What you need to know:

• Generative AI is blowing sustainability targets out of the water, with Google and Microsoft the latest to show massive emissions rises as a result. Locally banks, telcos and many other sectors piling into AI and AI-powered automated decisioning will inevitably see carbon footprints balloon as a result.
• Yet there is little discussion of the connection between AI and carbon emissions at the business level and even less at board level where mandatory reporting requirements have put ESG issues high on the risk register.
• Elsewhere, the programmatic advertising supply chain – highly carbon emitting due to the number of players involved in taking ad from advertiser to and user – is improving as global advertisers turn the screws on carbon-heavy digital placements.
• The side-effect is that streamlining supply chains tends to also weed out the worst performing ads.
• With progress happening on the media supply chain, Scope3 boss and AppNexus founder Brian O'Kelley is now pushing into the creative side of the equation. 
• And no, those 100 per cent renewable energy contracts hyperscalers like Google and Amazon say they have in place, and Microsoft says it will have next year, while worthy and necessary, don't solve the problem. While renewable energy reduces the carbon footprint, the overall demand for energy still increases.

Sure, you can generate thousands of images a day for creative executions using tools like Midjourney, Open AI's Dall.E, and Adobe's Firefly, but maybe you shouldn’t.

Each one may require the equivalent of a full iPhone recharge (and sometimes even more) to push out that picture of a six-fingered bunny rabbit juggling kittens. Meanwhile, if you have switched to generative search on platforms like ChatGPT you will be using an order of magnitude more energy than you would on Google for a comparable search. Goldman Sachs says that ChatGPT is 10x more energy-hungry, other estimates put it as high as 25x.

It's also very thirsty, requiring significant amounts of water for cooling data centres.

By 2030, when a lot of ESG and net zero promises are due, Gartner estimates that AI will be eating through 3.5 per cent of the world's total electricity pie.

It's not just generative AI. Other AI such as that used in automated decisioning is also chewing up energy, and given such systems operate at vast scale – Commbank says it generates 55 million next best actions a day – the carbon hit is dramatic.

It’s little wonder then that Scope3, the ad tech emissions company started by AppNexus founder Brian O’Kelley, will soon expand its attention to the creative side of marketing as new generative AI and other AI tools create the potential for an explosion of carbon emissions out of marketing and other customer-centred functions.

Marketers around the world are rushing into generative AI chasing productivity gains and creative flexibility. (As are publishers – Mi3 has a generative AI-powered news service.)

According to Salesforce, whose Einstein AI solution introduced many marketers to the power of AI, the most common use of generative AI among marketers is: basic content creation (76 per cent), writing copy (76 per cent) inspiring their creative thinking (71 per cent) analysing market data (63 per cent) and generating image assets (62 per cent).

The clear advantages of generative AI need to be weighed against the desires of consumers around the world who want the brands they support to act in a sustainable way.

A NielsenIQ survey in 2023 revealed that 46 per cent of consumers expect brands to lead in creating sustainable change, while Kantar’s Sustainability Sector Index 2023 underscores that consumers are increasingly aware of the environmental and social impacts of their choices.

Ultimately all of this ties back to revenue according to researchers from consulting firms like McKinsey and Company, and Bain whose data suggests a strong correlation between the ability to embed sustainability in brand values, and growth. Both firms suggest that brands can unlock significant market opportunities when they get the balance right.

And it's not like marketers are unaware of their role in the debate.

Even before the hockey stick growth of generative AI after the public release of ChatGPT in late 2022, marketers already saw themselves as sustainability laggards. A 2022 World Federation of Advertising report called Sustainable Marketing 2023 revealed that 74 per cent of delegates to the association's Global Marketer Week agreed with the idea that marketing as it was being practiced was not compatible with a sustainable future.

But no one is suggesting the AI toothpaste can or should be pushed back into the tube. Instead, sustainability advocates like O’Kelley say, “We need to be pro-progress, but pro-green progress.”

That explains why, as O’Kelley confirmed the move to Mi3 last week, “We are going to be doing a lot of work around extending what we do into creative. That’s where a lot of AI is being used.”

