Gartner recently predicted that by 2024, 69% of a manager’s routine work will be done by AI, which lends credence to the fact that AI is here to stay. Many products now integrate AI functions into their software and apps and this influences the way people think, feel and behave towards the product. So what does this mean for designers? Designers can help build AI products based on thoughtful designs that are useful, easy to understand and improve the AI model.
One of the most common mistakes that AI software companies make is not understanding nuances. Although AI has improved greatly in a very short amount of time, the machines could stand to be a little more intuitive and natural in conversation with humans. AI can also show biases if based on flawed or incomplete data. So how can we be better prepared than our AI UX is effective? Here are some strategies.
Type of AI
First, a little background on AI. Artificial Intelligence or AI is a branch of computer science that tries to replicate or simulate human intelligence in a machine, so machines can perform tasks that typically require human intelligence. There are three types of AI:
- Artificial narrow intelligence (ANI) has a narrow range of abilities. Examples include Siri, Alexa, Cortana, and Google Assistant.
- Artificial general intelligence (AGI) is on par with human capabilities. K, built by Fujitsu, was one of the best attempts at building an AGI, but it took 40 minutes to simulate 1 second of neural activity meaning it is unlikely we will see a legitimate AGI in the near future.
- And, artificial superintelligence (ASI) is more capable than a human. ASI is currently hypothetical and more in the realm of sci-fi movies and TV shows.
What is an AI-enabled interface?
A human uses AI through a user interface or a UI. An AI-enabled interface is an interface with simulated cognitive functions which can facilitate interactions between the human and the machine. Examples include Alexa, Netflix, Spotify, etc. In AI-enabled UI, the AI has to understand the user’s commands and present the user with the best possible prediction based on all the available data. This user experience (UX) between AI and humans should be simple, easy to navigate, and efficient. The challenge designers face is how to best design an AI-enabled UX which users will want to use.
List user expectations before designing tool interface
The user’s role and the user’s goal have to both be clear at the outset when you start designing the tool interface. After all, you can only address the user’s needs if you generate insights about them in the design process. Not everyone has the same expectations, experience, and level of trust in AI and so the design will have to be different for different users. So instead of diving headfirst into the algorithms, think about how and why the user will use the AI tool to get what they want. A user does not have to understand the math behind an algorithm, but the UI could visualize the algorithm in such a way that it makes the AI’s decision-making process clear to the end-user.
Plan for unexpected scenarios
Experienced designers will know that users have a tendency to use products in a way you didn’t plan for. This is where fallbacks come into play. Fallbacks are an alternative plan that can be used instead of the planned for or expected user path. By adding fallbacks into the design, you are making sure that your product is always ready for unexpected scenarios. A good example of this is dead ends. This is when a user’s journey in interaction with the AI comes to an end and there are no next steps available. Good design’s goal is to offer the next steps to the user and track their choices. For example, if a user hits a dead-end in asking the AI to search for something, the fallback can be asking them what they would like to know and building a knowledge base based on their responses so the responses can be smarter in the future.
Share a brief description of what to expect from the algorithm
These days users hear and read about AI so much that their expectations of AI are all over the place. They might have heard about Tesla’s self-driving cars or Revolt’s IoT-enabled bikes closer to home, seen a futuristic show on Netflix, and used a Qubo device at home, all leading them to form very different conclusions about what AI can and cannot do. They may underestimate AI or overestimate it. The easiest way to navigate these expectations is to clearly define what your AI product can and cannot do and where its limitations are. Generally, it is always a good idea to under-promise and over-deliver. Over time users will realize the scope and limitations on their own after repeated interactions.
Getting to conclusions, without getting the context
What takes AI to the next level is the addition of context. When humans communicate with each other they rely upon the environment, body language, tone, and a hundred other variables to decide the context for the conversation, something AI is not very adept at. If the AI reaches conclusions without the addition of the context of what is being said, it will often give the users the wrong predictions. Context is what frames information and gives it meaning. This requires giving the AI exhaustive training in connecting the dots to form a context. In the case of self-driving cars, for example, teaching the vehicles to drive in rainy conditions is difficult because of the variability that the wet conditions bring. But if the AI is connected with contextual information like rain, light, traffic congestion, and temperature, it is possible to combine information from multiple contexts and allow the vehicle to decide the next action forward.
Test user behavior at different stages
During the testing phase, you can see how users are interacting with your prototype and gather valuable feedback. User testing is not something you should reserve for the final stages, rather you should test user behavior early and often. In the early stages, it will help you get feedback on the initial test prototype to steer your thinking towards directions you might not have thought to fully explore. And towards the final stages, testing user behavior will help you fine-tune how your AI interacts with them. So you should include user behavior tests at each stage of the design process, since the AI design is a human-centered process, and let the users dictate how the end product will be used.
Evoke trust by protecting user data
An AI shouldn’t collect data it doesn’t need since it can easily depend on machine learning to learn more about the user. Data should only be collected when absolutely necessary and the user should be made aware of what data is being collected and what it will be used for. A user should be allowed to choose what private data he or she is comfortable sharing. And once the data is on your servers, it is your responsibility to protect it and make sure that it is not sold or hacked. Determine early on who needs access to the data and for what purpose. This is also a great way to evoke trust among your user base who will be more confident to use your product.
Conclusion
Designing AI UX is a new and exciting field full of challenges and learnings. A good designer should aim for products that are easy to understand, easy to use, helpful, and trustworthy. The above-listed design principles should form a good base for building an effective AI UX.