AI usage policy

Context

Using artificial intelligence (AI) involves ethical tradeoffs and we want to be honest about how we view those tradeoffs. AI allows us to be more productive, so we can deliver more value for the same budget. On an individual level it allows us to offload the more repetitive work (under supervision) and concentrate on the more interesting tasks. However, there are environmental and authenticity issues that need to be offset against these positive sides of its use.

Climate and environment

AI use requires data centres and the associated impacts of their construction and operation. There is a particular focus on (a) the use and generation of the electricity to power data centres and (b) the use of water to cool data centres. There are other issues such as the inflated employment benefits for communities that data centre companies sometimes use when getting planning permission, but these are complex issues beyond the scope of this policy.

Different types of AI use require different amounts of electricity use. Although clear figures are impossible to come by, in one benchmark simple classification tasks consumed about 0.002–0.007 Wh per prompt on average (about 9% of a smartphone charge for 1,000 prompts), text generation and text summarisation each used about 0.05 Wh per prompt, image generation averaged 2.91 Wh per prompt, and the least efficient image model used 11.49 Wh per image (roughly equivalent to half a smartphone charge).

In terms of how the electricity is generated, many companies that run data centres make lots of noise about using renewable energy sources (e.g. Amazon and Google), but don’t provide granular data, and also use dubious accounting practices such as “matching”.

Water use is even harder to measure as it can vary hugely between data centres, for example, whether a data centre uses a closed-loop or open loop system.

There is plenty of greenwashing around these issues, and when there’s such a lack of reliable hard data it’s impossible to use any metrics to guide usage beyond broad brushstrokes.

Accountability and authenticity

AI makes it particularly easy to generate text from a simple prompt, raising questions of accountability and authenticity.

When we produce any written language, whether that be code, contracts, communications or anything else, we need to be accountable for that language.

To maintain authentic relationships with clients, we also need to use our individual or collective voices.

AI has the potential to reflect biases present in its training data. Although we don’t produce copy for clients, and in fact always recommend the use of a good copywriter, we need to be alert to these kinds of biases.

Our usage policy

To maintain authentic relationships with our clients, we will never use AI for direct communications such as email, instant messaging or video calls.

For functional language, such as coding and contracts, we may use AI for generation or suggestions. We may also use it for automatically generated notes from video calls. All generated or suggested language from AI will always be reviewed and edited by a human before being used.

We may occasionally use AI for research. If we do, we will always explicitly ask for references and check those references.

We will not use it to replace the employment of skilled workers like photographers, illustrators or copywriters, for both ethical and quality control reasons.

Client budgets often limit the availability of suitable images. We may occasionally use generative fills on an image where a client has a particular requirement (e.g. using a specific image in a hero layout) and there are no other reasonable alternatives.

We will never use AI for full image generation or any video generation. This is mainly so we only use trustworthy, representative and sympathetic imagery that is consistent with our clients’ brand and values, but also due to the significant energy requirements involved.