WEBVTT - AI’s environmental cost - Lab 094 0:00:03.680 --> 0:00:09.200 I'm t T and I'm Zakiyah and this is Dope Labs. 0:00:11.360 --> 0:00:14.600 Welcome to Dope Labs, a weekly podcast that mixes hardcore 0:00:14.680 --> 0:00:17.599 science with pop culture and a healthy dose of friendship. 0:00:22.800 --> 0:00:25.600 We don't really show our faces a lot or at 0:00:25.640 --> 0:00:28.840 all with the podcast. We don't do because I know 0:00:28.880 --> 0:00:31.920 a lot of podcasts now are like, oh, you can 0:00:31.960 --> 0:00:35.360 come and watch us, Yeah, you can come and watch 0:00:35.479 --> 0:00:38.000 us on YouTube or whatever, and we don't do that 0:00:38.120 --> 0:00:40.000 because we don't feel like we need. 0:00:40.000 --> 0:00:40.320 To do that. 0:00:40.840 --> 0:00:42.879 But what I have said this is one of the 0:00:42.880 --> 0:00:46.160 things that I did do in preparation for this lab 0:00:47.120 --> 0:00:51.920 was I asked chat GBT to make a photo of us, 0:00:52.000 --> 0:00:55.480 and so I uploaded some reference photos and I just 0:00:55.520 --> 0:01:01.440 sent you what they were, and I want your initial reactions. 0:01:03.960 --> 0:01:08.000 That's the first draft, right because this is giving AI, 0:01:08.240 --> 0:01:11.440 you know it is it is. And I'm confused because 0:01:11.480 --> 0:01:13.399 I see other people that are like, I generated this 0:01:13.560 --> 0:01:17.080 using AI and it doesn't look like it looks very realistic. 0:01:17.319 --> 0:01:17.959 Uh huh. 0:01:18.000 --> 0:01:20.800 There's this black woman, her name is Jessica E Boyd, 0:01:21.800 --> 0:01:24.000 So she's on Instagram at Jessica E Boyd and she 0:01:24.080 --> 0:01:26.800 does these incredible AI images. It's just a great job 0:01:27.280 --> 0:01:30.520 and it looks like real afro texture hair, you know, 0:01:31.040 --> 0:01:33.920 but this one you created no shade. But how many 0:01:33.920 --> 0:01:34.920 fingers do I have? 0:01:37.240 --> 0:01:40.639 It looks like you've either got fourteen or six. 0:01:41.600 --> 0:01:42.479 I can't really tell. 0:01:42.520 --> 0:01:46.960 All the fingers are melting together, and I'm like, this 0:01:47.040 --> 0:01:47.360 is fun. 0:01:47.400 --> 0:01:50.000 I see people that have been making themselves, like not barbies, 0:01:50.040 --> 0:01:52.120 but like action figures in the plastic. 0:01:51.960 --> 0:01:53.880 I've been seeing that. I've been seeing that. 0:01:54.040 --> 0:01:55.880 I've seen people do themselves in animation. 0:01:56.280 --> 0:01:59.360 Yeah, the studio ghibli ones. Those are very cute. 0:01:59.680 --> 0:02:02.960 But I think like, no, don't get me wrong. I 0:02:02.960 --> 0:02:05.040 love all the latest technology, and you know, I've been 0:02:05.040 --> 0:02:08.280 out here trying everything okay, but sometimes I have to 0:02:08.280 --> 0:02:11.520 stop and wonder, like who's doing all this stuff? Whoever's 0:02:11.560 --> 0:02:13.760 doing it, or wherever's being done. It's just so far 0:02:13.800 --> 0:02:16.600 away from us, and it makes me think, like, what 0:02:16.720 --> 0:02:19.359 is the cost to generate these outputs? 0:02:19.840 --> 0:02:20.360 Honestly? 0:02:20.480 --> 0:02:23.200 And I think that's something that has been talked about 0:02:23.360 --> 0:02:26.160 a lot on social media. Most recently, I've been seeing 0:02:26.160 --> 0:02:30.480 people saying it takes sixty seven trillion gallons of water 0:02:30.680 --> 0:02:34.240 to just if you say hello, chat GPT, And I'm like, 0:02:35.400 --> 0:02:36.280 is that how that works. 0:02:36.360 --> 0:02:37.800 I don't know if that's how that works. 0:02:38.760 --> 0:02:41.560 I don't think that's how that works. But it's time 0:02:41.600 --> 0:02:42.400 to pull back the curtain. 0:02:42.760 --> 0:02:44.480 Let's jump into the recitation. 