Generative AI's hidden carbon cost: What marketers should know

The rise and increased usage of generative AI tools like ChatGPT, Midjourney, and Stable Diffusion come with significant hidden environmental costs.

The development and use of these tools are energy-intensive, mainly due to data centres and GPU chips, which have carbon footprints far larger than traditional CPUs.  

The carbon footprint of these AI models comprises three key factors: training, inference, and the hardware and data centre infrastructure. Training these models is incredibly energy-intensive, with large models like GPT-3 consuming hundreds of tons of CO2.

So, how do marketers ensure that their work is environmentally sustainable when they use these tools? 

Laurent Thevenet, head of creative technology for APAC and MEA at Publicis Groupe, tells Campaign precise targeting will reduce the number of assets wasted in reaching the wrong audience. It means one can reduce the carbon footprint by prioritising quality over quantity. 

In addition, he says marketers can optimise the assets themselves more sustainably. Do marketers need a video when an image may do the job, or can they use more straightforward visual layouts to reduce the asset's size, Thevenet questions. 

"There are emerging conversations online about the colour black requiring less energy to be displayed with OLED screens, which more and more smartphones are equipped with. Dark mode is an excellent example of this, as it activates on some devices when low on battery," explains Thevenet 

"The technology stack that delivers content and experiences for a brand can also be sustainable. Some cloud hosting platforms already provide the choice of picking a green data centre when using their solutions. If a brand decides to use Gen AI, which consumes a lot of energy, it would be good to compensate with a green cloud solution." 

(L-R) Magda Griffiths, Tom Jones-Barlow, Laurent Thevenet and Brian O'Kelley

When sourcing sustainably, whether it's a generative AI tool or any other tool, three fundamental pillars apply, says, Magda Griffiths, senior product marketing manager at Microsoft Advertising. 

First, sustainable businesses find ways to boost revenue and business growth without hurting their community, environment or the health and well-being of their employees. Secondly, they can invest in products that positively impact the environment.   

"According to research from Axioma, companies with better environmental, social and governance standards typically record more robust financial performance and beat their benchmarks," says Griffiths. "In today's world, it's no longer a case of profits vs. planet; what's good for the earth can also be good for business,"

Thirdly, there is social sustainability, which promotes a good quality of life, encourages diversity, and provides equitable opportunities for all.  

In the business world, this pillar includes critical issues like health and safety, employee empowerment and inclusion, professional development opportunities, and work-life balance. People buy from companies that stand for something more significant than just what they sell, as values drive value.  

Griffiths says it begins with marketers shifting their company from product-centric to people-centric. Marketers must go deep into diversity to uncover what all people value, not just their intended audience. 

"Protecting the environment is a crucial objective for organisations prioritising sustainable development. Environmental responsibility drives climate action and establishes long-term sustainability goals for the organisation," explains Griffiths. 

"Marketers play a decisive role within their company to build trust and business value through purpose-driven marketing." 

Optimising content quality while tackling data waste

Generative AI enables the generation of diverse media content across various domains, such as video, music, and images. 

The tool can also streamline content editing processes, assisting with video editing, audio improvement, and image manipulation tasks. Streamlining content editing processes benefits media professionals by simplifying high-quality content creation.  

In addition, generative AI aids in real-time content moderation, enabling streaming platforms to detect and remove inappropriate or harmful materials swiftly. 

Streaming platforms now harness generative AI to deliver personalised ads and content recommendations that cater to individual viewer preferences and behaviours.  

By analysing extensive datasets, including viewing history, demographic information, and user feedback, media companies can curate tailored content libraries, significantly enhancing user engagement and satisfaction. 

However, analysing extensive datasets can result in data waste, such as pre-loading out-of-view ads that nobody sees. As marketers navigate the complexities of creating impactful campaigns, the method of adaptive streaming, which only streams data that the user consumes, has seen a rise in popularity. 

Adaptive streaming is seeing more adoption as embarking on a global data waste cleanup while maintaining quality is a win-win for all, reducing the Internet's data transfer and excessive carbon footprint without jeopardising the user experience. 

