Design For AI (Artificial Intelligence)
A comprehensive guide on AI Design Principles for Product Designer.
How to override the Blackbox theory and Create a trustworthy AI/ML product by keeping users in the loop.
When we talk about AI product, there are two major groups,
one Build AI product and the other use AI product.
Same way being a product designer, we may design a better AI product or use AI tools to enhance our design.
In part 1- Design for AI
How to use UX design principle to build an AI product
Part 2 — Design with AI
How to use AI tools to enhance and faster UX design.
We can’t dive deep into this topic unless we start from the basic. So first, let’s discuss a bird’s eye view on what is AI/ML, real time examples, how it works and evolution of human interactions. Then dig deeper into the design principles to design a better AI product.
What is Artificial Intelligence (AI)
Artificial (Man made) Intelligence (Power thinker) is a computing concept that helps a machine think and solve complex problems as we humans do with our intelligence. A machine can easily mimic human brain and execute tasks.
Artificial Intelligence can be divided in various types, there are mainly two types of main categorization which are based on capabilities and functionally of AI.
Based on Capabilities of AI -Type 1
- Narrow AI — Dedicated for one task.
- General AI — Perform like human.
- Super AI- Intelligent than human
Based on the Functionality of AI- Type 2
- Reactive — React to stimuli and does not have memory.
- Limited Theory — Use memory to learn and response.
- Theory of Mind — Ability to perceive human’s thoughts, emotions.
- Self-awareness — human like intelligence and awareness
What is Machine learning (ML)
Machine Learning is a subset of artificial intelligence (AI) that focuses on developing computer algorithms that improve automatically through experience and using data.
Three approaches used in data science are supervised learning, unsupervised learning, and semi-supervised learning models.
Supervised learning:
Supervised learning is a type of machine learning in which machine learn from known datasets (set of training examples), and then predict the output.
Reinforcement learning:
Reinforcement learning is a type of learning in which an AI agent is trained by giving some commands, and on each action, an agent gets a reward as feedback.
Unsupervised learning:
Unsupervised learning is associated with learning without supervision or training.
Where we use AI (Realtime examples)
AI and ML have been deployed in a wide range of businesses and solutions here are a few examples — Face Recognition, recommendation and suggestion for movie, music, product recommendation in online shopping, Email smart replies, google navigation, Autonomous Vehicles, credit score, fraud prevention, Diagnosis, Robotic surgery, speech recognition, image recognition, Virtual assistant, personal assistants, and many more.
How AI Works
![](https://miro.medium.com/v2/resize:fit:700/1*4WlNjdWsVgOzX-gykuPLSA.png)
INPUT layer receives the data > HIDDEN layer processes the data > OUTPUT layer produces the results.
Evolution of human interaction
- Human to human
- Human to machine
- Human to AI
Of course, everywhere we start with the human and the context of interaction.
Human to Machine
For example, A doctor & patient interaction. A doctor explains the root cause of the disease in a simple way/ laymen terms. Suggests or recommends solutions. We can ask questions for any concern in between and may be go with an alternate solution.
He should be transparent about his expertise, suggest the treatment based on the diagnosis report as an evidence, incase of surgery inform the risk, post-surgery effect. Also if he is not able handle suggest the best doctor etc. Lastly, showing empathy and suggest the solution which is fit for our body.
Human to Machine
Here we take an example of commonly used microwave. The microwave should explain the user how to use it. User can control the operation or intervene in the middle if something goes wrong. It should be transparent — highlight the Caution, maintenance trips, what to cook what not to, safety feature and security like child lock Lastly it should be easy to use and understandable.
Hence — For any interaction there are 4 core elements to make the interaction successful and trustworthy.
![](https://miro.medium.com/v2/resize:fit:700/1*y2bdHOLzkqxgXDlWfwkaYA.png)
Explainability
1. Explainable AI — XAI
Explainable AI refers to a process of explaining the user why it behaves, recommends, and suggests the result by providing a clear and human-understandable explanation for the decisions generated by AI and machine learning models.
