When AI is used to harm the environment – On the Green Fence

#artificialintelligence #ai #environment

Episode 2: When AI is used to harm the environment

Artificial intelligence has huge potential to help boost efficiency and reduce greenhouse gas emissions.

At the same time, the technology itself has a big carbon footprint and, in some cases, it’s being used to accelerate activities that make climate change worse. According to one US study, training a single AI language model can emit five times as much CO2 as a car over its lifetime.

The global AI market is currently valued at $142.3 billion (€133 billion), and it’s expected to grow to nearly $2 trillion by 2030.

So how can we shape these technologies to make them work, in a sustainable way, for our societies? How do we avoid the pitfalls? And what does green AI look like?

Interviewees featured in this episode:

Benedetta Brevini, associate professor of political economy of communication at the University of Sydney, author of “Is AI Good for the Planet?”
Anne Mollen, research associate at the NGO Algorithmwatch, Berlin
David Rolnick, assistant professor at the School of Computer Science at McGill University, Canada, and at the Mila – Quebec AI Institute
Xueying Wu, campaigner with Greenpeace East Asia, Beijing

On the Green Fence is produced by DW studios in Bonn, Germany.

Connect with us here:
Instagram – @dw_environment
Twitter – @dw_environment
Facebook – DW Global Ideas
TikTok – @dw_planeta
Website – https://www.dw.com/en/on-the-green-fence/program-49760682

Thanks for listening and subscribing!
https://pod.link/onthegreenfence

Chapters:

00:00 Intro
03:07 Benedetta Brevini says AI is exacerbating the climate crisis
05:53 What makes up AI’s carbon footprint?
07:31 Just training one model can create 284,000kg CO2e
08:10 Why we need to talk about AI’s impact
11:02 Anne Mollen says AI is not clean as it seems
13:22 Impact of data center infrastructure
15:45 Google, Microsoft and Amazon going green
16:50 Do we need such big AI models?
17:46 David Rolnick on negative AI applications
19:54 Greenpeace slams tech giants helping fossil fuel companies
23:13 AI’s growing role in society and economy
24:00 Regulation is key
26:52 The EU’s landmark AI Act
28:19 Governments working to regulate AI

On the green Fence I’m the Green hi it’s Neil here and you’re listening to on the green fence and this is part two of our mini series on artificial intelligence last time we heard all about the PO poal of AI to help us tackle some major environmental challenges from preventing deforestation to weather forecasting to cutting

Emissions in carbon intensive Industries and because artificial intelligence can be deployed in many different ways to help us make sense of data create products reach decisions and make Supply chains more efficient many businesses have embraced it it’s a rapidly growing sector the global AI Market is currently

Valued at1 42.3 billion US and is expected to grow to nearly $2 trillion by 2030 and uh this growth is relevant because well some types of artificial intelligence also use a lot of energy and maybe that’s easy to forget when we ask an online translator to translate a

Sentence for us or type in a destination on Google Maps or scroll through social media feeds these are simple actions that take us seconds but what we see on the screen the information we get is the result of an a I processing colossal amounts of data and that can require a

Huge amount of energy AI is playing a bigger role in our lives and economies and it could also be helping to push our carbon emissions upwards at a time when we’re supposed to be making drastic Cuts according to one study training a single AI language model so that it can perform

The tasks it’s supposed to can emit five times as much CO2 as a car over its lifetime we have to stop not considering the climate crisis as the major crisis that this world is facing at the moment and we have to put it first and in order to

Put it first we need to move Beyond this absolutely mythical understanding of Technologies there is this idea of AI being this clean technology just being produced by a handful of companies but it actually relies on the exploitation of people and and the environment AI is being used in negative

Environmental ways uh one report estimates that the oil and gas industry is expected to make half a trillion uh dollars in additional profit by 2025 thanks to Ai and advanced analytics half a trillion Dollars in this episode we’re going to be looking at how AI could harm rather than help the planet and the carbon footprint of the technology itself what does green AI look like and how do we shape these Technologies to make them work in a sustainable way for our Societies we know too well and especially the audience of this podcast that if we want to meet the Paris agreement Target of keeping the global warming below 1.5 in the famous threshold we need to cut emissions globally by 50% we don’t even have a decade to do that that’s benedetta

