Episode 2 - Can AI Save the Planet? Technosolutionism, Climate Change, and Ethical AI Development.

Hi, this is Clamber Challenge, a podcast by the International PhD College of the University of Boronia on the Social Environmental Crisis. Welcome to today's episode of Climate Change, Climate Challenge, where we dive deep into the complex world of technology, especially larger in which models and the impact on our planet. LLMs, the driving force behind some of the latest advancements in smartphones and AI power devices, are not just about technology, they are also about the immense infrastructure required to support them. And SDD, the demand of LLMs, so does the need of massive data centers to house digital giants. And companies like Microsoft are at the forefront with the expansive facilities such as the one in Arizona highlighted by journalists carrying how recently, and these data centers are not only vast in size, but also in their consumption of resources. And in a world where summers are getting hotter, these centers, using innovative cooling techniques like evaporating drinking water to manage the immense heat generated by high energy workloads. And the rapid expansion of data centers rejects them to increase by 50 to 100 annually from an existing base of 300. And this growth raises pretty cool environmental and ethical questions like how sustainable is the rapid development of such infrastructure and what does it mean for the local communities and the environment. And some even say AI will save the planet so as we impact these topics we consider a proactive question. Can AI, so a technology contributing to environmental strain, also offer solutions to save our planet like some scholars say and what are exactly the connections between technology, ethics and sustainability in the age of AI. Today Dr. Tamarani's will be joining us. Dr. Tamarani's is a senior researcher and project director at the algorithmic impact methods lab at the data and society research institutes. And everything is colored at UC Berkeley Center for Science Technology Medicine and Society for the 2023-2024 academic year. She's also faculty member in the Mastering Design for Responsibility AI Program at PILI SAVA and Preservally Privacy Issues Serve Then Say Director of Developer Engagement on the Green Software Team at Intel and was actively involved with the Green Software Foundation's Policy Working Group. Dr. Tamarani's first book, Death Glitch, how Technosolucionism fills us in this life and beyond, explores the unforeseen long-term impacts of social media platforms aside of morning and memorization. In her research, we have been featured in outlets like Wire, Divergent, the Bethler, Spalts, Penhamist, Technologies, Pathful, Labour and Climate Activity Index. Dr. Tamarani's in Earned Her PhD from New York University, a Master's from the University of Chicago and a Bachelor of Fonkini in College. Her work has been supported by prestigious organizations such as the Melo Foundation and the American Council of the Society. So first of all, good morning. I thank you so much for joining us. The first question I had was what got you into climate activism in fact? Yeah, thank you so much for having me on the show today. So I came to thinking about sustainability through both my dissertation and book research, which were both focused on the planned obsolescence of technologies thinking about all of the e-waste and general waste that goes into the rapid production of technologies, particularly digital technologies and their hardware and infrastructures. And I began thinking about the need to pass on digital objects from one generation to the nest as a problem of reuse. So in the way that laptops that are falling into disrepair in one context say within a large corporation where they refresh all of their laptops or all of their desktop computers. A lot of those computers then have to be refurbished and modified so that they can be used in a secondary market. So by people who maybe don't have as much access to computing technology. And often those are people in the global south and other marginalized groups who are inheriting these second hand devices. And so looking at the problem of inheritance through the lens of death is also a way to get at this general problem of technologies that are meant to last only a few years then turning into something that should be passed down either from one group to another or from one generation to the nest. And the other way that I entered climate in technology in particular is through a lot of my organizing work as part of tech workers coalition and the other side of my research which focuses on labor practices in around the tech industry, especially platform labor. And so I've long been interested in the ways that tech workers find forms of solidarity and support each other across the supply chain and looking at the ways that tech workers can learn from and also form coalitions with activists in other realms and other domains so people outside of tech proper. And I was really interested in histories of tech worker organizing in relation to environmental justice which has been happening for many decades now at this point. So I see a lot of the current struggles that are happening both within the tech labor landscape and with our larger questions about the environmental impacts of AI, especially generative AI and other newer technologies. I think it's very important to situate these problems in longer histories which is the focus of a lot of my work. That's very interesting. Thank you. Also, I think that you were saying about this workplace culture in which there are tech workers that can self-organize what do you think about the fact that usually this happens but also it relies