Thevenet says adaptive streaming is 'fantastic', as was responsive design before because it is essential to not over-deliver what the targeted device can use or display. 

He notes that responsive design has taught us to create experiences and content from the bottom up rather than the top down. Designers and creatives will always focus on the best scenario in which consumers see content. 

"This scenario often uses heavy assets, rich visuals, videos, etc. The problem is that the mass of people who consume content and use experiences are usually not doing that in the best setup envisioned by agencies," explains Thevenet. 

"The impact is that vibrant/heavy content may be sub-optimised for less favourable scenarios (a small TV, 3G network, low-end phone, etc.). If we start with the worst-case scenario, we may ultra-optimise bottom-up and help to save the planet." 

Tom Jones-Barlow, the general manager for Asia at SeenThis, an adaptive streaming platform, says whether driven by generative AI or simply the fact that people are consuming more video, millions of people are still coming online via mobile devices every year or other yet unknown developments.  

Consuming more video means data consumption will increase massively every year, consuming more energy and having a negative carbon impact. 

"It means that focus must be maintained on efficiencies, removing data waste and upgrading our infrastructure. Think of it like our water supply; globally, 90 billion litres of water are lost due to leaky pipes daily, an issue that must be addressed as the world faces greater water scarcity,' explains Jones-Barlow. 

"Our data use is the same - there's no point leaking data via downloaded ads that never get seen, downloading 100% of a video file that is only watched for a second, or delivering unoptimised creative." 

How do agencies and brands prevent data waste?

The call for digital sustainability extends beyond content creation to encompass data transfer and the Internet's carbon footprint. 

Agencies and their clients must find ways to reduce the waste of data and resources while ensuring that their content is still high quality. 

They can start by auditing campaigns and experiences, explains Thevenet, who notes that media agencies are already doing some of this. 

However, Thevenet points out it is not happening enough on the experience side. He says there are ways to calculate the carbon footprint of websites and apps, like what the Website Carbon calculator is doing. 

"I would also like our industry to adopt a digital sustainability score. Like audience analytics or general performance monitoring, it would be constantly up-to-date but focused only on sustainability tracking," adds Thevenet. 

The industry can also measure its progress by working with partners like Scope3, which sends advertisers their cloud computing bills or emissions reports. 

When Scope3 analyses this data, the platform often finds that if an advertiser uses a significant amount of AI, their bill and carbon footprint are substantial. Scope3 then break this down to quantify it on a per-request basis.  

For example, suppose an advertiser currently emits 0.001 grams per request and starts using AI more extensively, which leads to the increase of their carbon footprint tenfold. In that case, Scope3's model reflects this tenfold increase.  

Reflecting the increase alerts advertisers to the environmental impact, prompting questions about whether using a particular AI tool is justified, especially if it is not substantially improving performance. 

"Things like reporting, such as graphing my campaign results over the last six months, are now effortless. You can ask it to do so, and it gets done promptly. This streamlines analytics significantly," Brian O'Kelley, the chief executive and co-founder of Scope3, tells Campaign.

O'Kelley also suggests that advertisers can work with generative AI platforms like Memorable that use AI-driven insights to make ads more memorable and effective, helping brands achieve their advertising goals efficiently and reduce their carbon footprint by not resorting to quantity.

"Interestingly, humans pay a lot of attention to hands in ads. An ad featuring a hand, especially with a product in it, tends to perform better than something like a product just sitting on a plate," explains O'Kelley.  

"This kind of insight is built into their creative generation process, which is fantastic. They can even analyse a creative and provide feedback quickly. For example, if your ad involves someone touching their hair, it may divert attention from the main message. This rapid feedback loop is an excellent example of how AI can enhance effectiveness." 

Effective ads, it seems, are not only better for brands, but for the planet too. Unless they're selling environmentally harmful products, of course. Measuring sustainability, therefore will continue to a complex process, but nonetheless critical to consider.