The hidden carbon costs of automation and digitisation

Credit: 

It is difficult to imagine the extent of greenhouse gas emissions when they come from a source so seemingly disconnected from physical pollution. When watching fumes trail from aeroplanes as they fly through the sky, or from the exhausts of cars, or the cooling towers of power plants, it is easy to understand their harm to the atmosphere. But the emissions produced from collecting, transferring, using and storing data, abstract as it is, are more difficult to picture, and arguably more difficult to quantify.  

And yet they exist, in huge quantities that grow, seemingly exponentially, as all aspects of society and industry gradually become digitised. Estimates suggest that worldwide, data will exceed 180 zettabytes by 2025, almost double that seen in 2022 and the equivalent of more than 6.8 billion years of continuous high-quality Netflix streaming. Decisions made on the visual appearance of websites and apps, even down to the colours used, can impact the extent of pollution produced when digital spaces are used. 

The mining industry has, in recent years, begun to digitise and automate physical operations at speed, often in the name of efficiency and sustainability. Recent studies have highlighted the ways in which AI and enhanced digital processes can improve the sustainability credentials of industrial activities and significantly aid the achievement of the UN Sustainable Development Goals. But the invisible environmental cost of producing and storing data in such high quantities is still often overlooked and could be negating efforts to decarbonise physical operations. 

Researchers at Loughborough University in the UK recently developed the world’s first publicly available tool, known as the Data Carbon Ladder, which allows companies to calculate the carbon footprint of their digital activities.  

Mining Technology sat down with the creators of the Data Carbon Ladder, Tom Jackson, professor of information and knowledge management, and Ian Hodgkinson, professor of strategy, to discuss the ways in which their tool can be used by businesses and the future of sustainable data management. 

Tom Jackson: Digital Decarbonisation was born back in February 2022. We started thinking, “there must be some CO₂ [carbon dioxide] that comes from digitising everything”. Data centres, of course, need energy to run. To store the data, to move the data from across the globe and within data centres, and also the compute time. For example, if you think about chat GPT, you need a lot of CPU time to do that. 

But we have only focused on the energy costs of digitalisation from data networking and computing. We haven't looked at things like the cooling of data centres, which we know is also an issue. Then we asked: how bad are data centres compared to, for example, the aviation industry? Data centres account for 3.7% of global greenhouse gas emissions (GHGs), in the worst-case scenario.

Ian Hodgkinson: It sits somewhere between 2.5 and 3.7% of total greenhouse gas emissions. That's not our research, that is numbers that are already out there. Regarding aviation, depending on the report [you use], that figure is between 2.1 and 2.4%. We focus on aviation because it’s clearly polluting, you can see it in the sky. Data is intangible, you can't see it, so often nobody even asks the question: what's the impact of this data that we're creating?

Currently, there’s an awful lot of data being generated and there's a lot of digital waste. This can take the form of ‘dark data’, which is data that's used once or never at all, or ROT data, which stands for redundant, obsolete or trivial data, and this is data that is kept even though it has no value.

In terms of the energy sector, there's been a big move towards Internet of Things sensors as a way of monitoring things like usage and end of life of data. A recent IBM report states that up to 90% of data from Internet of Things sensors is never used. This really highlights the difficulty of moving towards digitalisation for the benefit of the organisation or the industry and then, at the same time, thinking about how this data can actually be used. It’s one thing to generate the data, but can you then use it effectively? 

Tom Jackson: So we coined the phrase “digital decarbonisation”. From there, we started to create a number of tools that forecast the carbon footprint your data project will have, looking at things like the types of data sets needed, where they're going to be stored, how frequently they will be updated, what sort of technology will be used and what sort of AI.

Before you embark on a project, our tool allows you to go through these things step by step to look at the output needs. Then a company can ask: do we actually need all these different data sets? Do we need them updated every single second? Do we need to have the latest AI, or could we just do simple predictive analytics? It’s all about thinking about our carbon footprint and starting to predict what it would look like for any new project.

Ian Hodgkinson: It’s important to recognise that digital data processes are not inherently good or bad. It’s how human beings choose to use them, and how organisations are drawing on them, that impacts things. The better we get at [managing data] and the more we optimise processes, the benefits become far greater than any negative outcomes. It should get to the point where any environmental footprint being produced from digital processes becomes justifiable, but it's all down to how we as human beings use software.