on unpaid labor and within these employee volunteer resource groups this kind of labor can affect marginalized community. For example, usually there is a lack of institutional support for AI ethics initiatives and then the burden falls on individuals to advocate for this kind of work. Absolutely. So the people who are doing work within large tech companies or other organizations that are focused on DEI initiatives or other justice oriented concerns and a number of large tech companies do have employee resource groups that are actively working on climate-related issues. That can be a boon to companies that want to kind of rely on these groups for marketing purposes and to point to them to say, "Hey, look, we're doing good things in the world." And yet those employees, one, are often among the first to be cut when there are light off, especially if they are working in "non-technical positions." And two, often they do not have any decision-making power within organizations. So as you said, they are often siloed. They are not necessarily able to actively implement the recommendation that they're making. And they're kind of expected to do this work on their own time. They're not paid for this additional labor so they're doing their full-time job. Then they may be doing volunteer work on top of it. And it can also be dangerous. The other thing that I observed, especially working within a large tech company, was that within the context of our queer employee resource group, because we have an evangelical CEO. There was a lot of concern from employees if they sent anything to controversial. So people were afraid to push certain lines or appear too radical because they thought that that might jeopardize not only their position, but also alienate the C-suite. And if you don't have the supportive C-suite for your employee resource group, then the work that you're doing is not going to be effective. So I saw that chilling effect up front within that context. And I have been witnessing it as an observer at other tech companies where people who are passionate about climate change and are very worried about the role of the tech industry in perpetuating climate change, many of them are in these sort of volunteer positions within their companies and are very upset about the fact that their employers seem to be investing in technologies that are going to be detrimental to the environment. And there are some organizing efforts within tech companies, but it is also very difficult right now because there is such a hype cycle around gendered VI and companies, clearly care mostly about their bottom line. And so having this sort of visible employee resistance can be a problem for companies and it may make people more of a target in the time of mass layoffs. and we've witnessed us also within the context of the pandemic. the note tech for apartheid, activism at places like Google, where employees are being fired outright, not only for participating in direct action, but also for just being on the periphery, being involved with organizing efforts at all. - Yeah, it seems that there is a big mismatch between people who actually do the decision-making and people who work and they just serve as like a facade for the company to do a green washing, but also we can call it like social justice washing. So make it seem that some activities are for the social good, but actually the only importance in this profit. And of course we can understand that, but at the same time, there are some efforts, but by some companies concerning climate activism and climate change are kind of lazy. So it seems that for example, of course, some tech companies now advocate for data transparency and they make it seem like data transparency is the solution to I don't know solving the impact of the technologies on the environment. And okay, some organizations have made initial efforts to work with transparency, but their commitments are still a bit lazy. So for example, the fact that model training can be considered to be scheduled at different times of the day and that could help maybe a little bit the environment. At the same time, this is not implemented. And also there is the fact that often regulation of mandate certain types of environmental impact measurements such as carbon footprint calculations, which we can see maybe on many websites of many companies or I mean, it's pretty no now, but some scholars have pointed out the limitations of this kind of metrics. So on the one hand, we have the fact that it seems like the companies are taking action and the other hand is action maybe not enough. What do you think? - Yes. So the idea of building tools for carbon awareness, I think when you're within a large-scale tech enterprise, people who are developers are going to focus on a narrow part of the problem most likely. And the idea of creating data visualization and telemetry in tools that will help inform developer decisions sounds really good in theory. And it also fits into a business model. So there's an ocean of this can be something that will be marketable is an idea with business value which is very different from saying maybe we should build systems that last longer so that we manufacture fewer things, which is what would probably need to happen in reality for our companies to be greener. But the focus on tooling makes sense within the context of a company. But as even people who are very much involved with the green software movement have pointed out, including people like Ishmael of Velasco, he and some others, a green web foundation, including Hannah Smith, were looking at the problem of carbon awareness and this idea of developers sort of deciding to train their models at a time of day when there's more renewable energy on the