“I think by the end of the summer, we should be ready,” he says.

But wait, renewables?

Hyperscalers like Google and Amazon use 100 per cent renewable energy and Microsoft says it will get there in 2025. However, while that is good for encouraging the development of the renewable sector, it doesn't help at a systemic level. (The same goes for non-renewables like uranium for nuclear). That's because they are purchasing the energy, not producing it, and energy is a zero-sum game. Then of course there are lifecycle and indirect emissions to consider.

In fact, the hyperscalers have recorded huge emission increases due to AI. Google for instance is up 50 per cent since 2019, putting its 2030 targets are risk. Likewise, Microsoft's emissions are surging, up 29 per cent since 2020.

It's important to understand the distinction between matching energy consumption with renewable energy purchases and directly powering data centers with renewable energy. The hyperscalers achieve their 100% renewable energy goals by purchasing renewable energy credits (RECs) and entering into power purchase agreements (PPAs) with renewable energy providers. These actions ensure that an equivalent amount of renewable energy is added to the grid to offset Google's consumption.

And while they purchase renewable energy to match their consumption, their data centers also still rely on the electrical grid, which may include a mix of renewable and non-renewable energy sources. Therefore, not all the energy used by data centers is directly sourced from renewables.

AI is increasing the total amount of energy the world uses so the fact that the hyperscalers have contracts in place for so much renewable energy just means other organisations are buying a larger share of non-renewable. The net result is more carbon emissions in the atmosphere.

Fast track

Generative AI is on the fast track to adoption in marketing departments, especially for creative work.

In Australia, Mi3’s recently released Marketing Customer Benchmark report found that 44 per cent of B2C marketers are focusing AI efforts on content creation and 58 per cent of the B2B marketers, i.e. roughly half of the market.

According to the report, the adoption of Artificial Intelligence (AI) into marketing practices is significant – 67 per cent of all respondents acknowledged the use of AI within marketing. That incorporates all AI not just generative AI.

Furthermore, the bigger the budget the greater adoption of AI, with companies that have budgets over $50 million leading the way. B2B sectors show a higher rate of AI adoption compared to B2C.

When it comes to Generative AI in particular, a survey by global influencer marketing agency Billion Dollar Baby found that 91 per cent of marketers in the UK and Europe have already used generative AI to create content.

Furthermore, it found that almost three in four (70 per cent) marketers plan to increase marketing spend on creator content featuring generative AI in the next 12 months. But that uptake will come with a huge carbon hit

According to Natalya Makarochkina, Senior Vice President at Schneider Electric, the huge German manufacturer whose core business involves helping companies manage energy efficiency and sustainability, a simple comparison between a traditional search and a generative AI search – like ChatGPT – is instructive. “If I search on one word on Google I will use only 0.3 watts per hour but if I go to GPT4, it will use 10 times more energy consumption.”

What’s more we are already seeing an explosion in the number of large language models being deployed around the world.

“In the last 10 years, there were only about 25 LLMs created. But in the last two years, another 100 have been created – so this is really exponential.”

This order of magnitude impost on carbon emissions is often being driven within businesses by people who have little to no understanding of the energy impacts of different AI models – and the differences are huge.

Image generation for instance requires 62 times more energy than text generation, and 1,450 times more energy than text classification, according to a paper by Researchers at Carnegie Mellon University.

Alexandra Sasha Luccioni and Yaccine Jernite from Hugging Face, and Emma Strubell from Carnegie Mellon, – authors of “Power Hungry Processing: Watts Driving the Cost of AI Deployment?” – examined 88 models across 10 tasks and 30 datasets, encompassing both natural language processing (NLP) and computer vision applications. The mean of the energy consumption required for 1,000 inferences varied widely depending on the task:

• Text Classification: 0.002 kWh
• Image Classification: 0.007 kWh
• Text Generation: 0.047 kWh
• Image Generation: 2.907 kWh

By comparison, the energy used to charge an iPhone 10 is about 0.01216 kWh, according to ChatGPT which ironically used the equivalent of 4 full iPhone recharges to arrive at that conclusion. (The iPhone X has a battery capacity of 2,716 mAh (milliampere-hour) at a voltage of 3.81 volts, and charging efficiency is between 80 and 90 per cent for those interested.)

