Introducing the Antistatic generative AI statement
Today we published the Antistatic generative AI statement on our website. It sets out the parameters and considerations we will apply when using tools that generate text and images based on patterns from existing large datasets (GenAI).
As long-time writers about digital technology, one of the hills we’ll die on is that it’s not helpful or sufficient for a business to say “we use AI to improve efficiency and productivity.” What do you mean by AI? What do you mean by efficiency? The blanket term “AI” covers many types of model and capability, each with different levels of risk and reward. To get the best outcomes, ensure transparency, and foster trust, there needs to be specificity — both in discussing the types of tools you’re using, and in how you’re applying them. This is especially important in moments of transition, when new technologies are made available and heavily promoted, and people are still working out how best to integrate them safely and effectively.
We decided it was time to put our money where our mouth is, do the thinking, and set some clear parameters for our own business around how we’ll use GenAI in our work. We kept the scope of our statement pretty narrow, with a focus on the functional aspects of these tools and where we will and won’t use them. We have much more to say about the ethical aspects, business considerations, and political impacts of GenAI when it comes to work and writing, but we’ll save that for a different format.
At the core of our GenAI statement is a clear assertion: the writing that appears in our client deliverables and internal project outputs will be written by us, and that any use of GenAI in other aspects of our client work will be validated by a person and supported by human judgement.
Our decision not to use AI to generate content for our client deliverables and project work reflects a number of factors, including:
it’s easier for us to track provenance, ensure appropriate tone and voice, and produce distinctive and unique outputs when we draft content ourselves (we also really enjoy writing!)
editing AI-generated outputs to ensure accuracy and make style and content adjustments can be time consuming, and there is a risk of missing convincing-looking errors and gaps in thinking
the writing we do for clients is often produced through a process that includes multi-stakeholder discussion, deliberation, synthesis, negotiation, and collaborative drafting — the process itself is important to ensuring a trusted output
LLMs are limited by their training data and other parameters, which can lead to bias in the outputs produced — for example they may not adequately reflect minority and Indigenous perspectives
keeping our writing and analysis skills sharp is important so we can provide high quality advice; completing some tasks manually is a way for us to do this.
It feels risky to state your position clearly, even when it’s paired with careful reasoning and a strong value proposition. This is a work in progress: we welcome your thoughts and feedback on our statement. We hope that it provides a useful insight into how we work at Antistatic, and food for thought for others working through how they might apply these tools and talk about their decisions.
Please get in touch if you’d like to discuss further: hello@antistaticpartners.com
Acknowledgements
Thank you to Dave Moskovitz and Mandy Henk for reviewing our draft statement providing thoughtful feedback to make improvements.