grid. And this doesn't really change the problem of the demand issue. And there's the sense of renewable energy being a ethereal and infinite resource, even though as we know, renewable energy also relies on infrastructures and produces its own kinds of e-waste and is limited. There is just not enough renewable energy infrastructure. And so the demand right now for things like generative AI and crypto, which is still a thing, is causing coal plants to reopen because of the energy demand. And so these tools, I think the idea is that at scale, they can make an impact. So if you tweak the amount of carbon being emitted by your model through using carbon awareness, carbon aware computing, then at scale, it can make a sizable impact. And so that is the theory. But there's also the question of implementation. And as you've said, a lot of these tools are not necessarily taken up with organizations. People are not using them. Individual developers who are really passionate about this may use this kind of tooling, but it is not necessarily something that companies are pushing from the top down. - Yeah, and the demand also is, I am not really sure that all the actions, even if made in good faith, will be able to be at pace with the demand. Nowadays, I don't know in the department store, and I see like the refrigerator and oven and every kind of appliance staying AI in it, even the toothbrush, which is, I mean, kind of crazy, it seems like it has become like a jolly term to say advanced or saying, but we are doing something that is kind of state of the art. But I'm not sure how much we actually need to have AI in everything, and the point is also that it looks like maybe some people will have the perceived need of things with AI because everything says that it has AI in it and something that is new. So how much do you think really need it also? Why is it so pervasive now like in everything that we buy basically? - Sorry, why is what so pervasive? It's a yeah, this kind of, I mean, apparently, yeah, sometimes it can mean many things. - Of course, of course, yes. Yeah, I think that it is right now being sold as the thing that will create more efficiency. It is, it reminds me of when people were really excited about the internet of things, and this idea of kind of networking every object imaginable. And nobody really talks about the internet of things anymore, even though in many ways integrating AI in everything is a form of that. But I think it just speaks to this desire to create efficiency in reduced friction. And there's a sense that if you add automation to existing systems or objects in the world that it will reduce the probability of human error, it will also make things more personalized. So another emphasis that I've seen in a lot of the LLM-related marketing materials for companies like Amazon, for example, there's an emphasis on deeper levels of personalization. And that is what AI will afford you. And so the reason that you would want your smart fridge, or the reason that you would want to integrate AI in your fridge is to have it cater to your preferences. And so there is the sense of optimizing your life and making yourself more productive and more efficient through the use of AI. And so I think there is also an attempt right now to create use cases for generative AI because people are really excited about it. And think that the models are quite powerful and they can do some pretty neat tricks. And so there's a desire to locate potential use cases. And it kind of reminds me of just if we think back, I don't know, a year and a half ago to the metaverse. And all of the attempts at integrating the metaverse into Microsoft Teams so that we could all be legless at the same time we were in our really boring video calls. I think this desire to find a reason to have the technology after the fact is why we're seeing this. And it happened with crypto, of course, as well. And so this is where we got all of the arguments that crypto was going to create the capacity for a financial inclusion and the global south. We heard that it was going to create more supply chain transparency. There are all of these kind of fantastical uses for blockchain that didn't really come into fruition. But I can't tell you within my job at a major tech company how many presentations I sat through about the future of the metaverse and what the future would look like. And all of it seemed to involve Web 3. And now Web 3 as a concept is very outmoded. Nobody wants to remember that. And a lot of the big crypto companies have now pivoted to AI. Yeah, that's very interesting. We'll see what's the next big thing. But usually when these claims are made, the other side of it is that there is the exploitation of workers for doing annotation work. So the way they can be maybe customized, then of course, maybe they can be customized because they get a lot of data and the data comes in. also from us giving the companies our data and allowing them to cluster us into some kind of profile is also that yes there are these workers that are doing annotation work and they are often very low paid and often this is also this can raise some errors so we we never think about where the data that this lm come from especially those that are maybe more domain specific ones and there are also workers that work on the more practical side of infrastructure such as coal mines and this also another another problem and it's something that raises also the concept of elemental ethics by this term coined by Sebastiana Laudet which interrogates the AI-valid chain's problematic relationship with the elements that make up the world for example the fact that we need to understand the environmental impact of AI and which is also a social impact through the