinite number of typewriters, 0:19:19.160 --> 0:19:21.040 sooner or later one of them would produce the complete 0:19:21.040 --> 0:19:24.280 works of Shakespeare just by chance. Because the definition of 0:19:24.320 --> 0:19:27.440 infinite is in an infinite number of monkeys. I always 0:19:27.480 --> 0:19:31.000 thought it was this finite number of monkeys given infinite time. 0:19:31.680 --> 0:19:34.960 Either way, it's same thing right. I feel like the 0:19:35.000 --> 0:19:38.840 Douglas Adams version was infinite monkeys, infinite typewriters, and infinite 0:19:38.880 --> 0:19:41.639 If it were infinite monkeys and just one typewriter, boy, 0:19:42.119 --> 0:19:45.360 who I'd imagine. But the infinite monkeys get the Shakespeare 0:19:45.440 --> 0:19:48.960 done a lot faster than finite monkeys, you would imagine, Yeah, 0:19:49.040 --> 0:19:51.520 infinite monkey. I don't know, do they. I mean, at 0:19:51.520 --> 0:19:53.960 a certain point, is adding more monkeys help or hurt 0:19:54.000 --> 0:19:58.400 the situation? It always helps. Adding more monkeys always happens. 0:19:58.800 --> 0:20:01.800 Why you're the optimist. Three things. If you take nothing 0:20:01.840 --> 0:20:04.800 else from this show, Ladies and gentlemen, take the words 0:20:04.840 --> 0:20:10.119 of wisdom from Joe. Adding monkeys always helps. Um So 0:20:10.720 --> 0:20:13.920 in this case, obviously the monkeys weren't typing up Hamlet 0:20:14.040 --> 0:20:16.840 purely by chance, they were transcribing it, and in fact 0:20:16.920 --> 0:20:19.600 only one was doing it. But yeah, and so, so 0:20:19.680 --> 0:20:22.119 what what the researchers did here is they they implanted 0:20:22.160 --> 0:20:25.800 electrodes into the monkey's motor cortexes and and the monkeys 0:20:25.800 --> 0:20:29.520 were trained to move cursors around the screen first with 0:20:29.560 --> 0:20:33.320 actual gestures, and the camera watched their movements while the 0:20:33.359 --> 0:20:36.600 software was watching their brain waves. Thus the computer learned 0:20:36.640 --> 0:20:40.040 what the monkeys intent to move looked like right, And 0:20:40.320 --> 0:20:42.600 keep in mind, they had no comprehension of the text 0:20:42.680 --> 0:20:46.320 as far as we are aware, they weren't suddenly all. 0:20:46.359 --> 0:20:48.520 But they didn't all become subscribers to The New York 0:20:48.520 --> 0:20:53.760 Times or aficionados of The Bard As far as we know, 0:20:54.160 --> 0:20:56.879 none of them, as far as I'm aware, contacted Kenneth Bruna, 0:20:57.040 --> 0:21:01.159 but several of them did become oxford Ians. There were 0:21:01.160 --> 0:21:03.199 a couple of bacon Vans in the group. I'm not 0:21:03.240 --> 0:21:07.680 gonna lie. Uh. They were literally just just replicating those 0:21:07.720 --> 0:21:11.560 words by doing that that kind of cursor click approach 0:21:11.600 --> 0:21:16.560 we were talking about, by concentrating. But despite those limitations, 0:21:16.640 --> 0:21:19.960 despite those qualifiers, they were able to reach a speed 0:21:20.000 --> 0:21:22.879 of around twelve words per minute, which was pretty impressive 0:21:22.920 --> 0:21:26.840 for a monkey, Really pretty impressive considering the technology at all. 0:21:27.560 --> 0:21:32.800 Absolutely so for fun, because I was curious. I thought 0:21:32.800 --> 0:21:34.320 it might be interested to see how long it would 0:21:34.359 --> 0:21:36.320 take a monkey to type out Hamlet in this way, 0:21:36.440 --> 0:21:39.320 keeping in mind the monkey is again transcribing Hamlet, not 0:21:39.400 --> 0:21:43.919 actually just spontaneously typing up the play. Right, Sure, but 0:21:44.160 --> 0:21:46.199 I guess this means you have to pick a certain 0:21:46.320 --> 0:21:49.480 version of Hamlet that's true. I I picked I picked 0:21:50.119 --> 0:21:53.520 a pretty standard version, and I was I actually did 0:21:53.680 --> 0:21:56.760 go to a site that gave like a number of 0:21:56.840 --> 0:21:59.800 words for every work. Hamlet, by the way, in case 0:21:59.840 --> 0:22:04.800 you're curious, is the longest of Shakespeare's plays by number 0:22:04.880 --> 0:22:08.639 of words and by running time. If they're doing the 0:22:08.680 --> 0:22:13.399 whole darn thing, it has thirty seven words. So at 0:22:13.440 --> 0:22:15.280 twelve words per minute, it would take a monkey about 0:22:15.280 --> 0:22:18.040 forty two point four hours to type out the whole thing. 0:22:18.440 --> 0:22:22.000 But to replicate all of Shakespeare's plays and not as sonnets, 0:22:22.040 --> 0:22:24.199 I didn't include the sonnets. But if you were to 0:22:24.240 --> 0:22:26.600 replicate all of Shakespeare's plays, it would take about forty 0:22:26.840 --> 0:22:30.320 days of typing at twelve words per minute. Uh so 0:22:30.960 --> 0:22:35.200 now we know, right, Um, if you remove the chance part, 0:22:35.640 --> 0:22:38.080 we got chance in there. We still that's still a mystery, 0:22:38.240 --> 0:22:40.320 but it's it's still it's still like like certainly not 0:22:40.359 --> 0:22:44.000 an infinite amount of time. No, that's a pretty I mean, yeah, 0:22:44.359 --> 0:22:45.920 a month and a half right now, we do still 0:22:45.920 --> 0:22:48.280 need to add more monkeys to Joe's point in order 0:22:48.320 --> 0:22:51.760 to answer the other question. Now. They also pointed out 0:22:51.760 --> 0:22:54.920 that they were not using any completion algorithms. The kind 0:22:54.920 --> 0:22:58.120 of stuff that you might find in smartphones or some 0:22:58.160 --> 0:23:00.679 computer programs, where you start type being a word and 0:23:00.720 --> 0:23:04.000 it tries to guess what word you intend based upon 0:23:04.119 --> 0:23:07.040 context and your frequency of using certain words, and it's 0:23:07.080 --> 0:23:12.120 always wrong. It's frequently wrong, not necessarily. I actually had 0:23:12.200 --> 0:23:16.239 autocomplete back before I started using swipe text, which is 0:23:16.400 --> 0:23:22.000 just another exercise of futility that I apparently am determined 0:23:22.040 --> 0:23:24.919 to continue to pursue. Um. When I was typing in 0:23:25.080 --> 0:23:29.640 letter by letter, I noticed that autocomplete was more right 0:23:29.720 --> 0:23:33.080 than wrong. But when it was wrong, it was hilariously wrong. Yeah, 0:23:33.119 --> 0:23:35.160 and often in a way that makes you feel bad 0:23:35.200 --> 0:23:39.360