We did some work within the metals industry and found there is huge variation across the supply chain in terms of the sophistication of knowledge organisations have when it comes to measuring and mapping their data sources, where data is being captured and where it's being utilised. We saw quite significant variation in that type of capability.

Tom Jackson: So it's a case by case situation. We definitely know that a lot of organisations feel under threat because of the pressure to do something about sustainability but they don't really have enough metrics to make good decisions.

So, the easiest thing is to put sensors on everything. From the machinery in the mines to the furnaces, you can put sensors on absolutely everything to optimise how it works. And only the organisation itself will know what we call the ‘data resolution’ [of its activities], which is to do with having too much or too little data coming through. Very few [companies] hit that sweet spot where they've just got just enough to make the right decisions to optimise their operations without wasting energy on unnecessary data storage.

It's interesting that you say [companies] are putting more sensors in because it does cost them a lot of money. If you have more and more data, it’s going to cost you more and more because you've got to put it all into a data centre or data warehouse, or your own server within your own factory. It all costs money. And even if companies haven’t necessarily been very interested in sustainability, they have been interested in their bottom line. Cutting down [on unnecessary data] can save quite a lot of money, too.

Tom Jackson: [When operations go down] you can never bring that time back. It is genuinely lost revenue that [the companies] will never see again. So you can understand why companies chuck lots of money at AI and data because data is still relatively cheap compared to the downtime of a machine.

If you think about the greenhouse gas emissions in terms of the different scopes, most organisations are putting everything into Scope 3. That includes the data centres and data warehouses because they haven’t really got to worry about it, [Scope 3 emissions] are someone else's problem. 

We've been in discussions with the World Resource Institute [WRI] about this and our understanding is that they're starting to rework this, starting in October, and it should take two years. [The WRI] is looking to include digital decarbonisation in Scope 3 emissions reporting so that [companies] have to be accountable.

So, it makes sense to try and tackle this problem now. While predictive maintenance is important, do you need all those millions of data points, or can you reduce it down to maybe half a million data points? And on top of that, it’s also about the networking of data, the processing and the CPU time that is required to actually work out predictive maintenance.

Ian Hodgkinson: In terms of organisations asking themselves to question the CO₂ that's being generated from their data, we see high variation. For some organisations, they've never even come across the notion that data generates CO₂. For others, they are very much on it and are raising questions with regards to this. But they can't always get the exact figure from the [data] providers as to specific emissions quantities.

So, [some companies] are really challenging some of the cloud providers, but we see that huge variation from no understanding whatsoever all the way through to organisations demanding more information from providers. And accuracy in emissions reporting can be hindered by the existing knowledge base, rather than companies not being accurate on purpose.

Tom Jackson: In trying to make sense of all the new data constantly coming in, some companies can lose sight of what the whole business is about what they’re trying to achieve. You can see, [depending on the maturity level of some organisations] that it’s just overwhelming.

Tom Jackson: Better transparency. I think that's where it comes from. [Emissions from digital processes] shouldn't be a difficult thing for companies to work out. But if companies are putting data into the cloud, it has to be the cloud provider that calculates emissions. This can be done by looking at gigabytes exchanged and stored throughout the year, amounts of network traffic and CPU time for digital processes, and calculating overall carbon footprints from this.

But at the moment, the cloud providers don't do that. They don’t know the exact CO₂ footprint because often companies share data networks and use same cloud platforms, so it can be hard to break down specific emissions for different companies and processes. They struggle, but the cloud providers have certainly got to step up their game and begin producing hard figures for emissions.

Ian Hodgkinson: It’s also worth emphasising the importance of information governance at the organisation level.  Really understanding how data is being collected and stored. Re-evaluating the ways data is being used is important, as is not losing sight of digital technology as it progresses and advances. In the future, we’ll have far greater capabilities regarding the amount of data we can generate. We need to temper this with our own capabilities, and make sure that effective processes are in place internally to actually monitor and evaluate data use as a company ages and transforms.

What's exciting is that we're seeing far more discussions being had now in the media, in industry, and at the policy level than we were 18 months ago. But organisations have this responsibility to [monitor their digital activity].

Also, we are very pro-digitalisation. Digital processes have played such a critical role in decarbonisation efforts. But thinking about the other side of it is also important. We've never faced the sheer volumes of data now being created. As we look towards the future, if we carry on generating on this sort of trajectory, then we have to think about the impacts on the environment as well as the bottom line [for companies] as well.