Accedo – AI’s role in the OTT industry’s pursuit for sustainability

François Polarczyk, Sustainability Director, Accedo

Technology is set to play a crucial role in the fight against climate change by helping us to reduce greenhouse gas emissions, enhance energy efficiency, and promote sustainable practices. Is there potential for AI to also play a part in this? Google DeepMind certainly thinks so and is using the latest AI developments to help fight climate change and build a more sustainable, low-carbon world. But although AI has received a lot of attention since the launch of the large language model, ChatGPT, last year, AI and machine learning (ML) are not new concepts. Content creators, technology vendors, and service providers in the video industry have been using ML for some time. The difference now is that generative AI models have become more advanced, and are now being used by a wider audience. If organizations like Google DeepMind aim to use generative AI to fight climate change, can the video industry also use generative AI to optimize systems, create more sustainable consumption habits, and reduce the industry’s carbon impact?

But while there is much to explore in terms of how generative AI could support the OTT industry to reduce its carbon impact and become more sustainable, we also need to examine and consider generative AI’s carbon footprint. The use of AI has become widespread in an incredibly short amount of time, and while there has been a huge amount of debate around the risks and ethical concerns of using AI, there hasn’t been a lot of discussion about its environmental impact. And as with pretty much everything to do with sustainability, particularly around measuring environmental impact, it’s far from clear cut. Before we go on to consider AI’s carbon footprint, let’s first consider how AI could be used to help the OTT industry become more sustainable.

 Using AI to drive sustainability in the OTT industry

Already, we can see how generative AI could be used to optimize software engineering during OTT product development, to optimize content creation and delivery, and to promote more sustainable consumption habits. Firstly, generative AI can greatly enhance and optimize OTT software engineering by generating fitting code snippets which would speed up the coding process, as well as reviewing code to identify patterns and errors, and suggesting refactoring to improve efficiency of code to improve UX and reduce bandwidth. Additionally, AI can help developers to both document code, and to review documentation. Although AI-generated code and snippets still need to be reviewed and checked by engineers in case of faults, there’s little doubt that it is a useful tool that can be used to improve efficiencies in the video service development process.

Secondly, generative AI is also set to revolutionize how content is created. From a sustainability perspective, it can be used to support virtual production by creating virtual sets and backgrounds, reducing the need for physical set construction and minimizing resource usage. And AI algorithms could be used to predict the energy consumption of different production processes, to help content producers make more environmentally conscious choices during production.

Thirdly, AI-driven algorithms could potentially optimize content delivery based on user preferences, network conditions, and peak usage times. This reduces the need for excessive bandwidth and energy consumption, leading to more efficient distribution of video content. AI can also optimize resource usage within software applications, by enabling efficient memory management and processing, reducing energy consumption. Similarly, AI algorithms can be used to enable video services to better predict user demand patterns, so they can adjust server capacity accordingly, preventing over-provisioning of resources and reducing energy waste.

Lastly, more sustainable consumption habits could be promoted through AI-powered sustainability assistants. These AI assistants could help viewers understand the environmental impact of their different consumption choices, so that they can make more informed decisions.

Assessing AI’s environmental impact

But unsurprisingly, it’s not all roses in the garden. To train an AI model, colossal amounts of data are stored in data centers, which is then used to train a model using high powered, energy intensive GPUs. OpenAI reportedly trained the GPT-3 model on over 45 terabytes of data. Storing and processing this volume of data takes a huge amount of energy. According to Energy Innovation, some of the largest data centers require more than 100 megawatts of power capacity, which is enough to power 80,000 U.S. households.

Research carried out in 2019 into energy usage by four common AI models including GPT-2, found that training a model resulted in 284 tonnes of carbon dioxide equivalent. This is equal to five times the carbon footprint of an average American car, and its fuel, over the course of its lifetime.  Of course AI models vary massively in size, and as models become more advanced, energy consumption increases. A review of the energy use and carbon footprint of several recent large models including GPT-3 estimated that training that particular model consumed 1,287 MWh, and resulted in emissions of more than 550 tonnes of carbon dioxide equivalent – almost double the emissions that training an earlier model was estimated to have led to. And let’s not forget that it’s not just the training of these models that leaves a carbon footprint. The ongoing operation and use of these models also consumes energy.

It’s not just the data center’s electricity use and CO2 emissions that have an impact. According to Gerry McGovern, author of World Wide Waste, it’s thought that electricity probably accounts for about 10% of a data center’s emissions. Its infrastructure, particularly the data center’s cooling systems, is another big CO2 producer because of the energy it takes to supply those systems. The ‘water footprint’ also deserves a mention here because a massive amount of water is circulated around data centers as part of their cooling systems. One study estimated that training of GPT-3 in Microsoft’s US data centers would have used as much as 3.5 million liters of water. This figure includes the water used for cooling the data center, and also the water needed to generate the electricity that powers it.

So where does all this leave the OTT industry? Do the benefits outweigh the risks?

AI’s carbon impact in the video industry

 As the OTT video industry strives to become more sustainable, the continued integration of AI and ML into video workflows presents both promising opportunities and complex challenges. The fact that AI is so abstract and non-concrete means that it’s often overlooked when considering energy usage at different stages in the workflow. However, even when you recognize that when used by the video industry, AI does add to the industry’s carbon footprint, it’s not yet possible to accurately measure the CO2 emissions that incorporating AI into the various stages of the OTT industry will generate. For this, we need more transparency and improved carbon measuring tools.

While there’s no doubt that AI uses huge amounts of energy, and has a substantial carbon footprint, as data centers switch to using energy from renewable sources, it’s anticipated that this footprint will substantially reduce. The potential benefits of utilizing generative AI to drive efficiency, reduce carbon impact, and encourage sustainable practices across the industry are undeniable. Through optimization of software engineering, content creation, and delivery mechanisms, AI has the potential to revolutionize the way video services are developed, delivered, and consumed. To strike a balance between harnessing AI’s potential and minimizing its ecological footprint, the video industry needs to innovate responsibly and sustainably. This, along with improved means of measuring the carbon impact of AI, will be critical for ensuring a future where a sustainable OTT video industry thrives.

Search For More Content