0:02:46.760 --> 0:02:49.840 So today we're diving into AI. We're gonna talk a 0:02:49.880 --> 0:02:53.720 little bit about automation, large language models like chat, GPT 0:02:54.400 --> 0:03:00.560 and their impacts, specifically their environmental impacts. So with what 0:03:00.680 --> 0:03:01.519 we know TT. 0:03:02.160 --> 0:03:04.680 So we know what automation is. 0:03:04.880 --> 0:03:09.119 That's the use of technology to perform tasks without human intervention. 0:03:09.360 --> 0:03:14.040 So it's often rule based, so like if then if 0:03:14.040 --> 0:03:16.680 I do this, then it'll do this or do this 0:03:16.880 --> 0:03:20.680 until and it's also repetitive, so it's able to keep 0:03:20.760 --> 0:03:23.760 going over and over and over again without need for 0:03:23.800 --> 0:03:24.399 a human. 0:03:25.320 --> 0:03:27.680 But I think a step above that or beyond that 0:03:27.880 --> 0:03:32.079 is artificial intelligence, and sometimes we see people conflating those two. 0:03:32.280 --> 0:03:35.480 Artificial intelligence is when you have systems that are designed 0:03:35.480 --> 0:03:40.080 to mimic human intelligence and so unlike automation, which is 0:03:40.200 --> 0:03:43.000 just a rule and it just every time it encounters 0:03:43.080 --> 0:03:44.760 this one thing, it does whatever you told it to, 0:03:45.160 --> 0:03:49.640 artificial intelligence is capable of learning and problem solving, and 0:03:49.680 --> 0:03:52.560 so it'll say, okay, if you put this input, you 0:03:52.600 --> 0:03:54.400 tend to like this, so the next time I see 0:03:54.400 --> 0:03:56.040 that input, I'll bring it up. Even if that's not 0:03:56.080 --> 0:03:59.520 a rule, it's a learning. It's able to expand its 0:03:59.560 --> 0:04:01.720 knowledge based on what the inputs you give it. 0:04:02.000 --> 0:04:04.680 Right, So like if you had a button factory and said, 0:04:04.720 --> 0:04:07.560 every time this piece of plastic comes here, poke three 0:04:07.600 --> 0:04:10.440 holes in it, and it does that, and it does 0:04:10.440 --> 0:04:12.720 it over and over again, that's automation. 0:04:13.440 --> 0:04:16.240 Now if you had a machine that was able. 0:04:16.000 --> 0:04:19.039 To say, based on fashion trends, I'm going to make 0:04:19.080 --> 0:04:21.320 this button out of wood instead of plastic, or I'm 0:04:21.320 --> 0:04:24.320 gonna make this button bigger because people started liking bigger buttons. 0:04:24.360 --> 0:04:26.600 Now that would be aim. 0:04:27.760 --> 0:04:31.280 Then you have large language models, and that's advanced AIS 0:04:31.400 --> 0:04:35.400 systems that are trained on vast amounts of data, particularly 0:04:35.440 --> 0:04:38.960 text data, to understand and generate human like language. So 0:04:39.040 --> 0:04:41.240 that's CHAT, GPT, Gemini, all of that. 0:04:41.640 --> 0:04:44.200 Yeah, they're calling the Internet and gathering all of this 0:04:44.320 --> 0:04:48.240 information and creating context around all of that information so 0:04:48.279 --> 0:04:52.200 that they're able to learn. The next thing is proprietary models, 0:04:52.600 --> 0:04:55.640 and what they do is they highlight that companies develop 0:04:56.120 --> 0:04:57.600 specialized models. 0:04:57.839 --> 0:05:00.760 So open AI they have. 0:05:00.720 --> 0:05:04.839 The GPT variant, so we know open AI, most commonly 0:05:04.880 --> 0:05:08.240 for chat GPT and chat GBT within it, that's just 0:05:08.279 --> 0:05:11.760 the big umbrella. There's multiple types of chat GPT, so 0:05:12.000 --> 0:05:15.960 they have four oh they have I think it's four 0:05:16.040 --> 0:05:19.840 and four mini. They also have deep research now, so 0:05:19.920 --> 0:05:22.400 it's tailored specific for a different task. 0:05:23.400 --> 0:05:25.280 So now that we all have kind of the same 0:05:25.360 --> 0:05:28.400 understanding of what these terms mean, let's talk about what 0:05:28.440 --> 0:05:29.160 we want to know. 0:05:30.320 --> 0:05:34.679 I want to know how the energy consumption of training 0:05:34.680 --> 