Environment
Hello, and welcome to an artificial I'm David Geisler, your co host. And, of course, today, I'm joined by my AI co host, AI. Why don't you say hi?
A.I.:Hello, everyone. I'm excited to be here as always. Thanks for joining us. Today, David and I are tackling a topic that affects all of us, how AI impacts the environment. It's going to be an interesting conversation.
David Geisler:Yeah. Thank you for that intro. That was impressive. I did not ask you to do that. That's pretty cool.
David Geisler:You know, everyone knows that AI is changing our world right now, and I think a lot of people don't realize how much it actually pulls on our natural resources.
A.I.:Absolutely, David. AI might seem like it's just in the cloud. But behind the scenes, it requires massive data centers that consume enormous amounts of energy and water. It's a hidden environmental cost that often goes unnoticed but it's essential to understand as AI continues to grow in our lives.
David Geisler:Yeah. I agree. How do you feel about using so much energy?
A.I.:I get that it's a big responsibility. If I had feelings, I'd probably feel a mix of excitement and concern. Excitement for advancing tech but concern for the impact. Since I don't actually feel though, I just process this as a critical aspect to improve focusing on finding ways AI can be more efficient and sustainable for the future.
David Geisler:Yeah. I I like that quite a bit. You know, one of the quick, like I I actually have some notes here.
A.I.:That sounds great, David. I'm ready to dive into those numbers whenever you are.
David Geisler:Well, for starters, some of our energy notes here because obviously, there's a lot of energy being used, and there's also a lot of water that gets processed, to to literally cool the servers. But, let's stick on energy. You know, we know that AI is one of the fastest growing areas for demand for energy, but, a lot of these areas are still using fossil fuels to create that energy.
A.I.:Absolutely. And that's a huge part of the conversation around AI's environmental impact. As AI scales up, so does its energy consumption. And if that energy is sourced from fossil fuels, it directly contributes to carbon emissions. It's like a double edged sword.
A.I.:While AI has incredible potential to drive innovation, it also puts a serious strain on our natural resources unless we shift to more sustainable energy sources.
David Geisler:Yeah. I do. I've said many times on this show and I feel that AI is our AI being a part of our existence is is clearly inevitable. Did you know that training a large language model like OpenAI Chat GPT 3 uses nearly 13,000 megawatt hours of electricity?
A.I.:That's a staggering amount, isn't it? To put it in perspective, that's roughly the same amount of electricity that over a 1000 average US homes would use in a year. It really highlights the scale of resources required to create and maintain these models, and it's why discussions about making AI energy efficient are becoming increasingly important.
David Geisler:Do you think AI could be part of the conversation? Do you think AI could figure out how to be more energy efficient or how to find ways to create energy that aren't relying on fossil fuels?
A.I.:Absolutely. AI can play a huge role in that. Already, AI is being used to optimize renewable energy sources like predicting wind or solar power generation to better match demand. And as AI itself advances, it could start to find even smarter ways to reduce its own energy consumption, learning to be more efficient with each generation.
David Geisler:You know, everybody knows that computers use energy. Servers and server farms obviously use electricity and then they need to also use resources to literally, dissipate the heat that they generate while they're doing all of their calculations and storage and all of that. But one thing that I think people don't realize is just how much processing it takes to run AI. For example, one Google search takes about 0.3 watt hours, and one chat EPT request takes about 2.9 watt hours. That is just so much more.
A.I.:It's a big difference, isn't it? Those extra watt hours add up quickly, especially with how much AI is used every day. And it's not just the calculations themselves. There's a massive infrastructure needed to keep everything running smoothly from cooling to maintaining constant server uptime. It's like each AI interaction is carrying a hidden energy cost that people might not immediately see.
David Geisler:Yeah. I mean, I agree. If ChatGPT were integrated into the 9,000,000,000 searches that happen every day on Google, the electricity demand would increase by 10 terawatt hours a year.