Before showing some examples on Explainable AI, let’s understand what is Black box AI?
Black box AI is any artificial intelligence system whose inputs and operations aren’t visible to the user, or another interested party or in other word the decision-making process of an AI system is not easily understandable by humans.
Explainable AI systems aim to solve the black-box problem by providing insights into the inner workings of AI models.
Design consideration — Focus on visual and content
Clearly explain the reason with a simple language and in an elaborative way.
It should be clear on showing the user the input data, explain the process as much as possible (visually) and present the result.
![](https://miro.medium.com/v2/resize:fit:700/1*G5B4tOBWbeUaT6pooJ5t2g.png)
Above image is a Predictive forecasting with AI recommended targets.
Zia analyzes your historical forecasting and performance data to recommend the most accurate targets for your business.
![](https://miro.medium.com/v2/resize:fit:700/1*eIfZSegupQ89BcmznnKYVw.png)
A clear Explanation of the predictions in a simple language with visual.
Mostly Explain the what ( the score)and why of (driving factor)
2. Show Output with Details or drill down
Being able to show the evidence that used by the AI system to produce the output.
![](https://miro.medium.com/v2/resize:fit:700/1*SHakadvGg0fm7Tl1xD_S1A.png)
The Above image show the summary of conversations. Select a particular section, will explain the complete journey with clear text and visual annotation.
![](https://miro.medium.com/v2/resize:fit:700/1*8GWZRQwSDw2tingxdcw2ZQ.png)
Flexibility
1. Personalization — It improves the user experience.
AI personalization uses a customer’s demographic and past behavioral data — such as browsing and purchasing history, as well as social media interactions and learn the specific needs and preferences of that individual. The technology can predict what the individual may be interested in to make product recommendations in real time by Using those insights.
An example of AI personalization is when an e-commerce website includes a “Customers also bought” section to encourage shoppers to add more items to their cart.
![](https://miro.medium.com/v2/resize:fit:397/1*KTgkLrv9y7iqoGFEOSfn4w.png)
In the above image- AI system can make recommendations to your new customers based on the purchase patterns of existing customers, with relevant reasoning for each recommendation.
Here are a few Recommendation Engine Algorithms for ecommerce domain, where the system should explain why it show the recommendation list.
Personalized — Visitors that like this product also like that.
Cross Sells- Show your customers related products.
Upselling — Offer additional products in your cart that exactly match the ones already contained.
Last Viewed — Improve the navigation in your shop with a history of all visited products.
![](https://miro.medium.com/v2/resize:fit:700/1*FLzjA93I4McNaSqDKcGFJg.png)
2. User Feedback & Control
The users must feel like they’re in charge of the AI system. Allow user to intervene and change, undo, or dismiss an action or decision taken by the AI, as well as provide feedback and recommendations into the system.
2.1 Human in control
Giving users the controls to create or edit the rule
![](https://miro.medium.com/v2/resize:fit:700/1*SDta-uaXn4wPblxDt2MKlQ.png)
![](https://miro.medium.com/v2/resize:fit:700/1*rXcjGdpfQcuLhEJ_yu3dzw.png)
Customization
![](https://miro.medium.com/v2/resize:fit:700/1*GRP0isiaqEY20fnTcsRJOg.png)
User can easily Edit the result
![](https://miro.medium.com/v2/resize:fit:700/1*m_ZOkyNTOLBO_ypRA0PzJQ.png)
2.2 Teaching & Feedback
Human-in-the-loop is a blend of supervised machine learning and active learning where humans are involved in both the training and testing stages of building an algorithm. This practice of uniting human and machine intelligence creates a continuous feedback loop that allows the algorithm to produce better results each time.
The predictions made by AI tools are not 100% accurate. So, AI systems are not completely human free and require some amount of human interaction to rectify, monitor or train the system.
Design considerations:
Keeping user in loop and ask for quick feedback like is this information helpful “yes or no”, or little more specific about the suggestion to understand the user’s point of view. More feedback helps the AI model more smarter.
A simple popup with a question like ‘Is This useful?’ or ‘Did you enjoy my recommendation or tend ?’ will help the AI model to understand user opinions and train the algorithm.