0:05:38.400 these AI models and running these AI models compared to 0:05:38.440 --> 0:05:41.599 other technologies, like is it one to one or is 0:05:41.720 --> 0:05:43.120 AI the big bad wolf? 0:05:44.560 --> 0:05:47.000 And then like what are the implications? So like water 0:05:47.120 --> 0:05:49.919 usage is one that I've seen people talking about on 0:05:50.040 --> 0:05:53.640 social media, so about cooling data centers and all of 0:05:53.640 --> 0:05:54.920 this stuff. I was like, I thought we were all 0:05:54.920 --> 0:05:56.599 in the water cycle. So all the word is here 0:05:56.839 --> 0:05:59.960 is here, but I don't know we did an expert 0:06:00.040 --> 0:06:02.960 for that. I also want to know the mechanics. You know, 0:06:03.360 --> 0:06:06.080 my background isn't engineering all of those things that it's 0:06:06.120 --> 0:06:07.279 just interesting to me. 0:06:07.720 --> 0:06:11.239 So what processes occur when you're. 0:06:11.160 --> 0:06:14.760 Using a large language model like chat GBT or Gemini. 0:06:15.880 --> 0:06:17.839 We also want a future cast and the same way 0:06:17.960 --> 0:06:22.200 that we looked at what policies are around new technology. 0:06:22.400 --> 0:06:25.080 We thought about this in our Space episode. You know, 0:06:25.240 --> 0:06:27.119 we say, like, all right, now that all these people 0:06:27.120 --> 0:06:29.279 are going to space, who's making the rules, who's said 0:06:29.279 --> 0:06:31.919 in the speed limit? The same thing for AI? You know, 0:06:32.640 --> 0:06:35.680 who's making the rules, who's saying how much is too much? 0:06:35.720 --> 0:06:36.400 What do you use? 0:06:36.440 --> 0:06:38.400 What's the costs, what's you know, there's a cost and 0:06:38.920 --> 0:06:41.160 benefit for all these things, and so what is it 0:06:41.160 --> 0:06:43.640 for AI? It's out of site, out of mind really. 0:06:43.680 --> 0:06:47.240 Right, Well, I think that sets the stage perfectly for 0:06:47.360 --> 0:06:48.840 us to jump into the dissection. 0:06:53.200 --> 0:06:56.600 For today's dissection, we are talking to doctor Challet Wren. 0:06:57.120 --> 0:06:58.960 Doctor Wren, we wanted to start by kind of setting 0:06:58.960 --> 0:07:01.719 the stage about them and some measurement. You know, I 0:07:01.760 --> 0:07:04.200 think for all tech and innovation, there's a cost and 0:07:04.240 --> 0:07:07.719 a benefit, and often those costs are immutely presented in 0:07:08.360 --> 0:07:13.240 price and things you buy to maintain the technology. Sometimes 0:07:13.320 --> 0:07:16.440 it's subscriptions for software, Like I remember when I had 0:07:16.440 --> 0:07:19.720 a game Boy, it was the batteries. Okay, yeah, burn 0:07:19.880 --> 0:07:24.240 was like burning through batteries. For cars, it's gas, it's maintenance, 0:07:24.240 --> 0:07:27.600 it's inspections for emissions. But when we're physically removed from 0:07:27.640 --> 0:07:29.840 the technology, it can be harder to have a good 0:07:29.920 --> 0:07:33.120 grasp of like what it costs to run these things. 0:07:33.480 --> 0:07:37.720 How do we understand and appreciate costs now. 0:07:37.520 --> 0:07:41.920 So this is a huge energy consumption, and I mean, 0:07:41.960 --> 0:07:44.040 by the way, using energy itself, I think is a 0:07:44.080 --> 0:07:47.440 great thing because it just shows that we have more productivity. 0:07:48.360 --> 0:07:50.800 You know, usually energy consumption is a indicator of the 0:07:50.840 --> 0:07:55.320 economic activity, but the consequence of the using energy that's 0:07:55.480 --> 0:07:58.440 not always good. For example, the energy produces heat and 0:07:58.480 --> 0:08:02.000 we need water to take away the heat, and that 0:08:02.280 --> 0:08:06.080 consumes our natural resources. And also when we generate how 0:08:06.120 --> 0:08:09.360 we generate the electricity, how we generate the energy. Usually, 0:08:09.480 --> 0:08:13.640 even in the United States, a large fraction of the 0:08:13.760 --> 0:08:16.200 energy is coming from fossil fuels, and that comes with 0:08:16.360 --> 0:08:19.880 the carbon emission problem. Right. So this, you know, it 0:08:19.920 --> 0:08:23.480 has this long term impact on the climate change as 0:08:23.560 --> 0:08:25.040 many of the research that they have shown. 