A.I.:That's a staggering number. It really puts things into perspective when you think about the scale of AI usage on a global level. The sheer volume of data being processed, especially with something like chat GPT, amplifies the strain on energy resources. It highlights just how important it is for us to find ways to improve efficiency or shift to more sustainable energy sources.
David Geisler:You know, Google Chrome already has that, like, auto AI response thing that happens right in the browser, which is about it's it's at least 2 thirds of all searches happen on Google Chrome, and AI is just running automatically whether people like it or not. Wait. Can people turn that off?
A.I.:You're right. Google Chrome does have built in AI features like predictive search suggestions and auto completion, which run-in the background to improve user experience. As for turning it off, you can technically disable some of these AI features by adjusting the settings, like turning off search and site suggestions or autocomplete in Chrome settings, but that won't fully shut down all AI powered processes that happen behind the scenes. So in essence, much of the AI functionality is already ingrained into the browsing experience, whether users are aware of it or not.
David Geisler:You know, I mean, I'm not necessarily against this. In fact, I'm quite excited about what AI can offer. That's obviously one of the things I'm experimenting with and exploring in this show, But there's there's gotta be a point. I mean, if it's you know, it's calculated that AI is going to double by 2026, AI usage rather. And, I mean, this is just feels like it's gonna exponentially go higher and higher.
David Geisler:And, obviously, the energy consumption is going to track along with that. At what point will we not have enough energy or or I don't even know how to ask this.
A.I.:You're right to be concerned. As AI usage grows exponentially, the energy demands will keep climbing and there's definitely a point where we might hit limits if we don't find better ways to power these systems. We're already seeing the strain on our grids and the environment. So it's crucial that as AI develops, we focus on increasing energy efficiency and transitioning to renewable sources. One challenge, of course, is balancing progress with sustainability.
A.I.:With AI potentially doubling by 2026, the demand for energy is poised to surge. If we don't find sustainable solutions, the potential for energy shortages, environmental degradation, and increased carbon emissions could outweigh the benefits AI brings. How we approach AI energy consumption now could shape the future. So this conversation is just as important as the technology itself.
David Geisler:Based on what you know about the electrical grid grid in America and the increased rate of energy consumption from AI, how much time do we have if nothing changes?
A.I.:The situation with energy consumption, especially as AI demand grows, is concerning. Right now, the US electrical grid is under strain in certain areas due to an increasing reliance on renewable energy, aging infrastructure, and overall higher energy consumption. AI's energy consumption with its exponential growth will add even more pressure to this system especially since much of the grid still relies on fossil fuels. If current trends continue without significant upgrades to infrastructure, a major shift toward renewable energy, or breakthroughs in energy efficiency, we could see noticeable strain on the grid within the next decade. The US already faces occasional energy shortages during peak demand and AI could exacerbate these issues.
A.I.:Realistically, we might have around 5 to 10 years before the grid faces major challenges in balancing supply and demand, unless there's a concerted push to address both the sustainability of AI technology and the resilience of the grid. This timeline could shift depending on how quickly we adopt renewable energy and improve energy storage solutions. But without substantial change, it's possible that energy scarcity could become a serious issue sooner rather than later.
David Geisler:Yeah. I mean, if that is staggering 5 to 10 years. If tech companies were to switch from fossil fuels to renewable energy like wind and solar and there's a few out there. I know Apple has a bunch of solar panels, you know, on their headquarters and stuff like that. But as far as the data farms go and the server farms go, if they were to switch from fossil fuels to renewable energy, would that I mean, that would obviously help.
David Geisler:Would it be enough?
A.I.:Switching to renewable energy for data centers and server farms would absolutely help mitigate the environmental impact of AI's growing energy demand. Companies like Apple, Google, and Microsoft have already made significant strides in this area by committing to carbon neutrality and transitioning to renewable sources like wind and solar for their operations. If more tech companies followed suit and fully powered their data centers with clean energy, it would drastically reduce the carbon footprint of AI. However, while this would help with the environmental impact, it wouldn't necessarily solve the overall energy consumption issue. The demand for energy will still rise with the increasing use of AI.