![](https://miro.medium.com/v2/resize:fit:700/1*suGge9irzlAUvHrw0qxMww.png)
The deal performing card shows sales activities performed on the deal in the last 30 days and suggests the next best action that can be carried out on the deal. Also ask for the feedback in “yes” or “no”.
![](https://miro.medium.com/v2/resize:fit:678/1*xuKjQvHHgrcKa9TBdsM93g.png)
In Zendesk, a “Wrong score” LINK next to the prediction score, allow user to provide the feedback.
Train The model
![](https://miro.medium.com/v2/resize:fit:700/1*68ilenzI8re5xXGjEsuhtQ.png)
Above is an example to add the paraphrase shared by user to train the model.
2.3 Handle unexpected scenarios.
Even if an algorithm is almost flawless, there is always a possibility to receive an imprecise response. Depending on the user request, AI may provide either wrong or just not entirely right results.
Many funny and rude chatbot examples abound where the bot didn’t understand the context or people gave them simple, but unexpected, commands.
![](https://miro.medium.com/v2/resize:fit:511/1*Irpyoc3IskEYRxLNSVmTEg.png)
Design Consideration- Deep Understanding human behaviors are essential to train the AI system. Also test multiple scenarios will train the AI system more efficiently and allow it to handle unexpected cases.
Context switching
Amelia a Virtual Agent switches between threads and topics flexibly, providing quality humanlike experiences. She uses natural language processing and understanding (NLU) in conjunction with multiple Deep Neural Networks (DNNs) and natural language data sources to contextually understand and interpret simple and complex multi-sentence requests.
Design Consideration- solves this commonly encountered issue of context rigidity by combining knowledge graphs, natural language processing and computational linguistics to transform and refine the entire state of conversational experiences.
Fail Gracefully
AI is inherently probabilistic and so it should be designed for error and uncertainty. When the system fails, it should still retain some functionality or give useful feedback rather than just crash or show an error message.
Below is an example of showing graceful approach, when the Bot is not clear about the user’s query
![](https://miro.medium.com/v2/resize:fit:700/1*E89K7nHkYbDDIQTd7FKZLw.png)
Refer this article for error handling https://botfriends.de/en/blog/fehlermeldungen-bei-chatbots/
Transparency
AI transparency means provide a clear understanding of how the AI system works including how it makes decisions and processes data. AI transparency also involves being open about data handling, the model’s limitations, potential biases and the context of its usage.
Process transparency entails providing documentation and logging of significant decisions made throughout the development and implementation of a system. And it includes the structure of governance and testing practices.
Data and system transparency entails communicating to users or relevant parties that an AI or automated system will use their data. It also alerts users when they directly engage with an AI like a chatbot.
Tell the user when a lack of data might mean they’ll need to use their own judgment.
![](https://miro.medium.com/v2/resize:fit:521/1*gEFiUk33T5GqfUsRGtY0zw.png)
Tell the user why the AI system not able to provide the result.
![](https://miro.medium.com/v2/resize:fit:382/1*U44NMEP-K9atsEDXakcY1Q.png)
Set the right expectations.
One of the most crucial functions of AI machines is transparency. Be transparent about what the AI system can and can’t do, which elements are automated, and what the user should expect.
Clear communication can help setting realistic expectations and assist the user to build the mental models of what the system can and cannot do by being transparent about abilities and limitations.
![](https://miro.medium.com/v2/resize:fit:175/1*ylw5KWprxBJSlhEEshnlsw.png)
Try to explain in plain language what your AI can do, and where its limitations are. Generally, under-promising and over-delivering is a good way to build trust.
Design considerations:
Ensure clear UI communication: Use clear language in the user interface to explain the AI system’s capabilities and limitations. This can be via tooltips, help sections, or system responses.Prioritize user onboarding and feedback: Include information about the AI’s abilities and limits during user onboarding using interactive tutorials or guided tours. If the AI can’t perform a requested action, provide clear feedback and, if possible, suggest an alternative.