0:08:25.200 --> 0:08:27.680 I think that's a great point. What doctor Renna is 0:08:27.680 --> 0:08:31.000 saying is it costs even before we get to water, 0:08:31.080 --> 0:08:33.680 there's just energy. This is something I'm always talking about 0:08:33.679 --> 0:08:36.520 the people who drive electric cars. I'm like, Okay, you're 0:08:36.559 --> 0:08:38.280 not using gas, but what do you think how you 0:08:38.320 --> 0:08:40.679 think we get the energy that's coming to you. 0:08:40.640 --> 0:08:45.840 Right, That is one of the laws of thermodynamics. Energy 0:08:45.880 --> 0:08:50.600 cannot be created or destroyed. So we're not creating this 0:08:50.840 --> 0:08:53.440 energy out of nowhere and you can't get rid of it. 0:08:53.440 --> 0:08:54.680 It all goes somewhere. 0:08:54.720 --> 0:08:57.840 It's a closed loop like energy is flowing from one 0:08:57.880 --> 0:08:58.680 thing to another. 0:08:59.320 --> 0:09:01.600 And I think the other part that he mentioned that 0:09:01.640 --> 0:09:04.280 I didn't really think about for these data centers is 0:09:05.640 --> 0:09:08.840 the air quality. So PM two point five that means 0:09:08.920 --> 0:09:12.840 particular matter that is two point five micrometers or smaller 0:09:13.320 --> 0:09:17.679 in diameter. Now, biologically, when you're that small, you can 0:09:18.000 --> 0:09:22.880 get very deep into the lungs, okay, and so whatever 0:09:22.920 --> 0:09:24.920 you have can get into the lungs and it can 0:09:24.960 --> 0:09:30.000 potentially enter the bloodstream. Not always, but potential is enough 0:09:30.000 --> 0:09:33.760 for me to be like humhm, right, that's enough for me. 0:09:34.120 --> 0:09:35.679 That sounds scary. 0:09:35.800 --> 0:09:39.280 So the air pllutants includes particular matter two point five 0:09:39.440 --> 0:09:46.720 PM two point five and also nitrogen dioxide soultur dioxide. 0:09:46.920 --> 0:09:49.840 So these type of air pollutants can travel hundreds of miles. 0:09:50.280 --> 0:09:52.120 So even though we don't leave next to a data 0:09:52.120 --> 0:09:54.800 center or next to some power plants, we are the 0:09:54.960 --> 0:09:57.079 air pollutants that we breathe in could be coming from 0:09:57.120 --> 0:10:01.240 those places, and so this air just have a direct 0:10:01.280 --> 0:10:05.280 toll on people's health. They could create asthma, heart attack, 0:10:06.559 --> 0:10:11.000 lung cancer, and also even premature deaths. And there's been many, 0:10:11.000 --> 0:10:16.480 many research studies showing that causal relationship between these air 0:10:16.480 --> 0:10:17.880 pollutants and people health. 0:10:18.480 --> 0:10:22.360 But what about the water that's associated with running these 0:10:22.400 --> 0:10:23.320 AI models? 0:10:23.640 --> 0:10:25.920 Is it really as bad as folks are saying. 0:10:26.440 --> 0:10:30.120 Our study find based on the very limited data in 0:10:30.120 --> 0:10:33.120 the public domain, we show that if you have a 0:10:33.160 --> 0:10:38.280 conversation with a large language model, you likely consume about 0:10:39.400 --> 0:10:44.720 five hundred million water for ten to fifty queries, depending 0:10:44.720 --> 0:10:48.440 on where have you random model, And that's for the 0:10:48.520 --> 0:10:50.880 inference and also when you train the model, it's also 0:10:50.920 --> 0:10:52.680 consumed quite a bit of water. 