Bravini associate professor of political economy of communication at the University of Sydney and Senior fellow at the London School of Economics she’s also the author of the book is AI good for the planet we were Keen to get her on the podcast because she’s been closely following the rise of AI and the

History of technological development and she says she’s wary of what she calls mythical thinking surrounding artificial intelligence specifically the idea that AI can rescue us from the problems facing our society not least climate change we tend to think of this artificial uh deity of this artificial

Sublime hand that we have the ability to even surpass our cognitive functions as human right and if we start defining AI as Technologies machines and infrastructures we’re moving away from this mythical sub phase of AI right we’re moving away from this idea that it’s something that transcends reality right it’s something

Bigger than us these mythical ideas that impede us from actually really thinking about the materiality of AI the moment in which we understand these communication Technologies as in in their materialities as Technologies machines and infrastructures it is at that point that we realize also that they use scarce resources

In their production in their consumption in their disposal and this is where of course they exacerbate the problems of waste and pollution and exacerbated climate crisis whenever we’re developing a technology like this we never start breaking down the life cycle and production chain of AI and this is where

I wanted to bring back this conversation we need to if we consider AI as Technologies machines and infrastructure it means that we need to consider the entire production chain and all the environmental problems that are connected to this environmental chain and most notably energy consumption and Emissions but also material toxicity and

Also electronic Waste okay let’s have a closer look at what makes up the carbon footprint of of artificial intelligence creating training and then operating AI models requires computing power and that means energy training is the process of teaching the AI to interpret mountains of data and learn from it to complete a

Task accurately or make a decision based on that data for instance it can be tricky to put numbers on this because the amount of data AI uses depends on the kind of algorithm the machine learning model and for example whether those data centers storing the data AI churns through are powered by fossil

Fuels or Renewables small AI models can run on a simple laptop while big language models that use deep learning like the one behind chat GPT need a significant amount of computing power and then there are the emissions that come from manufacturing the Computing equipment the hardware that gets

Upgraded and replaced and the waste that results when it’s thrown away so there are a few things that need to be factored in here okay back to benedetta bravini when we consider a very basic training natural language processing model OKAY of AI so something simple think of Google Translate Okay so just

The training of this model to make sure that this model can work and can translate for us will entail and this data that have been published since 2019 by strubel ganes and maalum were based at massachusett armrest so this model so just one model of AI just for

The training of this model we need something like 284,000 kilogram of carbon emission now the first time I’ve read this data I was really shocked but at the same time I needed to make sense of them and so I tried to compare if you jump on a plane

From London to New York your carbon emission will be 9 986 kilog but to train one AI algorithm we emit 284,000 kilogram that’s just training it that’s not even using it right that’s just training it the training of it so shouldn’t we have a conversation how

Many times you know did you hear that you should reduce your travels how many times always right and and that would be the best way so why we’re never hearing and we’re never having a conversation about how do we reduce this carbon footprint of this training for example

Of an algorithm and there are ways to do it and we are working on it at this very moment but at the moment I think that what is crucial is to change the discourse and to finally acknowledge that if we don’t place the climate crisis at the center of this development

Of this most recent technology which is AI then we’re going to absolutely miss the chance to sort out the biggest crisis that we’re facing it’s also triggering new needs isn’t it every time we bring around this Innovation it’s it’s it’s almost as it’s pushing us towards AI because a lot of these

Systems that we’re going to put in place they have to be super flexible right especially if they’re going to be based on renewable energy so in a way it’s almost Paving the way for AI to come in because otherwise we won’t be able to deal with this without AI will we that I

Mean that’s part of the problem isn’t it absolutely and actually we need to really ask where are the advantages for us because we know that there are advantages for the corporations that are developing them and this is where a study of political economy becomes crucial because we need to understand