lens of those most affected by its expansion for example also the local communities where these data centers are but it's I mean it's a thing that it would become very integrated in the urban tissue and these local communities are not really thought of maybe what do you think? Yes absolutely and that is part of the work that I hope to do in the near term future is in particular a situated study in Virginia which is the data center capital of the world data center alley and to do some disciplinary workshops with impacted communities where data centers are affecting their lives and to really think about what that means on a local level and in a related project at the upcoming fact conference which is going to be in Rio this year I am working with two colleagues Lori Regitiri and Bobby Racovah who both recently were senior fellows and trust for the AI at Mozilla and Lori is from Brazil and she invited some colleagues from the Amazon Environmental Research Institute to talk about the work that they're doing with Indigenous communities and thinking about what it means to implement things like remote sensing to protect against deforestation in the Amazon what does it mean to implement that technology in a local context with the consideration of data sovereignty thinking about Indigenous protocols and what does that look like look like on the ground and so we're going to have a tutorial session at fact and we're hoping that a number of AI researchers and ethicists and also people from the tech industry come to our session to learn from the people who are doing that work in the majority world because they're often does seem to be a kind of disembodied and generalized way of talking about AI it's just so different when you hear the rhetoric of people like Sam Altman who are thinking on a like a galaxy wide scale and have so much power and to then zoom into a local context and look at all of the various impacts that happen especially if you're trying to consider the supply chain which for general purpose technology can be so difficult to trace particularly as you are thinking about the relationship between extraction and mining and manufacturing let alone you know training and deployment and e-waste and the entire sort of life cycle of AI that that can be it is a global thing and so how do you kind of talk about this massive scale of a problem while still situating it in a local context yeah I see that in this speaking of conferences I don't see that recently there has been some effort I mean there are when I maybe have to write a computational linguistics article for I don't know an artificial intelligence conference usually there is a netit statement but there is never an environmental statement which is something that has not implemented that this is also beyond academia even though I find that in a cabine and how even in non-stem disciplines like I don't know humanity is art linguistics whatever literature llem is a hot topic and so more and more people have been using for example chat gpt to write innovative work while of course putting a straight on the environment so on the one hand that has been like some people have turned the blind eye to this problem and also not only academia but in tech and what do you think that what practical measures do you think that can organizations both in academia and in tech organization can be implemented not only to improve transparency but also ensure that it leads to meaningful accountability and ethical compliance in AI development yeah it's a great question I think to this issue of measurement and the creation of standards and that is something that different standards bodies so in the u.s. nist can be actively part of but what I'm finding really interesting about the the way that academia is trading l.l.m.s for example is the notion that students should learn how to use them and the idea that we should be integrating things like chat gpt into the classroom aside from the the arguments about plagiarism and you know cheating and all of the sort of intellectual property conversations and debates there is also this problem of why would you be asking university students to engage with something that you know does have a horrible environmental impact and why are we so enchanted with the tool itself and why are we willing to sort of suspend our critical lens for what in many respects is a parlor trick and even if you're approaching it in a critical way isn't necessary and that goes back to this idea of integrating AI into everything and so a little paper that I was working on recently with some colleagues that is part of a larger policy brief that we were writing is looking at the need to incorporate social science expertise in evaluations of AI and that this is a problem in tech companies and this is a major problem with not just the tech development process but then also the problem of evaluation and how evaluation is also often solely technical and through the lens of red teaming in a very specialized way within tech organizations and everything is on the or the auditing side everything is sort of on the technical side but calling for more stakeholder engagement and graphic work the kinds of perspectives that come from having a background in the social sciences and humanities that creates a very different framework for evaluating these systems and so trying to come up with a way to better integrate people who have these critical perspectives into the development process across the life cycle and to work on things like co-design with communities is just one aspect of the work itself would be perhaps one way of changing the conversation that we're having right now because my fear is that even on the policy and regulation side a lot of the focus is on coming