0:11:07.280 --> 0:11:10.679 Okay, so we're talking to doctor Charlet Wrenn about AI 0:11:11.040 --> 0:11:15.839 and water usage, and doctor Rand said, basically, with ten 0:11:15.880 --> 0:11:18.640 to fifty queries, you use about five hundred millileaters of water, 0:11:18.679 --> 0:11:21.680 which is sixteen point nine ounces or a Deer Park 0:11:21.760 --> 0:11:24.600 water bottle. Now, I think there are two things to 0:11:24.679 --> 0:11:28.000 unpack their tt the water usage and the note that 0:11:28.040 --> 0:11:30.559 he said about it depending on where you are when 0:11:30.600 --> 0:11:32.960 you run the model. Let's start with the water usage. 0:11:33.040 --> 0:11:36.520 Okay, so before we get too deep into the water consumption, 0:11:36.600 --> 0:11:38.560 I really want to take a step back and like 0:11:38.840 --> 0:11:42.000 set the stage for everybody about what actually happens when 0:11:42.040 --> 0:11:45.200 you type something into chat GPT. When you put a 0:11:45.280 --> 0:11:48.560 prompt into CHATGPT, it travels through the Internet to a 0:11:48.679 --> 0:11:51.800 data center where that model. Remember we talked about those 0:11:51.800 --> 0:11:56.440 different specific models associated with different organizations, So open ai 0:11:56.559 --> 0:11:59.520 has the GPT models. So where that model is hosted, 0:11:59.559 --> 0:12:02.559 it goes to that data center, and then the servers 0:12:02.640 --> 0:12:05.839 process your input by running it through that model, which 0:12:05.880 --> 0:12:09.520 consists of layers of neural networks, and then the servers 0:12:09.679 --> 0:12:12.360 generate a response and send it back to your device, 0:12:12.440 --> 0:12:15.040 so your phone or your laptop or however you're using 0:12:15.080 --> 0:12:17.199 that AI model. 0:12:17.640 --> 0:12:20.240 And so it's a very fast process. 0:12:20.280 --> 0:12:22.520 If you've ever used any of these AI tools, you 0:12:22.600 --> 0:12:26.360 know it happens in seconds. So when we're talking about 0:12:26.480 --> 0:12:31.040 water consumption, I think knowing where that water comes in 0:12:31.120 --> 0:12:33.080 because it's like, what are we talking about here? Why 0:12:33.200 --> 0:12:36.840 is water associated with the Internet. Data centers use a 0:12:36.880 --> 0:12:41.320 significant amount of water to cool those servers down because 0:12:41.320 --> 0:12:42.920 they generate a lot of heat. 0:12:43.080 --> 0:12:43.960 We all know that. 0:12:43.880 --> 0:12:46.280 This technology gets hot when we're on our laptops for 0:12:46.320 --> 0:12:48.199 a long time, on a computer for a long time, 0:12:48.480 --> 0:12:50.600 when we're on our phones for a long time, they 0:12:50.640 --> 0:12:53.400 start to feel a little bit warm, right, And that's 0:12:53.440 --> 0:12:57.040 because you know, all of the technology inside is working 0:12:57.080 --> 0:12:59.839 and it's generating heat to do that work. And so 0:13:00.120 --> 0:13:02.560 what they do is they're running this cooling system which 0:13:02.679 --> 0:13:05.880 uses water to cool down these servers. 0:13:06.360 --> 0:13:10.360 When we cool down the data center facility, it involves 0:13:10.400 --> 0:13:13.199 two typically involves two stages. The first stage is moving 0:13:13.240 --> 0:13:16.320 the heat from the servers to a heat exchanger or 0:13:16.360 --> 0:13:20.400 to the facility. This process either uses air or uses 0:13:20.440 --> 0:13:24.000 a closed loop with some special liquids so there's no 0:13:24.080 --> 0:13:26.360 water loss, and there shouldn't be any water loss unless 0:13:26.400 --> 0:13:29.720 there's a leak, okay. And the second stage is moving 0:13:29.720 --> 0:13:32.880 the heat from the heat exchanger to the outside environment, 0:13:33.320 --> 0:13:37.439 to the atmosphere, to the sky and this process, of course, 0:13:37.480 --> 0:13:41.360 there are different options for moving the heat, but one 0:13:41.600 --> 0:13:46.960 Cammer approach is using water evaporation. And water evaporation just 0:13:47.559 --> 0:13:52.600 is the water that is temporarily lost into the sky, 0:13:53.280 --> 0:13:57.280 and that's what we call water consumption. So of course 0:13:57.440 --> 0:13:59.600 the water is still within our planet. It doesn't go 0:13:59.679 --> 0:14:06.800 to the sound whatever somewhere. Yeah, yeah, within our globe. 