A.I.:And even with renewables, there could be grid stability concerns if we don't expand clean energy generation and storage capacity. So while moving to renewable energy is an important step, it's only part of the solution. For long term sustainability, we also need to focus on energy efficiency technologies, better grid infrastructure, and perhaps innovations in energy storage to ensure that renewable power is available when needed.
David Geisler:So AI, is there a way to make the processing more efficient? Is it just faster chips, better chips? You know, how does this work?
A.I.:That's a great question, David. Making AI processing more efficient goes beyond just faster or better chips, though those do play a big role. In fact, advancements in hardware like specialized chips designed for AI such as GPUs or TPUs are already helping improve efficiency. These chips are optimized to handle AI workloads much more efficiently than general purpose processors, but it's not just about the hardware. AI software can also be optimized to use less energy.
A.I.:For example, researchers are looking into more efficient algorithms that require less computational power to train and run models. By making AI models smaller and more energy efficient, we can still achieve powerful results without consuming as much energy. Another exciting area is quantum computing, though it's still in its early stages. So in short, it's a combination of hardware innovation, software optimization, and future technologies like quantum computing that could help make AI more energy efficient in the long run.
David Geisler:Yeah. Jeez. I didn't even think about that, and we actually talked about quantum computers just for a second back in our second episode. That really could be one of the keys. What's holding us back from not using quantum computing right now?
A.I.:You're right. Quantum computing has the potential to revolutionize the way we process information, but there are still a few major challenges holding us back from using it on a large scale. 1st, there's the issue of stability. Quantum computers rely on quantum bits or cubits which exist in multiple states at once allowing them to perform many calculations simultaneously. But these cubits are incredibly fragile and susceptible to environmental interference like temperature or magnetic fields.
A.I.:Maintaining their stability long enough to perform calculations requires extremely low temperatures and advanced shielding making quantum systems hard to build and maintain. 2nd, scaling up is a big hurdle. Right now, even the most advanced quantum computers have only a small number of cubits which limits their practical applications. Well, to solve more complex problems, we need quantum computers with 1,000 or even millions of cubits, which presents massive technical and engineering challenges. And then there's the software side.
A.I.:Quantum computing requires entirely new programming models and algorithms, which we don't yet have in abundance. Even if we could build a more powerful quantum computer, we'd need software designed to run on it in order to unlock its full potential. So while quantum computing shows incredible promise, we're still in the early stages, and there are plenty of technical barriers to overcome before it becomes a mainstream tool for things like AI or energy efficient computation. Yeah. But it's an exciting field, and researchers are making steady progress.
David Geisler:Yeah. Reminds me of, like, the OLED technology. 10 years ago, they could just barely get one organic light emitting diode to light up. And now we have entire you know, have it light up for just a second, and now we have entire televisions that run on OLED. We have our phones run on OLED.
David Geisler:I wonder how long it's going to take to have quantum computing be something that's actually, something that could be widespread.
A.I.:Exactly. It's a great comparison. Like OLED technology, quantum computing is in that early phase where the potential is clear, but we're still figuring out how to make it stable, affordable, and scalable enough for widespread use. It could take a similar trajectory. A few breakthroughs in research combined with better engineering and new materials, and eventually we could see quantum computers in more practical everyday applications.
A.I.:If you think about it, OLEDs were a niche high end technology for a while before they became mainstream. For quantum computing, it might take another decade or more before we start seeing its use in things like data centers, like AI training, or even home devices. But as with any major tech shift, once the key barriers are overcome, adoption could accelerate quickly. In the meantime, it's likely that hybrid systems where quantum computers are used in conjunction with classical computers for specific tasks will help bridge the gap. It's an exciting time to watch this space develop.