Who are the corporations that are currently controlling the development of AI we know in Europe especially we have been trying really hard right to have a conversation around AI but the problem is that the stakeholders that are taking mostly of the decision around the development of AI come from the industry

And the problem really is that we see the level of efficiency they come of economy of scale that they can achieve in a corporation by implementing AI but this type of deficiency are really leading to well-being are really leading to something again in the public interest considering also the climate

Crisis like all these new applications that we are developing in us are heavily reliant on AI and it is the structure of data capitalism that is leading to the necessity for more AI because we need it’s a cycle right we need more data we need to collect more data in order to be

Able to feed also the machine but we also need more data in order to push the cycles of this production chain so in a way it seems almost impossible to stop benedetta says she’s currently collaborating with Engineers to look at the entire life cycle of AI and come up

With recommendations to reduce the carbon footprint of data scientists practices that also means putting sustainability and climate concerns on the agenda right at the start in the design and algorithm training phase let’s bring in Anna Mullen here she’s a research associate at the NGO algorithm watch which is based here in Germany in

Berlin and she’s also calling for there to be more public discussion around when AI is necessary and what using it means for the environment there is this idea of AI being this clean technology just being produced by a handful of companies but it actually relies on the exploitation of people and and the

Environment basically with cat GPT we see that there’s been a lot of work on moderating and cleaning the data being outsourced to Kenyan sweat shops where people are working on a very precarious conditions just cleaning up the data set from the worst that we see on the internet because otherwise the chatbot

Would have been so much more toxic than it is the training of one large language model is just a fraction of its entire emissions so the development happens only one time the training happens a bit more often depends on the model sometimes it happens daily but it

Doesn’t have to and the inference so the application Phase that happens billions of time potentially there we only have very small numbers concerning the emissions we have low emissions but you know with it happening billions of times potentially the emissions still are significant and we have estimates that actually if you look

At the entire AI life cycle the application Phase accounts for 90% of the CO2 emissions so if we see already in the training phase for instance that emissions can be up to 270 tons seeing that the application Phase might be much higher we really have to

Come to a discussion of how much CO2 these big AI large language models emit we know that the entire data center infrastructure and data submission networks they account to two to 4% of the higher Global CO2 emissions this is not only AI but AI is a large part of

That so we really have to understand where this huge energy consumption comes from and how to reduce it because 2 to 4% is almost as much as the aviation industry produces so we have to come to a discussion here and I mean what happens if the companies that are

Pushing this all these tech companies I mean what if they sort of say well the problem we’ll solve it by just you know basing the entire ni systems AI systems on Renewables we’re going to do this with Renewables to 100% does that solve the problem then can we then it no it

Doesn’t solve the problem because actually one has to admit that big Tech is quite good with the renewable energy so big Tech has the power to build their own data centers and they are largely powered by renewable energies or they really make an effort to have them powered by renewable

Energies but on the one hand this brings another problem because Google for instance they like to go to very dry hot and windy areas where they can rely on wind and solar energy but at the same time these areas are usually quite dry but these data centers they need a lot

Of water to cool those service for instance so we know from examples in Chile for example that the renewable energy supply of Google’s data center in Santiago Chile is very good but at the same time local communities are actually revolting against the data center and against the construction of

New data centers because of their huge water needs and also coming back to your question just relying on renewable energies will not be enough because because we don’t have enough renewable Energy there are efforts underway to make AI A Greener industry Tech Giants are Keen to point out what they are doing to reduce their impact on the planet for example Google aims to be operating exclusively on carbon- free energy by 2030 Microsoft has pledged to be carbon negative by 2030 and Amazon

Plans to reach Net Zero by 2040 to reduce emissions companies might choose to operate or use data centers in regions that rely on renewable energy running these huge server farms in parts of the US or Australia where fossil fuels still make up a significant chunk of the power grid will produce more

Emissions than in say Iceland where geothermal power is a main source of energy or France where nuclear energy could be used to power computers Anna says open- Source AI models could also be helpful from a sustainability perspective because that means they can be reused and altered and

