1 00:00:01,080 --> 00:00:03,240 S1: I mean, have you listened to something that I think 2 00:00:03,240 --> 00:00:09,120 S1: captures extraordinarily well? Why? Arguments that AI don't understand anything 3 00:00:09,119 --> 00:00:15,120 S1: and can't possibly understand anything are completely misguided and empty. 4 00:00:16,360 --> 00:00:19,239 S1: This is a blues version of Without Me by Eminem. 5 00:00:19,800 --> 00:00:24,360 S1: It's from the 1950s, which means it's not real. And 6 00:00:24,360 --> 00:00:27,240 S1: he's also never done a blues version of Without Me, 7 00:00:27,240 --> 00:00:31,120 S1: to my knowledge. And so it's AI generated and it's 8 00:00:31,120 --> 00:00:35,720 S1: objectively a stunning piece of music, and it's quite different 9 00:00:35,720 --> 00:00:38,440 S1: from the original. So let's listen to it. 10 00:00:46,240 --> 00:00:59,000 S2: Guess who's been back again? Shady's back. Tell a friend. 11 00:01:00,970 --> 00:01:03,890 S2: Guess who's back? Guess who's back? Guess who's back. Guess 12 00:01:03,890 --> 00:01:06,890 S2: who's back. Guess who's back. Guess who's back. Guess who's back. 13 00:01:06,890 --> 00:01:14,289 S2: Guess who's back. Guess who's back. Guess who's back. No 14 00:01:14,290 --> 00:01:24,570 S2: no no no no no no no. I've created a monster. 15 00:01:25,050 --> 00:01:28,930 S2: Cause nobody wants to see my shoes no more. They 16 00:01:28,930 --> 00:01:32,530 S2: won't shake I'm chopped liver. Well, if you want shady. 17 00:01:32,569 --> 00:01:34,530 S2: This is what I give you. A little bit of 18 00:01:34,530 --> 00:01:38,009 S2: me mixed with some hard liquor. Some vodka that'll jumpstart 19 00:01:38,010 --> 00:01:40,369 S2: my heart quicker than a shark. When I get shot 20 00:01:40,370 --> 00:01:43,890 S2: at the hospital by the doctor. When I'm not cooperating, 21 00:01:43,890 --> 00:01:47,290 S2: when I'm rocking the table while it's operating. Hey, you 22 00:01:47,290 --> 00:01:49,970 S2: waited this long to stop debating cause I'm back. I'm 23 00:01:49,970 --> 00:01:52,490 S2: on the. I know that you got a job, Miss Cheney, 24 00:01:52,490 --> 00:01:56,130 S2: but your husband's heart problems complicating. So the FCC won't 25 00:01:56,130 --> 00:01:59,050 S2: let me be. Or let me be me. So let 26 00:01:59,090 --> 00:02:02,370 S2: me see. They try to shove me down on MTV, 27 00:02:02,810 --> 00:02:06,730 S2: but it feels so empty without me. 28 00:02:07,450 --> 00:02:12,370 S1: So every time I listen to that, I feel compelled 29 00:02:12,370 --> 00:02:16,170 S1: to move. I think if music makes you dance and 30 00:02:16,169 --> 00:02:21,330 S1: feel things, it is real. If AI models and scaffolding 31 00:02:21,330 --> 00:02:25,370 S1: can be assembled into a product that can replace human workers, 32 00:02:25,889 --> 00:02:30,369 S1: it's intelligent, i.e. it has the ability to understand, pursue, 33 00:02:30,370 --> 00:02:35,450 S1: and accomplish goals. If a technology can perform a task 34 00:02:35,450 --> 00:02:42,889 S1: and produce an output that requires understanding, it understands. So 35 00:02:42,889 --> 00:02:45,450 S1: in this frame, understanding is the ability of an actor 36 00:02:45,450 --> 00:02:49,049 S1: to interpret a given task and desired outcome well enough 37 00:02:49,050 --> 00:02:53,810 S1: to create an acceptable result. AI can clearly do that 38 00:02:53,810 --> 00:02:58,290 S1: now across so many domains. It's true that if you 39 00:02:58,290 --> 00:03:01,660 S1: break open a neural net or a human brain and 40 00:03:01,660 --> 00:03:04,180 S1: start poking at it with a stick or a scalpel 41 00:03:04,180 --> 00:03:07,860 S1: or an electron microscope. There is no place to point 42 00:03:07,860 --> 00:03:12,500 S1: to and say this is understanding, or here is the intelligence, 43 00:03:13,380 --> 00:03:18,019 S1: but it is there in both human brains and in 44 00:03:18,020 --> 00:03:21,900 S1: neural nets, because we see the outputs that prove that 45 00:03:21,900 --> 00:03:26,380 S1: it's there. We should stop wasting cycles on does it 46 00:03:26,380 --> 00:03:30,700 S1: understand or is it intelligent or it can't be intelligent, 47 00:03:30,700 --> 00:03:36,140 S1: because all these behaviors in both animals and technology are 48 00:03:36,140 --> 00:03:41,060 S1: the result of emergent functionality. And the core issue here 49 00:03:41,060 --> 00:03:46,300 S1: is that we still lack transparency into emergence itself, not 50 00:03:46,300 --> 