We Talk About Responsible AI. But Are We Forgetting the Environment?

I've been thinking about a question that feels slightly uncomfortable for someone whose business is helping organisations use AI.

What about its environmental impact?

I spend a lot of time thinking about responsible adoption. Whether people understand the tools they are using, what happens to their data, how they check the answers and where human judgement needs to remain involved.

But if we are encouraging organisations to use more AI, I think we also need to be willing to ask what that means for the environment.

An organisation might be working hard to reduce waste, asking difficult questions of its suppliers and making commitments about carbon reduction. At the same time, it might be giving employees AI tools and encouraging them to experiment.

I can understand both decisions.

I am less sure how often they are being considered together.

It is easy to forget what sits behind the screen

When you ask AI to help with something, the answer appears on your laptop. You don't see the equipment or resources involved in producing it.

There are data centres to run, servers to cool and chips to manufacture. There is electricity consumption, water use and equipment that will eventually need replacing.

The International Energy Agency's latest outlook expects global data-centre electricity use to roughly double between 2025 and 2030. That includes other computing as well as AI, but the demand from AI-focused data centres is growing faster still.

That feels significant enough to deserve our attention.

The same research also makes a point that can get lost: individual AI tasks are becoming more energy-efficient, and different uses can have very different demands. A short text response and a generated video are hardly the same activity. A tool working through a complicated problem may do considerably more computing than one answering a simple question.

Which is why I struggle with claims that every AI prompt has one fixed environmental cost.

What was it asked to do? How much work did that involve? Where was it processed?

Those details matter. I would want to understand them before using a striking statistic to tell someone what they should or shouldn't do.

AI might also help us reduce the impact

This is where I find the conversation becomes more complicated.

AI can help improve energy use in buildings, make vehicle operations more efficient and reduce the resources needed in industrial processes. The IEA has also examined the potential environmental benefits of those applications, alongside the reasons those savings might not be fully realised.

I can see a strong argument for exploring that.

If a system helps an organisation use less heating or avoid wasted materials, I would want to know whether the improvement outweighs the impact of running it.

I would also want to know whether the saving actually happened. A supplier saying that its product supports sustainability would leave me with a few more questions.

There is another distinction I think we need to make. Helping an HR team complete a task more quickly may be valuable. It doesn't automatically mean that task has become better for the environment.

We might have saved time without reducing energy or material use at all.

That wouldn't necessarily make it a poor use of AI. It would mean we should be clear about what benefit we are claiming.

Can a business afford to step away?

This is probably the part I find most difficult to settle.

Some businesses may decide that limiting AI is the right choice for them because of their environmental or ethical concerns, which is an acceptable stance to take.

But what happens if their competitors use it well?

There's clear evidence that useful gains are possible. Anyone running their own business can tell you how much time can be saved, but that doesn't mean the same gains will apply to every business.

It does make the commercial question harder to dismiss.

If a competitor can provide a good service more efficiently, it may be better able to hold its prices, invest in its people or cope with rising costs.

Customers might care deeply about a business's values and still need to choose the option they can afford.

I don't think we can assume that an ethical position, however sincerely held, will always be enough to keep a business viable. For an organisation already working with tight margins, that could eventually affect jobs and its ability to keep operating.

Some businesses may accept that possible trade-off. Others may find that personal service and human expertise are precisely what their customers are willing to pay for.

I wouldn't want to dismiss either choice. I would want it to be an informed one.

And the same scrutiny needs to apply to adoption. Buying licences, training people and checking the work all take money and time. If employees spend longer correcting an answer than they would have spent doing the task themselves, the promised efficiency starts to look rather less convincing.

The question is whether AI is helping with the work that particular business needs to do.

Where does HR come into this?

I can imagine an HR team reading this and wondering whether they now need to become experts in data-centre cooling as well.

I don't think so, but HR does have a part in how organisations introduce technology, what they expect of employees and how people learn to use it. Environmental awareness could become part of those conversations.

The government's Data and AI Ethics Framework already includes environmental sustainability. It is written for the public sector, but I think the underlying question is useful more widely: have we considered the resources involved alongside the benefit we expect?

That might mean asking a supplier what it can actually tell us about the service. It might mean discussing whether a simpler tool would do the job. It should also mean bringing the people responsible for sustainability into decisions about wider AI adoption.

For employees, I would start with the work in front of them.

Where does AI help? When does it add another step? Do they know how to judge the result, and do they feel able to say that a particular use hasn't been worthwhile?

I have written before about the importance of giving people space to learn and talk about what hasn't worked. That feels relevant here too. If employees think every AI experiment has to be presented as a success, we make it harder to understand what value we are getting.

The organisation also needs to take responsibility for the tools it buys and the expectations it sets. An employee can make thoughtful choices, but they cannot answer questions their employer has never asked its suppliers.

Responsible AI needs to include environmental responsibility

I still believe AI can be incredibly useful for HR teams and the organisations they support. Looking more closely at its environmental impact hasn't changed that.

What it has changed is how broadly I think we need to define responsible AI.

We already talk about data protection, security, bias, transparency and human oversight. Environmental impact deserves a place in that conversation too.

That doesn't mean measuring the carbon footprint of every prompt or avoiding AI because it uses energy. It means being more deliberate about where we use it, choosing the right technology for the task and asking suppliers better questions about the infrastructure behind the tools we're buying.

And I think there's a wider organisational point here.

If a business is setting ambitious sustainability targets while rapidly increasing its use of AI, those two strategies can't sit in separate rooms.

The people responsible for AI, technology, procurement and sustainability need to be having the same conversation.

For me, that's part of what responsible AI adoption needs to look like.