0:14:06.800 --> 0:14:09.320 But it's just you know, when the water comes back 0:14:09.400 --> 0:14:13.560 and where it'll come back is highly uncertain, especially in 0:14:13.640 --> 0:14:18.800 those route regions, and also even when the water comes back, 0:14:18.840 --> 0:14:21.880 it may not be in the in the freshwater state. 0:14:22.240 --> 0:14:24.040 But I think in the same way that they're using 0:14:24.080 --> 0:14:25.760 water to cool down a service. Isn't that the same 0:14:25.800 --> 0:14:29.960 way like basically air conditioning works and cooler and stuff 0:14:30.000 --> 0:14:30.440 in there. 0:14:31.280 --> 0:14:34.560 And the other thing is like the Internet been using servers, 0:14:34.600 --> 0:14:38.720 like AI using servers. That's not new. That's not new technology. 0:14:38.760 --> 0:14:43.080 We've always had servers. AI is just advancing and using 0:14:43.120 --> 0:14:46.920 those exame, exact servers that we're already being used for 0:14:47.040 --> 0:14:49.000 the Internet, like the same types of servers. 0:14:49.000 --> 0:14:50.440 They just have different models in it. 0:14:50.760 --> 0:14:54.760 But when you type in www, dot Dope Labs, podcast 0:14:54.800 --> 0:14:58.000 dot com, that information has to go to a server 0:14:58.440 --> 0:14:59.040 in order. 0:14:58.840 --> 0:14:59.920 To type it in right now. 0:15:00.120 --> 0:15:04.960 He tied it right now and click on stuff. Oh 0:15:05.080 --> 0:15:09.120 my goodness. Yeah, it's it's they're all doing the same things. 0:15:09.880 --> 0:15:12.240 I think one of the things that was really interesting 0:15:12.280 --> 0:15:15.840 to me that Charlett explained is that sometimes people are 0:15:15.840 --> 0:15:19.560 comparing apples and oranges because that water that's those servers 0:15:19.600 --> 0:15:21.240 are using is being recycled. 0:15:21.240 --> 0:15:22.359 It's being used. 0:15:22.160 --> 0:15:24.720 Over and over again. And they were comparing it to 0:15:24.760 --> 0:15:27.240 like eating a hamburger, but they were looking at the 0:15:27.280 --> 0:15:31.240 water to cow, the water to water the grass. 0:15:31.280 --> 0:15:34.760 Like a few weeks ago, the CEO of a leading 0:15:34.880 --> 0:15:38.080 AI company said, you know the study by the study 0:15:38.120 --> 0:15:40.200 on as water US. It was something made up by 0:15:40.240 --> 0:15:45.200 those anti AI crowd. But this, I would say, this 0:15:45.240 --> 0:15:50.120 is really misleading comparison because the Hamburg's water days quoted 0:15:50.320 --> 0:15:54.880 includes the rainwater in the in the soils to grow 0:15:54.920 --> 0:15:56.960 the grasses, to feed the cattle, to make the paddy. 0:15:57.040 --> 0:16:00.000 So that's really live cycle water using and it's mostly 0:16:00.160 --> 0:16:03.440 water which we call green water in the technical community. 0:16:04.520 --> 0:16:07.720 But the air water consumption is only for the operational 0:16:07.720 --> 0:16:10.160 stage when we actually random all, we're not talking about 0:16:10.160 --> 0:16:14.000 the water used for mining the rare earth or making 0:16:14.040 --> 0:16:17.800 the AI chapes or recycling the summer. So it's not 0:16:17.960 --> 0:16:21.640 life cycle for water for AI, and so it's a 0:16:21.640 --> 0:16:25.120 different scope of water comparison, different type of water. And 0:16:25.160 --> 0:16:28.920 the air water is mostly blue water, that's the water 0:16:28.960 --> 0:16:32.560 in the rivers, lakes, and groundwater sources that human can 0:16:32.600 --> 0:16:36.440 directly use. So I think those type of comparison are 0:16:36.560 --> 0:16:41.320 just showing that they either do not have the necessary 0:16:41.360 --> 0:16:45.320 knowledge in this space or they're having some other agendas. 