JAAG EXPLAINER N° 6

What is greenwashing
in the world of AI?

Companies in the ICT sector make some of the boldest claims to be green…

Greenwashing is a corporate practice of making misleading claims to be, or giving misleading impressions of being, ‘green’. 

Many of the sector’s biggest companies – the likes of Apple, Google, Amazon, Microsoft – now report that their energy supplies are almost 100% renewable, and that, in consequence, their operational greenhouse gas emissions are consistent with, or even ahead of, a trajectory to be Net Zero within a decade.  

These reports sound encouraging, but they are based on a misattribution of Greenhouse Gas (GHG) emissions. Whether the greenwashing is active or passive, direct or indirect, the reality is that there is a gulf between the companies’ reported emissions – and their public profile – on the one hand, and a more holistic assessment of their climate impact on the other.  

In broad terms, there are two dimensions to this problem: embodied emissions and additionality.

Embodied emissions

With any manufactured product, there is an environmental impact associated with its construction.  For relatively complex products like a laptop computer, this impact is likely to be significant. A typical estimate of the embodied emissions for a laptop is about 350 kgCO2, while a typical European who uses their laptop for say 8 hours a day will add a further 35-40 kgCO2 per annum from the associated electricity consumption.  (There is a wide range but these numbers are typical.)  That means that on average it takes nearly a decade for the energy-related emissions to catch up with the emissions associated with the laptop’s manufacture.  As most of us don’t keep our laptops for a decade before we replace them, the chances are that the embodied emissions are greater than the energy emissions over the product’s lifetime.

However, when a company reports its operational emissions, it is not including the embodied emissions associated with manufacture of its products, or related products, or their disposal, or related infrastructure.  Nor is it likely to be taking account of resource implications – for instance, the use of rare earth metals – or the associated environmental, social and ethical considerations of their mining. In the case of the computing industry, operational emissions are largely confined to the emissions related to electricity consumption.

While some of these considerations are difficult to quantify, it is usually possible to estimate embodied GHG emissions, as in the laptop example above; and when this is done, the ICT sector’s percentage contribution to global warming roughly doubles, i.e. it is probably closer to 4% than the figure of 2% that is usually quoted.   

For AI specifically, the impact is indirect: models and programmes that require more computing power and data storage lead indirectly, via hardware and data centres, to a bigger carbon footprint than ones which require less power or storage.

Additionality

This dimension is less appreciated but probably more important still.  When a company claims its operations are ‘carbon free’ or ‘net zero’, it is usually basing this on a combination of renewable energy supplies and offsetting. For both of these, it is important to ask: what is the additional impact of the operations on GHG emissions?

In the offsetting world, the concept of additionality was established in the early days of carbon trading.  It has long been recognised that when a company offsets its emissions via some carbon-saving scheme in Asia or Latin America, say, the benefit might be double-counted.  Perhaps another organisation or government is already taking credit for the scheme.  Or maybe the green activity upon which the scheme is based – be it tree-planting, water purification, burying waste, improved energy efficiency, cleaner energy production, or whatever - would have happened anyway, even if the scheme did not exist.  If either of these is the case, then it is best not to give the company credit for offsetting emissions by someone else somewhere else, and to limit emissions measurement to the impact of their own corporate activities.

With renewable energy, again one needs to ask – would the renewable energy be generated anyway? Usually the company is signing contracts, at market prices, with existing renewable generators.  A feature of renewable generation is that once it’s been developed it generates electricity whenever the wind blows, water flows, or sun shines. If the renewable generation assets are new, then probably they would be developed anyway, because the contracts are at market prices, i.e. the company is not paying any premium for the fact the energy is renewable.  (A simple analogy is to imagine you are selling your house. You live in an active housing area, with plenty of market activity, and you’ve priced your house at what an experienced estate agent considers to be its market value.  Someone quickly comes along and offers you the asking price. If they just as quickly pull out because of lack of funds, it is likely that someone else will come along and offer the asking price, and your house will sell anyway.)

If there is no net impact on the overall quantity of renewable energy, the extra demand from the company’s activities must be met by non-renewable generation (usually natural gas-fired power stations, in the case of Europe).  And the additional GHG impact is the carbon emissions from this extra non-renewable generation.

When both embodied emissions and additionality are properly taken into account, the measured carbon emissions are likely to be several times higher than claimed or implied by the ICT or AI developer.  This is illustrated in the diagram below (the numbers are made up, but hopefully the idea is clear). 

The gulf between perception and reality is the measure of greenwashing.