Design principles for Human-AI interaction


  • AI technologies are advancing exponentially. We're playing catch-up – trying to get a grip on how we should use them and what we want them to do for us

  • Designing AI's the right way is critical in ensuring they meet the needs of their users, deliver the results we want, and don't confuse, mislead or do harm

  • Good AI is a framework for designing the right AI

Artificial intelligence is rapidly changing the way people interact with the products and services they use. New interaction patterns are being invented daily with new technologies powering new customer experiences.
We have the opportunity and responsibility to design Good AI – an approach built to benefit everyone.

To help guide us, we’ve created eight Design principles for Human-AI interaction.

Design principles for Human-AI interaction
  1. Good AI solves human problems

  2. Good AI is clear about what it can do

  3. Good AI is easy to use

  4. Good AI is easy to understand

  5. Good AI is honest

  6. Good AI keeps its user in control

  7. Good AI is always learning

  8. Good AI is inclusive to all

1 – Good AI solves human problems

As with any business challenge – success comes from solving real human problems.

Make sure you’re solving one, and not diving in with an AI already being your solution.

Google puts it like this – “If you aren’t aligned with a human need, you’re just going to build a very powerful system to address a very small — or perhaps nonexistent — problem.”

Where to start…

  • Don’t start with an AI, start with a person – use all of the tools in the UX toolbox (ethnography; contextual inquiries; shadowing; support ticket analysis; deep hanging out; etc.) to really understand your users and their needs – Psst, Good AI can help you along the way!

  • Bring your user experiences to life in journeys. Humans are hard-wired to understand stories, so make your journeys tell the story of how your product and your customer connect…
    …and only then find where AI fits by plotting out where Good AI will add meaningful value

  • Bring everything together in a solid service design before you commit to coding your AI-powered vision.

2 – Good AI is clear about what it can do

Today, it seems as though everything from advanced self-driving cars to mundane washing machines are ‘AI-enabled'.

There are many preconceived ideas and expectations about what AI can do in the minds of users – not all of these ideas are baked in reality.

Avoid misunderstandings, prevent disappointment and build trust.

Where to start…

  • Introduce your user to what your AI can (and can’t) do early on in your product experience, setting their expectations of how your AI adds meaningful value to their life

  • Be transparent about how your AI works and what data it is using –this will help users to understand and build trust with your AI

  • Always inform your user when your AI updates or adds to its capabilities – don’t forget to take your users along with your 
rapidly improving AI.

3 – Good AI is easy to use

Simplicity drives success because people prefer easy-to-use products.

Ease-of-use creates value by minimising effort while maximising rewards – so design your AI to be Always Intuitive, a seamless user experience that’s instinctive and instant.

Make that your mantra and turn your AI into Good AI.

Where to start…

  • There are many rules for making products easy to use: Simplicity; Using clear and concise language; Using familiar design patterns, and many more. Here are two we think you should start with:

    • Design for the context of your user’s action. Think about where and when your user will be interacting with your AI and let that context guide your design

    • Design for the skills of your user. Match the complexity of your AI interface to the level of experience your user has within your domain

  • Design for your user to spend more time with your AI than they might expect – developing intuitive interfaces and fostering ongoing digital relationships can reduce the learning curve and increase personalisation, making your AI experience a user-friendly experience.

4 – Good AI is easy to understand

Your AI has the power to shape your users' behaviour, guide their decision-making, and determine their next actions.

It’s critical, therefore, that your users can easily comprehend the messages conveyed by your AI and grasp the underlying processes that generated those conclusions.

Don’t leave your users guessing, make it easy to understand what’s going on and what to do next.

Where to start…

  • Display the results of your AI-generated information in a way that’s relevant or expected given your user’s current task and environment

  • Ensure the language that your AI uses is in keeping with your user’s expected language of your brand or system

  • Toning down your user interface, or altering the layout to something more conversational, can help interrupt your user’s preconceived expectations of given results – helping to guide behaviour.

5 – Good AI is honest

Your AI will not be perfect and can make mistakes.

Having an AI that is right some of the time but wrong at other times can be difficult for users to understand, and can lead to confusion and possible negativity.

Honesty is the best policy – be trustworthy, be transparent, and make Good AI.

Where to start…

  • Alert your user to the fact that they are about to engage with an AI – don’t hide it behind a false facade

  • Don’t be afraid to say you don’t have all the answers all the time – this can be a positive opportunity to enrich an ongoing experience and sharpen the accuracy of results if designed into the full user experience

  • Provide users with a transparent explanation of your AI’s decision-making processes, so they can better understand and trust its behaviour.

6 – Good AI keeps its user in control

AI should be a tool that amplifies human abilities, not replaces them.

Your users should be in control of your AI and should be able to intervene, provide feedback, and reverse bad actions. Design your AI to be as flexible, extensive and reliable as possible, while always keeping your user firmly in control.

Good AI is most useful when it works with users, not instead of them.

Where to start…

  • Design your AI to amplify and augment your users' abilities – don’t turn them into spectators. Focus your AI on the hard parts of their journey and keep the experience wholly rewarding and fun!

  • Make it easy for your user to dismiss or ignore undesired AI services – if you can, provide a gracefully degrade of AI-driven features

  • Make it easy to edit, refine, or recover when your AI does something wrong or delivers incorrect or unexpected results.

7 – Good AI is always learning

Your AI should be used to improve people's lives over time, not just be the next big thing.

By continually learning and adapting to your users’ needs, your AI can become more helpful, and more useful with every interaction.

Switch-on your AI to the power of lifelong learning.

Where to start…

  • Learn from user behaviour – ensure your AI personalises your user’s experience by learning from their actions over time

  • Enable your AI to ask as much as it tells – your AI can learn considerably by simply asking questions

  • Make it easy for your user to provide feedback indicating their preferences, thoughts and reactions during regular interactions

  • Allow your user to easily filter, control or modify your AI-generated results so that your AI can continually learn from their interactions.

8 – Good AI inclusive to all

AI is for everyone.

By designing for the widest range of people, you benefit everyone. People use technology in different ways and for a variety of different reasons. Your AI should consider everyone - their requirements, limitations, and impact.

Good AI aims to create a great user experience for everyone who touches it.

Where to start...

  • Start by identifying situations where people are excluded from using particular technologies – empathise with the widest group of people you can – and use your findings as your starting point for designing inclusivity AI

    Here’s an example – due to my age, I need reading glasses to read emails on my phone while I’m on the tube, if I forget my glasses I’m temporarily excluded from that technology. Someone with a visual impairment may be permanently excluded. Solve the problem for one, solve the problem for all.

  • Ensure you’re aware of any inherent biases within your AI and do everything you can to design them out of your product

  • Don't wait until the end of the development process to think about inclusivity and the resulting accessibility. Make sure they are a core consideration throughout your entire design process.

One last word…

Prototype with real data and fake AI

Test early. Learn fast. Prototype your AI assumptions with users before you commit to code. This can sometimes be hard without the tech in place, so we use the wizard-of-oz method - fake it until we make it!

Good AI is human by design

Are you looking to create Good AI?

I'd love to talk more, drop me your thoughts:

Thinking of starting a new project?

Let's talk

Interested in joining our team?

Join us