Encouraging safe AI adoption: A principles-based approach

Many companies recognise that the use of GenAI tools such as ChatGPT could lead to massive productivity gains if used by their workforce. These gains potentially come from two avenues.

1. Automating tasks - Reducing employee time spent to generate the same output. This frees up time for employees to focus on other (potentially higher value) activities.
2. Augmenting tasks - Giving the employee an additional resource which they can use in order to produce higher quality work. This might include outputs which are better researched, more likely to solve a customer problem or more clearly worded.

Two major forces slow the rate at which this adoption proceeds.

1. Harm avoidance - Businesses rationally fear that using Generative tools will lead to reduced quality or new risks. As a result, they take top down steps to prevent or slow adoption of this new technology. These fears are similar to those arising from any change in process, although made more acute because of the low barriers to entry to using GenAI tools.
2. Behaviour change - Getting employees to do their work in a different way is rarely something that can be brought about by instructions alone. This is even more the case when asking employees to figure out the best new way of doing something themselves. For most employees, except those who have an alternative extrinsic source of motivation (such as an interest in the technology itself), carrying on as they have before is likely to cause much less friction, even at the cost of longer term benefits to themselves.

A Principles Approach

Principles, communicated from the top of an organisation, can have two major benefits in leading employees to explore and adopt positive behaviour changes, like the integration of GenAI tools into their workflows. Principles communicate values and boundaries. They are particularly useful in situations where employees must learn by doing as they can provide employees with a north star when navigating unfamiliar situations. Whilst alone they aren’t sufficient to mitigate all harms or encourage everyone to make a change, they are a key part of most behaviour change programmes leading to the adoption of new technology.

Principles work alongside other tools such as: training, formal policies, incentives and social demonstration & proof. Where the task is to get employees to figure out: what is the most useful new thing I can do rather than simply getting them to obey do as I say, they have a particularly crucial role in making employees feel safe experimenting and learning, whilst protecting the company (and its clients) from harm that might arise during this process.

Ten principles for safe AI adoption

Below are 10 principles that organisations might use when trying to create a positive environment for the safe adoption of ChatGPT-like tools. Please feel free to use or adapt them for your own business.

1. Continuously improve. Innovate your ways of working, particularly on tasks that you do regularly.
2. Celebrate efficiency. Finding ways to do the same work quicker is a positive thing and should be celebrated - so long as quality and understanding are maintained.
3. Take ownership. You are in charge of the quality of your work. Editing and improving things written by teammates or machines is an essential skill.
4. Examine critically. Think about what might be missing. It’s rarely a good idea to accept the outputs of Generative AI tools wholesale (except in very low value tasks). Treat outputs as a starting point to build on, not necessarily the finished product.
5. Share carefully. Trying new methods is good, so long as you stay within the boundaries set by your team. Use approved tools and follow your organisation’s rules for personal, confidential and client data. Where a tool has not been approved for that information, do not share it.
6. Play as a team. Discuss new & improved ways of working with those who do similar roles. You aren’t in a race with your colleagues, you’re in a team. Sharing failures is valuable too, because you can help your teammates avoid making the same mistakes.
7. Be a good manager. Treat delegation to AI as a management decision. Good questions to ask yourself include: How can I phrase my instructions to get the best results? What am I able to responsibly delegate? Set clear expectations and check the work against the task’s requirements; do not assume that familiarity alone will make the system reliable.
8. Think long term. Some tasks are learning tasks. Particularly if you are a new joiner, the process of doing something may be more important than the result. Check if this is the case and don’t cheat yourself by using shortcuts when you are supposed to be developing.
9. Specialise. Prioritise automating work that isn't your core function, like formatting. Use freed up time for the real reason you are here.
10. Be you. Remember that human interaction is core to so much of what we do - be protective where there are risks to losing this.

‍