After reading The Checklist Manifesto, let’s just say, I believe in the power of checklists! I think we should be adopting them everywhere and a place I think we really need to adopt them is in the development of AI solutions.
Responsible AI principles from Microsoft outline 6 key areas we should consider for building solutions that minimise risk in the solutions we build. In their eyes, a responsible AI solution hits these checkboxes:
- reliable and safe
- privacy-maintaining and secure
How responsible AI fits into manufacturing
I’ll briefly cover how these relate to solutions for the manufacturing space:
- A fair solution delivers similar outcomes for people rather than perpetuating systemic biases in our culture. For manufacturers, the biggest risk of unfair systems comes from computer vision systems where there has been a track record of working unevenly for different groups.
- An inclusive system helps a broad range of people get benefit. Fairness is about outcomes, whereas inclusiveness is about access. If you get a speech AI system to help warehouse staff report problems, you need to make sure you have a fallback for people who have difficulties with speech AI interfaces. This can include people with strong accents!
- Reliability and safety are key considerations in manufacturing so you should be able to assess whether your solution will work well in the environment you put it in and it should prioritise human safety. For instance, computer vision defect detection systems can help monitor for problems in areas that might be high risk for humans (boosting safety) but may require a strong WiFi connection which your factory lacks (bad for reliability).
- Privacy conscious AI development is usually more in-focus for solutions involving people rather than solutions involving machines but you do need to be careful that for solutions like health and safety tracking, you make sure the outcomes and monitored feeds are protected so people’s privacy isn’t breached. Similarly, you need to make sure solutions are secure so people can’t game them, can’t access management interfaces, or reverse engineer them to steal your IP.
- Transparent use of AI is critical as surveillance and computer-driven decision making is a big concern for everyone. Be upfront about the use of AI, how you’ll use it, and where it’s limitations may be. For instance, you build a predictive maintenance solution that will generate alerts. It might only be useful for your latest machines so it’s important to let technicians know to watch out for potential problems on your older machines.
- Accountability, last but definitely not least, is about ensuring that you as the developer or deployer of an AI solution understand the risks and compliance considerations and work to ensure the other five outcomes. No technology is value agnostic and AI for automation is part of the long line of changing manufacturing industry. AI systems will impact jobs and being accountable means planning things like your longterm re-skilling plan of workers.
There’s a lot to think about right? It can definitely seem like a minefield but proper planning prevents poor performance afterall.
Building your responsible AI checklist
As the owner of a business, a business unit, or a tech team, you should be starting to hold yourself accountable by getting a checklist in place as soon as possible for any AI projects you’re thinking about.
A good checklist is an improving checklist so you can start with just the six principles and use it as a thinking or discussion aid. Longer term you can be developing out a more robust checklist and framework as your use of AI matures.
Checklists and frameworks you can be looking at for insight or use include:
- The UK Government Data Quality Framework (since Garge In = Garbage Out)
- Microsoft’s AI Fairness Checklist
- Guidelines and cards for Human-AI interaction
- pwc’s Responsible AI toolkit
- UK Government’s checklist for AI procurement
Getting started with a checklist should be your first step and it should be built with buy-in from others since it is a mechanism for attaining good outcomes.
You should be comfortable discussing the outcomes of technicology and thinking things through as you adopt it. You don’t have to be an IT specialist to do this - technology supports business change and you know your business. Give Ethical considerations for AI monitoring a read to start understanding a bit more about some of the areas in computer vision in the workplace for instance. PS It has another checklist
Ideally, responsible AI should be part of your overall business strategy and the specific section on how AI aligns in your organisation. If you’re unsure about AI and it’s strategic value, you can do some further reading and/or book in a chat with me to discuss!