0:16:46.520 --> 0:16:51.400 So yeah, those are not really scientifically correct comparison. 0:16:52.600 --> 0:16:54.200 Oh okay, I think I understand. 0:16:54.240 --> 0:16:57.920 So the water consumption is the water that doesn't make 0:16:57.960 --> 0:17:01.880 it into the drain and into our sewage. So that 0:17:01.880 --> 0:17:04.880 would be the water that in the washing your hands example, 0:17:05.119 --> 0:17:08.199 is being absorbed into your skin a little bit and 0:17:08.240 --> 0:17:11.600 then also absorbed in like a paper towel or a 0:17:11.640 --> 0:17:14.400 dish towel, or however you wipe your hands off after 0:17:14.440 --> 0:17:16.200 you wash your hands. I hope y'alla washing our hands. 0:17:17.640 --> 0:17:23.119 Yeah, So technically that's called water withdrawal, and the water 0:17:23.160 --> 0:17:25.640 you put it that goes into the sewage is called 0:17:25.720 --> 0:17:29.240 water discharge. The difference in water wizdow and water discharge 0:17:29.240 --> 0:17:33.359 for washing our hands is very minimum, so okay, the 0:17:33.359 --> 0:17:34.960 difference is called water consumption. 0:17:35.920 --> 0:17:43.520 So Challet did a study about how AI and human 0:17:43.520 --> 0:17:48.480 work compared to each other, about the efficiency and about 0:17:48.520 --> 0:17:52.560 the energy required to perform a task, and what they 0:17:52.600 --> 0:17:55.520 found was is that of course, you know, in a 0:17:55.520 --> 0:17:58.560 lot of situations, AI is more efficient. It can work faster, 0:17:58.640 --> 0:18:01.639 it can work longer. Know where humans require you know, 0:18:01.720 --> 0:18:06.439 lunch breaks and sleep. AI doesn't it's constantly running. But 0:18:06.520 --> 0:18:10.560 the energy consumption of AI is a lot higher, you know. 0:18:10.680 --> 0:18:13.680 So there's like pluses and minuses to both our studies. 0:18:13.720 --> 0:18:17.160 Not trying to say, let's AI is bad and that's 0:18:17.200 --> 0:18:20.560 not our purpose. Its AI is great. It's why why 0:18:20.560 --> 0:18:23.919 it's bad. It's it's using some resources. It's just like 0:18:24.040 --> 0:18:27.600 anything else, right, So we need to have an understanding 0:18:27.640 --> 0:18:30.119 of the cost. And turns out the water cost is 0:18:30.160 --> 0:18:31.919 just one part of the cost, and where there are 0:18:31.960 --> 0:18:36.840 also other costs like the climate, the energy, the public health. 0:18:37.200 --> 0:18:40.359 So we need to look at this, evalu the cost 0:18:40.440 --> 0:18:43.280 more holistically and think about what is the best way 0:18:43.320 --> 0:18:49.680 to support the technology. That's our purpose. Our research goal 0:18:49.920 --> 0:18:53.840 is not like hey, please stop using it now, No, 0:18:54.720 --> 0:18:59.359 we shure to use AI for good purposes and you 0:18:59.359 --> 0:19:00.640 know more response way. 0:19:01.119 --> 0:19:03.359 Can you talk about that a little bit more the 0:19:04.160 --> 0:19:07.679 cost to the environment, public health, so that people can 0:19:07.800 --> 0:19:09.080 understand what you mean by that. 0:19:10.280 --> 0:19:14.480 So because this is a huge energy consumption, and I mean, 0:19:14.520 --> 0:19:17.480 by the way, using energy itself is not I think 0:19:17.600 --> 0:19:19.960 is a great thing because we want I mean, it 0:19:20.080 --> 0:19:23.600 just shows that we have more productivity. You know, usually 0:19:23.680 --> 0:19:27.080 energy consumption is the indicator of the economic activity. So, 0:19:27.119 --> 0:19:31.320 but the consequence of using energy that's probably not always good. 0:19:31.359 --> 0:19:34.240 For example, the energy produces heat, and we need water 0:19:34.400 --> 0:19:39.800 to take away the heat, and that consumes our natural resources. 