00:03:49,980 S1: just for tech, not just for llms, not just for AI, 51 00:03:50,020 --> 00:03:54,460 S1: but for humans and other animals as well. So let's 52 00:03:54,460 --> 00:03:58,460 S1: not confuse that opacity of emergence itself, which is a 53 00:03:58,540 --> 00:04:04,140 S1: universal human problem in curiosity, with a specific implementation of 54 00:04:04,180 --> 00:04:08,940 S1: that emergence, opacity and a new intelligence stack judge capabilities 55 00:04:08,940 --> 00:04:13,460 S1: by their ground truth outputs. In other words, in your lexicon, 56 00:04:13,860 --> 00:04:17,380 S1: did the creation of that output require understanding and or 57 00:04:17,380 --> 00:04:21,739 S1: intelligence if it were a human doing it? And if so, 58 00:04:21,980 --> 00:04:27,740 S1: then did a non-human actually produce that? Did a non-human 59 00:04:27,740 --> 00:04:31,300 S1: technology produce that same thing that if you saw it 60 00:04:31,300 --> 00:04:37,060 S1: from someone else, it would have required intelligence? Then guess 61 00:04:37,060 --> 00:04:43,140 S1: what that is? Intelligence. Intelligence was used to produce the output. 62 00:04:43,339 --> 00:04:46,900 S1: We can use the output itself, and the fact that 63 00:04:46,900 --> 00:04:51,020 S1: we have defined it as requiring intelligence to say that 64 00:04:51,020 --> 00:04:55,740 S1: anything that could have produced it had intelligence itself. I 65 00:04:55,740 --> 00:04:59,230 S1: think this framing helps clarify the whole situation a little 66 00:04:59,270 --> 00:05:02,310 S1: bit because we can start from ground truth, which is 67 00:05:02,550 --> 00:05:06,190 S1: what we already know and accept as being the product 68 00:05:06,190 --> 00:05:10,070 S1: of intelligence, right? If you hear a song like this, 69 00:05:10,070 --> 00:05:13,390 S1: if you see a work output from an AI digital 70 00:05:13,390 --> 00:05:16,309 S1: worker or something, and you say, well, if a human 71 00:05:16,310 --> 00:05:18,670 S1: would have made that, I would have thought it was 72 00:05:18,670 --> 00:05:22,070 S1: a good product. I would have thought this definitely required intelligence. 73 00:05:22,550 --> 00:05:26,909 S1: That statement there we can use as ground truth. And 74 00:05:26,910 --> 00:05:30,590 S1: then from there, it's a quick step to say anything 75 00:05:30,589 --> 00:05:35,590 S1: that can produce that then also has that intelligence. And 76 00:05:35,589 --> 00:05:38,270 S1: notice that this is completely separate from being able to 77 00:05:38,310 --> 00:05:41,470 S1: explain how it got it. We just have to remind 78 00:05:41,470 --> 00:05:44,830 S1: ourselves we don't know how we got ours either. We 79 00:05:44,830 --> 00:05:48,430 S1: have no idea how. When you look at a spongy 80 00:05:48,470 --> 00:05:52,270 S1: pink brain, how you can store memories in there, how 81 00:05:52,270 --> 00:05:55,550 S1: you can have ideas, how you can have thoughts. We 82 00:05:55,550 --> 00:05:59,830 S1: have no idea where inside of that brain any of 83 00:05:59,830 --> 00:06:04,790 S1: this stuff is actually performed or stored. Now, in humans, 84 00:06:04,790 --> 00:06:07,230 S1: we are not tempted to say, well, since I can't 85 00:06:07,230 --> 00:06:11,350 S1: find it, we are clearly not doing understanding. We are 86 00:06:11,350 --> 00:06:14,990 S1: not doing intelligence. Those things are not there because I 87 00:06:15,029 --> 00:06:18,589 S1: cannot find them by looking at the substrate. We're not 88 00:06:18,589 --> 00:06:21,510 S1: tempted to say that with humans, and we're not tempted 89 00:06:21,510 --> 00:06:23,830 S1: to say it, because we can actually look at the 90 00:06:23,830 --> 00:06:29,470 S1: outputs of ourselves doing those exact things. So why are 91 00:06:29,470 --> 00:06:33,589 S1: we making this mistake with a different type of intelligence? 92 00:06:34,150 --> 00:06:36,950 S1: Why are we looking at outputs that we would judge 93 00:06:37,190 --> 00:06:42,790 S1: as being intelligent or requiring intelligence to make and saying, well, 94 00:06:42,830 --> 00:06:45,190 S1: because I can't find where it was made or how 95 00:06:45,190 --> 00:06:49,390 S1: it was made, it must not be intelligence. It just 96 00:06:49,390 --> 00:06:53,030 S1: doesn't make sense. And hopefully this frame will help you 97 00:06:53,029 --> 00:06:56,349 S1: have the conversation with yourself or with others. We'll see 98 00:06:56,350 --> 00:06:57,190 S1: you in the next one.