Generative tasks, such as text generation and image generation, were found by the study to be the most energy-intensive. In contrast, tasks that involved classification (both text and image) required significantly less energy.

The lack of visibility over the environmental impact of marketing’s use of generative AI is not surprising, says Lucio Ribeiro, a former lecturer in AI for RMIT and Deakin, and an executive with extensive experience in marketing and innovation at leadership levels in telco (Optus) and media (Seven).

“Sustainability might have come up in the conversations around marketing systems (though rarely) but it was never central to the conversations,” he said.

“ESG is a board issue,” he said, suggesting pressure for change will need to come from the top down.

It might be a while before that happens.

Mi3 spoke with directors representing half a dozen ASX companies including some amongst the largest ASX200 set. There is little discussion on the issue of the impact of AI and generative AI in particular and that’s likely to remain the case – despite increased pressure from regulators on ESG reporting –  until the use cases are firmed up.

According to Cheryl Hayman, a non-executive director at Silk Logistics, and a NED at AI-Media, the issue is not yet fully understood. “The sustainable piece is sitting mostly in agri and resources sectors and [there is] little decision or focus on it in other sectors.”

By the time the marketing fraternity finally wakes up to the potentially huge negative sustainability impact of their rush into generative AI, it may be even harder to understand that impact.

That’s because there’s a shift in the research community away from transparency as markets do what markets have done for thousands of years – enclose what was previously common, and stick a price tag on it.

According to the Carnegie Mellon paper, “While we encourage continued work analysing open source models, we note that the growing lack of transparency in model architecture and training details makes this line of work, alongside many branches relating to fairness and accountability in machine learning, increasingly difficult to carry out.

2 steps forward, 20 steps back

“We need to help people understand that there are ways to generate creative that is more efficient,"  Scope3’s O’Kelley. "There are different kinds of models. There are different ways we can route models around to make them more effective. Let's find the most efficient way to get great outcomes from AI.

“What’s really changed in the last year is that while adtech is getting better, you actually have AI going up exponentially in pretty shocking levels," he added.

“We are just entering the next phase of the conversation, which is ‘if adtech is 7 million tonnes of carbon. AI is 70 million tons. It's 10 times bigger.”

However, the good news is that the lessons learned from tackling sustainability in the adtech supply chain can be applied to the much more substantive problem of AI, he suggested.

“The real-time decisioning that we're doing with AI, the real-time model inference, it's exactly what we've been doing in adtech for 20 years.”

Real progress has been achieved in the programmatic supply chain in the last year thanks to things like the creation of the Global Alliance for Responsible Media (GARM) sustainability framework. O'Kelley described this as 'the biggest advertisers in the world saying we want this problem solved and we want to control how it's solved.'

“So instead of having a bunch of scrappy adtech companies inventing their own metrics, we now have a set of metrics and a framework where advertisers have said 'this is what we want our sustainability platform to do and measure'."

Progress, per O'Kelley, is now clear.

“What we've seen is that the activation side – agencies, marketers, adtech companies – are creating green media products that will actually block high carbon, block, made for advertising, block a bunch of the bad stuff. I think we're now over a $100 million run rate for those products. That's starting to be meaningful.”

“I'd like to see it be a billion dollars,” he told Mi3.

Eventually, O'Kelley suggested, every major marketer will have a sustainability platform. "Any publisher or any platform that says, 'I don't want to be bothered by this', it’s like saying 'I don't want to be bothered by brand safety'.

“If you're Elon Musk, you can say I don't want to be bothered – but that's why 12 months later, he's back at Cannes begging for forgiveness and rejoining GARM because it turns out the biggest advertisers in the world kind of matter.”

X rejoined GARM on 1 July. Of course, a week later, and true to form, Musk threatened to sue those very same 'biggest advertisers in the world.' Perhaps he's hoping hot air is a sustainable energy source.