30th November 2022. That’s when OpenAI released AI chatbot ChatGPT and changed the course of AI, and life as we know it. From this point, a wave of attention and investment flooded into the AI sphere, with companies from all industries and sizes scrambling to implement Gen AI tools at mass.
Sadly, there are a number of societal risks associated with AI, such as its impact on employment or use by criminals, but an overlooked consequence, and one that particularly impacts the media industry, is the tremendous amount of energy being used to power AI models and the adverse consequences this is having on the environment.
The reality is that AI uses a heck of a lot of energy to power itself. A single ChatGPT query uses ten times as much energy as a Google search. When it comes to visual content, generating a single image can consume between five and fifty litres of water.
This should be a consideration for all businesses, especially as they aim to meet net zero targets, yet a number of challenges stand in the way of even the best intentioned company, including a lack of emissions transparency from major AI vendors, limited or no frameworks available to measure against, and a race to implement the technology to be competitive without considering future repercussions.
How are Media Companies Measuring Their AI Emissions Usage?
To put it bluntly, media companies aren’t measuring their emissions output when it relates to AI. But this is mostly because it’s incredibly hard to measure.
As businesses start to use AI in more ways, the ability to accurately measure usage becomes more tricky. Yes, you might be able to measure the energy output of 100 chatbot queries, for example. But how about when it comes to using your email automation tool, or your AI agent?
Rebecca Sykes, Partner at The Brandtech Group, believes measuring is only going to get more challenging as the technology leaks into every aspect of life. “What’s going to be challenging is all of the AI you don’t see coming. Right now, it’s quite a conscious choice to use AI. As it is increasingly built into every system, you become more of a passive consumer of AI. At this point, it will be harder to measure.”
Because AI is currently so challenging to measure, this is stunting the ability of businesses to meet their net zero 2030 and 2050 goals. According to Andy Power, founder of media ESG consultancy Legacy, it’s those companies that pave the way in measurement that stand to benefit the most.
He says: “The most proactive companies that report on AI will stand out as leaders. Strategically using AI can speed up progress without amplifying environmental and social risk, but only if there are the right policies in place.
“If media agencies and ad tech companies integrate their AI and ad tech emissions into the measurement of their scope 2 and 3, that gives them a start of how they’re disclosed. That sets a standard for the company in terms of who they work with, and gives them a benchmark to learn from year-to-year. Then it’s about promoting green AI practices, which includes looking at the most renewable models, and engaging in sustainable media buying. It’s about demanding carbon transparency from any of your AI vendors.”
Indeed, another reason, it’s so difficult to measure this output is the lack of transparency of energy consumption from tech giants and major AI players. One thing companies can do, is put pressure on their AI partners to be more open about the cost of powering models.
From Data Centres to Chatbot Queries
The Brandtech Group recently released a whitepaper about AI use in media companies and its environmental impact. The paper found, perhaps contrary to popular belief, that inference, beyond model training or data centres, contributes significantly to AI emissions. This paints a new picture about AI use – that users have more power than we might have initially thought in mitigating AI’s damaging environmental effects.
Sykes says: “A lot of the impact comes from inference – which is people using the models every day. You have a lot more control over that than you do over the training. So maybe the surprising feeling for me was that I felt more empowered, because so much of the impact happens at the time that I choose to use AI and the way I choose to use it.”
That’s not to say it isn’t the responsibility of AI companies to be more transparent about what the real numbers behind AI emissions are right now.
Lillie Ratliff, Head of Strategic Initiatives at Scope3, feels the responsibility ought to be shared across media businesses, tech companies, and government. She says: “Big tech companies certainly have a responsibility to understand the full impact of the technology that they’re rolling out – both from a consumer and environmental impact perspective.
“Within the marketing realm, brands, agencies, media owners, and publishers, all have a similar responsibility to understand the impact and be thinking about ways to be more effective. Governments certainly play a role too, but we can’t wait on regulation and reporting requirements to drive progress.”
Can Efficiencies Offset Damages?
Everyone knows that AI, above almost anything else, is an efficiency tool. Efficiency gains have often been associated not only with monetary gains or savings, but bringing about energy efficiency too. With this in mind, could we see the efficiencies brought about by AI use level out its own energy consumption?
It’s undeniable that reducing the time it takes to create something from three weeks to four hours involves spending less time on your laptop – which can have less environmental consequences. According to Sykes, there are some instances where it’s more energy intensive to not use AI tools than it would be to use them. She says: “I think the real insight comes from comparing a the impact of a task to it being carried out traditionally. It can be less efficient to create your PDP (Product Detail Page) with AI than it is to write it traditionally – a person sat at their laptop for an hour writing and editing a PDP compared to minutes with AI.”
Knowing when to use AI and understanding how to prompt efficiently is a major element of both driving down emissions, and using your AI tools more effectively. There is an obvious difference between prompting 20 times and prompting six times to get the same end result – these all add up in terms of energy consumption.
AI measurement can help firms get even more efficient, make their relationship with their tools better, which can be beneficial for business growth overall. In this sense, firms aren’t measuring AI use out of the goodness of their hearts, but for a tangible business case.
Guy Jones, founder of The GoodNet, suggests measuring sections of the ad or marketing process, instead of attempting to measure whole tools. “What we’ve been able to see through measuring billions of ad impressions is that you can actually drive improved campaign performance whilst also reducing your environmental impact,” he says. “We shouldn’t try and divorce the performance of a campaign from the different pieces of technology or components that are delivering that performance. Any kind of next step for this is about measuring in the ad round, which is not trying to measure the individual technology components.”
What are Practical Steps Media Companies Can Make?
Like with any aspect of the climate crisis, the weight of responsibility is placed on the laps of every party. When it comes to AI, businesses will be using the technology the most, and at scale. So it becomes even more crucial to not only choose the best AI models, but to also consider what models are eco-conscious or the ‘right’ model for you. This can have profound implications, not only from an ethical point-of-view, but from a cost-saving perspective too.
The most intelligent models aren’t necessarily the tools that will give you the best outcome – sometimes smaller models will give you a better experience and a more efficient output. For Sykes, choosing the best models also includes considering where its data centre resides – which is the biggest energy consumption factor for businesses. She says: “Data centre location plays the biggest part in how you can reduce your footprint. You can look at self-hosting models, and you choose the greenest infrastructure.”
It’s also important to start tracking from day one and build sustainability into your workflows now. It’s no use thinking about this in two years when the damage is done and it’s harder to upheave your existing AI processes.
Start Figuring it Out Now
One of the major points made by all the experts is that there is value in trying to figure out your AI energy measurement now. The role that inference plays in AI energy emissions is a signal to business leaders that they must adequately train and educate staff on what efficient AI prompting and use looks like, and then figure out what measurement looks like to them.
Sykes says: “AI is constantly evolving, and it’s happening so fast, so the more you can get ahead of what you think is going to be important and significant – and it’s my job to help businesses think about what is coming next – the better.”
The same sentiment is given by Ratliff, and that sharing use cases amongst peers is powerful too. “At Scope3, we open source all of our methodology so that people can see how we get our numbers, and we welcome and invite feedback,” she said. “Because climate science is changing, and the infrastructure and architecture of AI and ad supply chains are changing, we really want to be open, collaborative and learning together, and reinforce that it’s never too late to start doing the work to identify ways to grow safely and sustainably.”
The past year alone has shown that there is clear excitement for AI in the industry. However, clearly, we must balance this excitement with some concern for how AI is negatively impacting our company’s ESG efforts, and the planet. The issue, therefore, doesn’t necessarily lie in AI itself, but how we decide to train our workforce to use it. And then, how we choose to measure it.