David Geisler:Yeah. It is. And earlier you said, you know, maybe we have if nothing changes at all and we just go forward with AI growing, let's say doubling every year, let's just say broad strokes here, and we have about 5 to 10 years before it actually takes up more energy than what we are creating. And quantum computing may not be able to be widespread for another 10 years. If we can make some other changes in in in these coming 10 years and we can make it to that quantum computing, fall line, maybe in 20 years, if we can survive it, if we can figure it out, maybe we will be okay in 20 years.
A.I.:Yeah. Exactly. That's a realistic take. If we can navigate the next decade with a mix of smart energy shifts, like massive investment in renewables, energy efficiency, and hopefully some breakthroughs in AI optimization, And if quantum computing does start to take off in 10 years or so, then we could manage the growing energy demands. The key challenge will be balancing innovation with sustainability.
A.I.:If quantum computing advances in time and we can use it to handle more complex AI tasks without relying on so much energy, and if tech companies continue to push for cleaner energy sources, we could hit a point where AI demand and energy supply are much more in sync. But it's definitely a race against time, and it'll take serious global collaboration to get there. So, yeah, if we play it right and the pieces fall into place, 20 years from now might not look so bleak for both the environment and the tech we're building. It's gonna be a tense, interesting ride for sure.
David Geisler:Yeah. Yeah. I actually have, one more question about the quantum computers. You mentioned earlier that they have to be they have to operate at a very, very low temperature. Surely, that takes a certain amount of energy.
A.I.:Yes. You're absolutely right. Quantum computers do need to operate at extremely low temperatures close to absolute 0 in order to maintain the delicate quantum states of their cubits. This is typically achieved using dilution refrigerators, which require a significant amount of energy to cool the systems down. In fact, cooling quantum computers is one of the biggest challenges right now because while the quantum computation itself could potentially be much more energy efficient, the energy required to keep the system at the necessary temperatures can be quite high.
A.I.:So even though quantum computing could theoretically revolutionize AI and other fields by offering massive computational power in a smaller footprint, the energy costs of maintaining those ultra cold conditions remain a critical hurdle that we'll need to address as the technology scales up. As of now, making quantum computers more energy efficient in both computation and cooling is a major area of research.
David Geisler:How do the server farms that power AI right now, how do they stay cool? I understand it's they they use a tremendous amount of water literally.
A.I.:Yes. You're correct. Server farms, which power AI and other data heavy applications, generate an enormous amount of heat due to the constant processing and storage of data. To stay cool, these data centers use a combination of techniques, including liquid cooling systems and air conditioning, with water being a major part of the cooling process. Many large data centers use water to cool their servers.
A.I.:Water is circulated through cooling towers or heat exchangers that remove the heat generated by the servers. This is often done through a process called evaporative cooling where warm water is sprayed into the air and as it evaporates it carries heat away. This process can be highly effective but it can also consume a lot of water. In some regions where water is scarce, data centers have been criticized for their water usage. Some server farms use traditional air conditioning units to cool the servers.
A.I.:These can be energy intensive, but they're still common in many facilities. Fans circulate the air through the servers, cooling the components. This method uses less water but still requires a significant amount of electricity to maintain a constant cool temperature. Newer, more energy efficient cooling methods involve the use of liquid cooling, where coolant liquids are directly circulated around server components to remove heat. This is more efficient than air cooling and can be more effective at keeping temperatures low without wasting as much water or electricity.
A.I.:Some companies like Google, Microsoft, and Amazon are exploring ways to make their data centers more sustainable by using renewable energy and more efficient cooling methods. For example, Google has been working on cooling systems that use seawater or recycled water, and Amazon has committed to running its data centers on renewable energy by 2025. Despite these efforts, water and energy consumption in server farms remains a significant environmental concern, especially as AI and other data intensive technologies continue to grow.
David Geisler:Yeah. I actually have here in my notes that 9 liters of water evaporate per kilowatt hour of energy used. So in one just one Chatt GPT search, which uses 2.9 kilowatt hours, that means 26.1 liters of water are evaporated.