Communicate Confidence.
Provide users with a guarantee that their data are safe and protected by AI system:
- Tell users what data you need.
- Allow them to monitor AI data.
- Ask for permission for data acquisition and processing.
- Allow user an access to control their data.
- Asks for permission to user for using their personal data.
![](https://miro.medium.com/v2/resize:fit:540/1*Qew-Q1LE66cCPz58ZNBvcg.png)
Work on data privacy and security policies.
Clear communication about how user data is handled and protected helps build user trust.
Craft transparent policies: Develop clear, concise, and transparent privacy and security policies, avoiding technical jargon and legalese to ensure comprehensibility for all users.
Ensure accessibility: Position these policies within easy reach in the AI system, such as in the settings or help menu, and consider sharing policy highlights during the onboarding process.
Safeguarding against bias: To check or confirm that an AI system is not using data. in ways that result in bias or discriminatory outcomes, some level of transparency is necessary.
COMPAS, a risk assessment tool used to detect the criminal. But the AI wrongly labeling and Bias about racism. So the AI should explain from where and what basis it generates the score
![](https://miro.medium.com/v2/resize:fit:700/1*G-Q77KG9yQxwemOSth7_Mg.png)
Usability
Start with the user.
The technology you use should be guided by the user experience you want to achieve. Instead of diving headfirst into algorithms, think about how people do the task today. Figure out what’s valuable, and how you can enhance the experience.
AI product should be easy to use, understandable, Usable which make the process more efficient and faster.
Differentiate AI content visually.
There should be a clear differentiation between AI generated content and manual content.
![](https://miro.medium.com/v2/resize:fit:684/1*V9456xEiNissZ0t5FHCDlA.png)
Focus on Visual Elements
Show the AI driven results, suggestions, prediction visually appealing.
Performance Trend: This chart will have the target amount plotted in a straight line with the subsequent plot of accumulated achievement of the user, providing insights on whether the target is achievable.
![](https://miro.medium.com/v2/resize:fit:700/1*rB--SLL-QxnKh6yAVYTDIQ.png)
Elaborate the reasons
The Anomaly Finder scrutinizes the presence of anomalies in the attained value for a given period, and elaborates on the factors behind them, the anticipated versus actual targets, and the extent of their influence on the target.
![](https://miro.medium.com/v2/resize:fit:700/1*mNaDIaovdIoLvuLzpIGPYw.png)
Show the confidence score or accuracy.
![](https://miro.medium.com/v2/resize:fit:374/1*B5Cmv-k-ifUKnOaV20IS8A.png)
Design with usability in mind from the start
- Prioritize user needs: Focus on understanding and addressing user preferences and requirements, ensuring your AI system simplifies tasks and enhances the user experience.
- Create intuitive interfaces: Develop user interfaces that are easy to navigate and require minimal prior knowledge about AI, fostering widespread adoption.
- Incorporate feedback channels: Establish mechanisms for users to offer feedback on system performance, enabling continuous improvement based on real user experiences.
- Use accessible language: Communicate with users using clear, non-technical language to ensure the system is easily understood by all.
- Implement regular updates: Stay committed to user satisfaction by frequently updating and refining the AI system based on changing needs and user feedback.
- Onboarding: we need to explain to the user how the system works, how the data is collected, set expectations and ask for permissions to collect data.
- Activation (Voice product): Design the activation/deactivation process. Is it a button, a word, or a gesture? Should it be on by default or is it an expensive operation that needs user authorization to start? How does the user know it is activated?
Conclude
While Designing an AI Application think from perspective of Human, Data, and AI system.
For human. Develop AI applications for benefit of the human, society. Human has a control on the action provided by system
Data Transparency. Data availability, data quality, and data privacy is important. Using user’s data in a ethical, secure way and consent for how data is used.
AI system. The system should explain the internal process and show the reasons behind the prediction or suggestion. Take input from user, process the data and produce end result which answer the what, why and how questions.
Useful Resources on Design for AI -
Please refer the link for -
Part 2 Design with AI — How to use AI tools to enhance and faster UX design.