Top 7 Design Considerations for AI Interactions
To date, all software has been programmed to complete a specific task. It may have parameters, by they have been considered and configured ahead of time to provide constraints on the product’s capability. AI however, isn’t created nor doesn’t function this way.
The new paradigm
Counterintuitively, similar prompt inputs can elicit multiple unique responses and to complicate things further, these responses could be partially or completely wrong. This means users will now not only have to prompt the system but critically review the results. This puts the onus on the user to vet the results and is a completely new workflow for the majority of users.
AI is a powerful new technology which bring with it unique UX (user experience) considerations. These 7 interactions tips provide guidance on key areas where we can leverage current best-practice to create familiar and usable AI interfaces that ground these new products in reliably good experience design.
1: Tell users what AI can and can’t do
Tip: Ensure users are comfortable with how AI can best help them.
Critical for users new to AI, providing an understanding of what the AI can do is pivotal in using it effectively. A simple understanding of things like what the AI can reference, its response length and how to direct its tone can give users greater control over its usefulness.
A core promise of AI is time-saving but without users bring empowered to use it, these benefits will never be achieved. Users not only need to be comfortable using AI but also have a basic understanding of how it works so they can marry their requirements with AI’s strengths.
Consider: Provide educational content to educate users on the AI’s abilities. Make this content concise but empower them to feel confident in engaging with this new tool.
2: Transparency builds trust
Tip: Allow users to see the references used to generate a particular response.
When reviewing generated content, users will need the ability to properly vet the results, not only at face value but by accessing any references that underpin them. Show what sources the AI has referenced in it’s results. These could be from internal file systems or publicly available content sourced online.
Depending on our product’s context these references may need to be cited on a per paragraph basis or more wholistically as a single reference appendix. Any transparency around the cited source provides context for users to evaluate the generated content whilst building trust.
Consider: How your sources are referenced and presented back to the user. Allow them access to sources so users can vet it’s reliability.
3: Give fast and clear feedback
Tip: Provide feedback as soon as possible on prompt suitability and AI status.
Well-crafted user experiences give clear feedback on system status, this includes states like loading, deleting or saving. However, prompting an AI comes with new dimensions of feedback that occur before even actioning generated content.
While each context will mean that a “good” quality prompt means something different, this feedback will assist users in writing better prompts initially as it dynamically assessing their prompt. For example, if a quality prompt needs to be highly detailed, the prompt length could be used to gauge its level of detail. Simple heuristics like this can be paired with real-time assessment using the AI itself for a more powerful report on a prompt’s quality.
This type of feedback builds an understanding of the AI requirements for the user and also lowers the number of iterations required to produce a quality result.
Consider: Providing feedback on the user’s prompt dynamically as it’s written. Hone the metrics indicating a “quality” prompt to your particular use case.
4: Data is your secret sauce
Tip: Only give access to data that you want the AI to reference
A common misconception is that the “data” reference for AI is the training data, but this isn’t correct. The AI has already been trained by the provider (Open AI, Microsoft, Google etc) and is ready to interpret natural language to generate results.
The data it’s given is then referenced to create contextually relevant and accurate content. Having your data appropriately partitioned for this purpose isn’t a usual data architecture consideration, so this may need reviewing in preparation for adopting AI.
Consider: What data would be relevant for your user’s particular tasks and how you can manage the reliability of these resources ongoing.
5: Augment existing familiar interfaces
Tip: AI can be an enhanced layer to your existing product. Don’t rewrite your UI, augment it.
Whilst chat interface is appropriate for uses where the primary function is interacting with the AI, this doesn’t mean a chat interface is right for your product.
AI models can be a powerful tool to extend your existing application and do not require a complete interface design. Augmenting your current user experience with AI can be done by thoughtfully integrating it contextually where it can benefits your users the most.
For example, anywhere a user has to author content could be supported with AI generation. Any workflow with clearly defined variables could be actioned by prompting an AI with a desired outcome as opposed to configuring pages of options.
Consider: Where do your users need the most assistance that lines up with the strengths of AI. Looking for opportunities to speed up workflows or improve accuracy can be ideal situations for AI. This augmented approach builds an understanding of AI’s abilities which means your users will be more accepting of it takin the future as you adopt it further.
6: Educate by showing, not telling
Tip: Provide concrete and usable examples to users to make the AI black box more clear
Users are familiar with selecting existing options to accomplish a goal. It’s very different however, to prompt an AI with natural language which then interprets a desired outcome and generates results.
This inverted way to accomplish tasks is very new for users and will take time and repetition to become familiar.
So by providing users not only guidance but tangible prompt examples users will build a deeper understanding of how to author quality prompts. Consider showing users concrete examples like a prompt and it’s generated result. Build on this by showing how changing small variables can lead to large differences in results. For example, defining a tone of voice can change the structure and feel of the generated content.
Consider: Providing users example prompts relevant to their work as well as visual tutorials where possible to build confidence and understanding.
7: Asking AI to take action
Tip: Clearly define what the AI can access to action tasks
As powerful as AI is today, asking it to generate various media is only the beginning. Soon we’ll leverage its natural language ability to understand and execute sequences of actions autonomously.
While this may inherit some of the accuracy issues of the current generation, this type of task automation has been traditionally too complex to develop. Whereas, an AI that’s able to identify documents, data types and semantically understand user requests can be used to automate many processes.
Consider: What tasks in your business are both repetitive and easily defined? These will be the next focus of AI and offer users huge efficiency gains. These tasks may be things like simple business logic, document archiving, data transfer, or glossary updates.
Planning for AI Success
As you can see, with more dynamic outputs and new ways to think about tasks, how we design for AI should be considered earlier in the planning process.
Best-practice foundations
User-centered design methodologies provide the foundational best-practice to ensure the usability of these new experiences. AI brings an even stronger importance to understanding not only your users goals but their mental models surrounding them. How users think about a task directly relates to how they’ll interact with AI.
These tips build on decades of best-practice design while considering the unique user experience afforded by AI. Buidling on this knowledge allows us to effectively explore the new opportunities AI presents and guide its broad abilities effectively.
Guiding your AI
What configuration and boundaries you implement greatly impact its effectiveness. A deep understanding of the user and the business goals is required to thoughtfully balance these sometimes opposed requirements. You’ll need to understand not only the desired output format but it’s context and how your users will frame their requests.
AI Success
These tips represent foundational design principles providing effective user experiences which can greatly improve your chances of success. Understanding your users thought processes is critical in this new software paradigm where intent based commands are the primary input. How you research and plan for your desired outcomes will require completely new questions and testing methods. Question like “What are the most important parts of this task” and “What would a low-quality result be missing” help to capture not only how users think about their goals but how to effectively get there.
Whether it’s cost reduction, enhanced user experience or productivity improvements, pairing tried and tested design with thoughtful research early on will ground your solution and provide clear outcomes to point AI towards.