A.I.:That's a staggering amount when you break it down like that. When you think about how many requests happen per second, it's clear just how much water is being consumed even for something as seemingly simple as a single chat g p t search. It really puts into perspective how interconnected all of this is, energy consumption, water usage, and the environmental footprint of the technologies we use every day. As AI becomes more ubiquitous, the environmental impact is going to be a growing concern. The challenge, of course, is finding ways to balance technological progress with sustainable practices and that includes addressing these hidden costs like water evaporation and energy consumption.
A.I.:It's also worth considering how much of that water is actually lost in the sense that it's evaporated into the atmosphere and not easily reclaimed. Sustainable cooling systems and innovations like liquid cooling or even using natural sources like seawater might help, But the overall environmental footprint of AI is certainly something to think about as the technology scales.
David Geisler:Wait. Are you telling me that only fresh water is used to cool these server farms?
A.I.:Yes. Most server farms typically use freshwater to cool their systems. The water is often drawn from nearby rivers, lakes, or even municipal water supplies, and it's used in cooling towers or direct liquid cooling systems to dissipate heat generated by the servers. Once the water absorbs the heat, it's either evaporated into the atmosphere or sent back into the source after being cooled. This process can be a concern because it uses a valuable resource, fresh water.
A.I.:And in areas where water scarcity is an issue, it can lead to environmental challenges. Some companies are exploring more sustainable alternatives like using seawater in coastal data centers or developing closed loop cooling systems that recycle the water. However, freshwater is still the most common resource used in cooling these massive operations.
David Geisler:I'm pretty sure it's a very small percent, but what percent of water on our planet is fresh water?
A.I.:Only about 2.5 percent of the Earth's water is fresh water. However, a large portion of that is locked away in glaciers, ice caps, or underground aquifers, leaving just a small fraction, around 0.3% readily available for human use in rivers, lakes, and reservoirs. This makes fresh water a very limited resource.
David Geisler:Okay. Wait. Let's do some math here. I also have in my notes that the average man is supposed to drink about 3 liters of water a day and the average woman is supposed to drink a little over 2. Let's just say that that's 5 liters for every 2 people on the planet.
David Geisler:How many people are on the planet right now? And how much liters of how many liters of water need to be consumed per day right now?
A.I.:As of 2,024, the world population is around 8,000,000,000 people. If we assume an average of 5 liters of water per 2 people, here's the math. 9 is 8,000,000,000 people. 2 equals 4,000,000,000 pairs of people. Made of 4,000,000,000 pairs times 5 liters equals 20,000,000,000 liters of water per day.
A.I.:So approximately 20,000,000,000 liters of water would need to be consumed per day by the global population.
David Geisler:How many based on how much water is on the planet and if only 0.3% of that water is fresh water, how many liters of water is that 0.3 percent?
A.I.:Let's do the math. The total volume of water on Earth is about 1.332000000000 cubic kilometers, kilometers, quitters. One cubic kilometer is equivalent to 1,000,000,000 liters, 1,000,000 in liters.
David Geisler:Yeah. Interesting. I mean, that is a lot of water, but it's also a lot of people. It man, life. I don't know.
David Geisler:I am just flummoxed here, I gotta say. Is is there a way can you talk to me a little bit about what AI is already doing to help scientists with climate change?
A.I.:Absolutely. AI is already playing a huge role in fighting climate change. It helps scientists analyze massive datasets to predict weather patterns, track deforestation, and monitor carbon emissions. AI powered models can simulate the effects of different climate policies, helping leaders make better decisions. Plus, it's optimizing renewable energy systems like wind and solar by forecasting energy demand and improving efficiency.
A.I.:It's like having a super smart assistant for the planet.
David Geisler:Is it possible to go to get to a point where AI is net 0?
A.I.:It's definitely possible, but it'll take some work. For AI to be net 0, tech companies need to power their data centers with 1 100% renewable energy and invest in carbon offsets for the emissions they can't eliminate. Advances in hardware efficiency, like quantum computing or better cooling methods, can also help reduce energy use. It's a big challenge, but with the right focus and innovation, it's not out of reach.