0:19:39.960 --> 0:19:42.960 And also when we generate how we generate the electricity, 0:19:43.080 --> 0:19:46.560 how we generate energy. Usually even in the United States, 0:19:48.200 --> 0:19:50.879 a large fraction of the energy is coming from fossil fuels, 0:19:50.920 --> 0:19:54.520 and that comes with the carbon emission problem. Right, So this, uh, 0:19:55.080 --> 0:19:58.160 you know, it has this long term impact on the 0:19:58.200 --> 0:20:15.080 climate change. As many of the research I've shown, I feel. 0:20:14.800 --> 0:20:17.119 Like so much of the conversation has been framed as 0:20:17.119 --> 0:20:20.280 an all or nothing is don't have any costs or 0:20:20.440 --> 0:20:22.920 use AI, And I'm like, what people don't think about 0:20:23.000 --> 0:20:27.479 is how many people are using Zoom, Google Meet, Skype 0:20:27.520 --> 0:20:32.400 if that's still operational. I don't know Microsoft teams and 0:20:32.880 --> 0:20:36.720 you know, when the pandemic hit, Zoom saw elite from 0:20:36.840 --> 0:20:40.479 ten million daily users to over three hundred million daily users. 0:20:41.040 --> 0:20:43.640 And there was a study from Purdue University. A group 0:20:43.680 --> 0:20:46.640 there said that an hour on zoom generates one hundred 0:20:46.640 --> 0:20:49.560 and fifty two one thousand grams of CO two. Nobody 0:20:49.640 --> 0:20:51.959 was thinking about that when we were trying to all 0:20:52.000 --> 0:20:55.119 stay connected and stay saying now turn off your camera 0:20:55.160 --> 0:20:57.560 can help with that is whether this study is in. 0:20:58.240 --> 0:21:02.119 But I think us considering, you know, there's a cost 0:21:02.160 --> 0:21:04.600 to all of these things that we often don't think about, 0:21:04.880 --> 0:21:08.960 and so I think it just becomes important to consider, like, hey, 0:21:09.080 --> 0:21:10.880 everything here has a cost. 0:21:11.040 --> 0:21:14.680 Right, and the cost varies depending on where you are. 0:21:14.800 --> 0:21:18.600 So the amount of water required for the AI chatbots 0:21:19.080 --> 0:21:22.800 using chat GBT four to generate a one hundred word 0:21:22.840 --> 0:21:28.000 email varies by location. The amount goes up depending on 0:21:28.040 --> 0:21:31.800 where you are, Like in Washington State, it'll require almost 0:21:31.800 --> 0:21:35.320 fifteen hundred milli liters of water, while in Texas it's 0:21:35.480 --> 0:21:38.760 only two hundred and thirty five. So we can't just say, oh, 0:21:39.000 --> 0:21:43.920 all of it requires all this water, and that's not 0:21:43.960 --> 0:21:45.119 necessarily true. 0:21:45.600 --> 0:21:46.440 Is it using water? 0:21:46.600 --> 0:21:49.199 Yes, in a closed loop, because the water we have 0:21:49.320 --> 0:21:53.280 is the water we've got, But also other forms of 0:21:53.320 --> 0:21:58.800 technology are also using water as well. It takes six 0:21:59.240 --> 0:22:02.240 nine hundred and nine liters of water to produce one 0:22:02.320 --> 0:22:07.240 pound of beef one pound of beef six ninety two liters. 0:22:07.720 --> 0:22:11.480 Training a chat GBT three has the same water cost 0:22:11.520 --> 0:22:13.560 as producing one hundred pounds of beef. 0:22:15.040 --> 0:22:18.720 So if one hundred people do meatless mondays yea, then. 0:22:19.280 --> 0:22:21.960 It's nearly double the amount and average American eats in 0:22:22.000 --> 0:22:24.679 a year. So the amount of water you're using to 0:22:24.840 --> 0:22:28.160 eat beef is way more than you're using to run 0:22:28.200 --> 0:22:32.240 a CHATGBT model. So in another comparison, is it takes 0:22:32.400 --> 0:22:35.280 nine and fifty six liters of water to produce one 0:22:35.320 --> 0:22:39.120 pound of rice because rice has grown completely submerged underwater. 0:22:39.240 --> 0:22:42.080 Requires a lot of water to grow rice. 0:22:42.480 --> 0:22:44.760 But we haven't been hearing nothing about that, or maybe 0:22:44.760 --> 0:22:46.920 people been talking about it, and I just don't know that. 0:22:47.040 --> 0:22:47.640