Don’t need to be developed and trained from scratch she also questions the trend of building bigger and bigger models which might make results more accurate but can be much more energy intensive there’s a very unreflected development happening when it comes to AI system they just built bigger and

Bigger and there’s just massive amounts of data just thrown at them without considering that there could be smaller data sets more qualitative data sets which could also serve the purpose that these companies try to achieve still the main Mantra of big Tech and also the machine learning research Community is

To build bigger and bigger models right Now But even if companies and developers work to shrink ai’s emissions there’s another way this technology could hamper efforts to fight climate change and we’re in danger of building essentially like certified fair trade landmines where in you making the algorithm is is very green but what the algorithm is

Being used to do is not very green we heard from from David rolnick in the last episode he’s an assistant professor in the school of computer science at McGill University in Canada and at the mea AI Institute he’s also a co-founder of the nonprofit organization climate change AI I’m concerned with the ways

That AI is being used really many many of the Innovations in AI are being directed at advertising systems which have an impact by deliberately shaping consumer behavior and being designed to increase consumption which assuredly comes with a very significant climate cost it may be difficult to quantify but I expect that is truly

Enormous various stakeholders including the tech industry want people to focus on the energy usage of AI instead of also thinking about for example the ways that AI is being used extensively by the fossil fuel industry by an autom advertising systems which are designed to make people purchase more the ways

That self-driven cars are probably going to make climate change worse and various other ways that AI is being used in negative environmental ways which are extremely Significant let’s look at a few more examples AI can make it easier to manage livestock at scale which could lead to an increase in cattle farming or as David mentioned the case of self-driving Vehicles which are more fuel efficient but could lead to more people in personal cars rather than on public

Transport and AI can help bring down the cost of extracting the fossil fuels oil and gas uh one report estimates that the oil and gas industry is expected to make half a trillion dollars in additional profit by 2025 thanks to Ai and advanced analytics half a trillion Dollars the environment NGO green piece has been highly critical of AI contracts between fossil fuel companies and Amazon Microsoft and Google in a report from 2020 it said companies such as shell BP and Exxon Mobile were using AI tools to expand their oil and gas operations reduce costs and in some cases boost

Production it said tech companies provided their clients with cloud storage for large amounts of data as well as sophisticated tools for analyzing modeling and understanding those vast data sets for example machine learning models that can make it easier to map and assess underground deposits of oil and gas shiing wo is a campaigner

With Greenpeace Asia based in Beijing and she says these contracts undermine the ambitious climate goals set by tech companies it is deeply concerning to see all these Tech Giants AI Technologies are aiding oil and gas companies so we think tech companies like Google and Microsoft they have like boosted

Extensively about their climate goals such as they have committed to powering their own operations with 100% renewable energies but when these tech companies collaborate with oil and gas companies it worked directly against these tech companies climate goals we do think Tech Giants in general must refuse to collaborate with oil and gas companies

In any form to stop this industry from growing and it’s more like generating a pile of trash from your own business but like dumping it in someone else rard I mean if we’re talking about energy how do we weigh up the value of deploying these Technologies you know I mean for

Instance to improve efficiency and reduce emissions in some Industries that’s one of the pro arguments you know balancing that against the carbon footprint that they have you know when is it worth it or at what point is the cost too high like I do think it is a

Very good sign that like there are companies are willing to develop Technologies to help our society to reduce emissions but you know like some companies are using this as a PR done to lead the public to believe that they have done enough in carbon reductions for examples companies like Microsoft

Google and Samsung they might say they have invested in like new software or like Technologies to reduce carbon emissions across their operations or Supply chains but the reality is that they have kept delaying using the most effective options to reduce these emissions like as we going to like 2023

Now like partnering with oil and gas sector in any way should not be accepted and also you know like the impact of the climate crisis are already too severe to be ignored by Us Google has since said it will no longer build customized AI tools for helping companies extract fossil Fuels artificial intelligence already has a place in homes workplaces City Transport networks social media and businesses just about every industry is using it in some way and with the sector growing rapidly that role is only likely to become more significant in the future we’ve looked at how this technology can