up with different metrics that we just need to kind of measure how much carbon or how much water is being used and we slap the number on it and that will do something but measurement doesn't translate into action so even if we measure it effectively and we have the perfect numbers that doesn't necessarily mean that the harm will be reduced it also doesn't mean that behaviors will change and it doesn't capture all of the other downstream impact so for instance with data centers the fact that there are things like noise pollution that can really disturb people who live near data centers that is a secondary impact that is not going to be captured if you're just measuring the carbon cost of AI and so these are the kinds of considerations that do require a different form of socio-technical expertise instead of just a technical form of expertise. Yeah that's that's very important it's also why we need the participatory studies and of course sometimes this kind of evaluation it takes more time than technical ones but it goes more in depth than the technical one and it would never not be that way and sometimes we need to agree and come to terms with the fact that we cannot do everything very fast sometimes we need to take more time to go more in depth into some problems and take the time to involve people and keep people in the loop which is also very important when designing every kind of application and besides trying to change the conversation what argue recommendations for moving past techno solutionism in next climate labor and AI. And do you think, by some scholars, I've said that AI itself can have a positive role in it? - So I think the question of can AI be useful in climate action? So that was the kind of question that we wanted to get at with this fact tutorial with the researchers from Brazil looking at what it really means to use AI for a very specific kind of use case. I think in many cases, the application of AI to climate purposes is untested. So there are, of course, ways that the technology could be useful, but the trade-off between the environmental impacts of AI and its capacity to be useful and helpful, it's very hard to weigh those costs and benefits. If you don't engage in some kind of participatory exercise, if you're not really thinking about the problem from multiple angles and over time. And so I think part of the way that we get beyond technosolutionism is, and this comes out a bit in the policy brief that I just referred to earlier, but sometimes the recommendation should be that AI is not the solution. And so how can AI researchers point city governments or tech companies to resources or structural changes that are not just about the production of the in deployment of the technology itself? In this often does come out through participatory engagement. So you might have a workshop with people who are in theory going to be using this particular chat bot for some kind of sort of social good purpose. But then as they interact with it, a lot of questions come up about whether or not it is really useful to their specific context. And I do think having the input from communities, especially from the majority world is incredibly important. And in the work that we've been doing so far at AMLAB, a lot of that really has been our focus to think about who are the people who are potentially most negatively impacted by the technology or who is the expected user or who is the expected community to be part of this. And what are their perspectives? What are the concerns that they raise? What are the things that they wish would be better? Are there things that they, in question, is in concerns that they raise that are about something outside of the technology itself? Because I think one of the most harmful things about the way that these cycles of AI research policy and development changes can be that we imagine that just by tweaking the tech, if we tweak the algorithm and make it less racist, if we make the chatbot more benevolent seeming and kinder, that it will fix some kind of social problem or that you can stop there. And there may be changes that could happen, that have nothing to do with a sort of technical tweak that can be done on a more social level. So how do you flag that and signal that? And I do think that is maybe one benefit of having these participatory exercises around the specific use cases imagined for generative AI. Yeah, it's funny that you talked about the chatbot because I remember a study talking about maybe a chatbot helping with environmental dissemination or making people more aware of the algorithm and if we train GPT a little to say, "Oh, remember that you're wasting water using me or something." Is it really necessary to use GPT? For example, for dissemination or a chatbot and I can't, we use books or social media posts even as before. I mean, in the end, why do we need necessarily a chatbot for that or for many things? It's also funny. But yeah, the trade-offs are also not so clear. So probably we will see in the future but having more perspectives and remembering that bio-relibering is not corresponding to the actual truth of life and the world is important. So thank you very much for giving your perspective on this. All the links of the things and people we talked about will be in the description and also social media account will be in the description. So thank you very much for your perspective on this and for joining us today. Yeah, thank you so much for having me. It's been a great conversation. You have been listening to Clammer Challenge a forecasted by the International PhD College of the University of Bronya. You can find all episodes, all the references and additional information on the college in the short description. (upbeat music)