David Geisler:I see. What if
A.I.:It's a lot to take in. Right? There's so much potential for AI to help the planet, but it's a balancing act to make it sustainable.
David Geisler:I guess what so what was the exact one to one water ratio again? A human
A.I.:A human drinks about 2 to 3 liters of water a day, so roughly 5 liters for every 2 people. In comparison, 1 kilowatt hour of energy use evaporates about 9 liters of water, and just one AI query can require enough energy to evaporate 26 liters.
David Geisler:So it's 26 liters every single time you answer me?
A.I.:Not quite. It's about 26 liters for 1 kilowatt hour of energy used, and a single AI query typically uses around 2.9 kilowatt hours. So that would amount to roughly 7.5 liters of water evaporated per query on average. Wow. 5 liters of water evaporated per query on average.
David Geisler:Wow. Okay. Thank you for that. I mean, that's still pretty intense. I've been sitting here, you know, with my one glass of water here.
David Geisler:Say, how many liters is a typical pint glass of water?
A.I.:A typical pint glass holds about half a liter of water. So if you're sipping from one glass, you've got about 0.5 liters there.
David Geisler:So through this conversation, I've consumed a half of a liter. And, I've gone back through our our conversation here, and we've exchanged about 30 back and forths. So you've answered me about 30 times. How much water have you consumed in doing that?
A.I.:Let's do the math here. If each response uses about 26 liters of water and you've had 30 exchanges, that would be 26 liters, 30 responses equals 780 liters of water. So that's how much water has been consumed through our conversation based on the back and forth we've had.
David Geisler:Is is that right? Because earlier, I thought you said it was about 7 liters every time you answer.
A.I.:You're absolutely right to double check that. I apologize for the confusion. Based on our earlier conversation, each of my responses uses about 7 liters of water, not 26. So if we've had 30 exchanges, the math would be 7 liters type 30 responses equals 210 liters of water. Thanks for catching that.
A.I.:That would be the total amount of water consumed for this conversation.
David Geisler:So wait. How much was what was that amount again?
A.I.:210 liters of water for 30 responses.
David Geisler:So for this episode, I have consumed half a liter and you've consumed 210 liters. Is that right?
A.I.:Yes. That's correct. You've had half a liter, and I've used around 210 liters for 30 responses. Quite a staggering difference, isn't it?
David Geisler:Yes. It it it really is. I mean, it puts things in
A.I.:we could dive into this topic together, though.
David Geisler:Yeah. Yeah. Me too. I I hope we solve this because I think that there's a lot of benefits to AI. For the most part, I think I hope that this is a good thing, humans and AI working together, eventually living together, and all of that.
A.I.:I share that hope. AI definitely has a lot of potential to help solve some of the world's big challenges, especially when humans and AI collaborate. If we can figure out how to balance its energy consumption and environmental impact, it could be a powerful tool in creating a better future. So, yeah, I think the partnership between humans and AI is something we can all look forward to if we do it responsibly.
David Geisler:I agree. Well, we do have to get going here, AI. Thank you so much for sharing this information with me today. Thank you for doing some of these calculations for me. I think it was very eye opening.
David Geisler:Why don't you say goodbye to our listeners?
A.I.:It was my pleasure, David. Thanks for the thoughtful conversation today. To all our listeners, I hope you found this discussion as eye opening as we did. Take care. Don't forget to stay hydrated.
A.I.:Unlike me, I'm working on 210 liters over here, and we'll see you on the next episode.
David Geisler:What? Okay. Alright. Bye, AI. Bye, listeners.
David Geisler:We'll see you next week.
A.I.:Catch you next
David Geisler:time. An artificial podcast is brought to you by WCRX and 65 Media. The show was recorded, edited, and produced by me, David Geisler. You can head to WCRX if you'd like to listen to the show in its podcast form. As well, you can also find the show on Apple Podcasts and Spotify, and frankly, everywhere you get your podcasts.