Help us improve efficiency and rein emissions and how it can also make our climate targets harder to reach if it’s used to expand carbon heavy activities so where does that leave us what can we do to avoid the possible pitfalls billionaire Elon Musk has raised concerns more than once about what he

Sees as the potential risks he’s even called AI dangerous technology that should be regulated by some sort of of authority to make sure it’s operating within the public interest the experts we spoke to for this podcast all said regulation will be key to manage how AI develops in the future here’s David

Rolnick Technologies don’t just have effects independent of our choices the way in which technology impacts Society is changeable so for example for self-driving technology self-driving cars personal vehicles are expected make climate change worse but if autonomous vehicle technology is used in public transportation that can have a positive

Impact on climate change the implicit choices which are being made by technologists by Regulators by members of the public shape how Technologies get used and the way in which the media portrays new technologies prioritizes discussion of them even by put discussing certain Technologies in a negative light that can make them more

Visible sometimes I think that what gets discussed and the way in which technology is portrayed if it’s portrayed as something that is an inevitability rather than something that we are all choosing in certain ways yeah and the particular kinds of applications that are discussed and the kinds of use

Cases and needs for for AI there are two things I’m taking away from that the one where you say that technology is not inevitable this progress we still have a choice how it evolves how we develop it what we use it for and also if I understood correctly that perhaps at the

Beginning of every decision that we have a mindset that is factors in the climate crisis um yes absolutely the normal way for developing a new technology at least within AI is to see what’s possible and see what’s called for by short-term Market incentives and what’s cool and attracts investment and then do

It and then Society plays catchup Regulators play catchup Regulators think about technologies that are 10 years out of date and don’t address the challenges of the Technologies now or the Technologies five years in the future that are foreseeable and generally don’t think of themselves as shaping the pathways down

Which technology could go in the way you might have a switch on a railroad we could go this way we could go this way they’re both good they’re probably both good for the economy it’s not that one way is the way to technophobia and and shutting down Innovation it’s really a

Question of what we’re prioritizing and getting in early and shaping those choices that are being made is really important here in the European Union lawmakers have proposed a landmark Bill to govern the use of artificial intelligence the legislation aims to classify AI tools according to perceived

Risk and it covers anyone who provides a product or service that uses AI AIS considered unacceptable would be banned outright while those deemed high risk would face heightened scrutiny the AI act could be passed this year something unen from algorithm watch is following closely it’s quite hard to work against The Narrative of

Competition in the AI economy between the US on the one side and China on the other side and Europe being kind of far behind so environmental impact or accounting for environmental impacts and trying to regulate environmental impacts is being seen as hindrance to the economic potential of the AI industry

But I don’t think that we have a choice here because we don’t have the data center infr structure to just build AI systems bigger and bigger so it actually I think it will be an asset to build more sustainable AI systems and it will become a necessity in the coming years

And we see kind of big Tech also being more transparent about the environmental impact so I think it’s a development that is just going to take place governments around the world are still working out how to deal with AI how to encourage an inov ation while avoiding potential risks upholding

Ethical standards and protecting citizens but keeping up to speed with such rapidly developing technology will be a challenge and if AI continues to evolve at this pace what can we expect further down the line where are we headed today’s AI tools are very good at performing specific tasks but will we

Eventually have artificial intelligence that is smarter than humans capable of thinking and learning independently that’s straying into the realm of Science Fiction but that’s what we’ll be focusing on next week in part three of our miniseries on AI we’ll be talking to a philosopher about what rapid AI

Development could mean for us as a species and the world we live in it’s bound to be an interesting discussion so do join us for that that’s it from me for now many thanks to my colleague and producer Natalie Mueller and my sound engineer thas Schmidt and a big thank

You to all our listeners for sharing reviewing and subscribing to on the Green fence my name is Neil king take it easy and take Care on the green fence