Artificial intelligence (AI) is transforming various domains and industries, from health care to entertainment. But how can you make AI systems accessible to everyone, regardless of their background, abilities, or preferences? In this article, you will learn some principles and practices that can help you design and develop AI systems that are inclusive, ethical, and user-friendly.
The first step to make AI systems accessible is to understand your users and their needs, goals, challenges, and expectations. You can use various methods, such as surveys, interviews, personas, and scenarios, to gather user insights and feedback. You should also consider the diversity of your users, such as their age, gender, culture, language, education, and disability, and how they might interact with your AI system differently.
Remember to cover the basics first: all that you would do to ensure accessibility to a non-AI system will also most likely apply here! And there's a massive amount of practical advice applicable, so there's no need to reinvent the wheel when it comes to understanding your users.
Leverage your current chat bot and text data like call transcripts, chat logs and emails to identify customer needs through keyword analysis, sentiment analysis and intent recognition. Analyze this data across risk profiles mapped to customers to understand pain points of high-risk, unhappy customers. Overlay journey maps and existing metrics like resolution rates to spot key drop-off points an AI assistant could address. Continuously analyze your customer conversation data, sliced by risk profile, to derive ongoing insights into the highest priority needs and use cases to tackle with AI.
Execute execute execute! I will refer to the build fast fail fast strategy. It is best to get some sense of the potential users to get an idea of the direction you want to head in and then launch. The most valuable feedback you can get on your AI system would be from actual users.
The second step is to design your AI system for accessibility, which means ensuring that it can be perceived, understood, operated, and enjoyed by everyone. You can follow some guidelines and standards, such as the Web Content Accessibility Guidelines (WCAG), to make your AI system accessible on the web or mobile devices. You should also test your AI system with different tools and users, such as screen readers, keyboards, voice assistants, and people with disabilities, to identify and fix any accessibility issues.
The current AI is a UX game! The more seamless your solution is integrated within your product/system, the better it is for the users, and the more they will love it. Great resource: https://kozyrkov.medium.com/whats-different-about-today-s-ai-380569e3b0cd
Apart from WCAG, WAVE and Apple's accessibility guide are useful for guiding the design process. We should also try to improve accessibility for those suffered from cognitive and neurological impairments. These kind of impairments maybe more difficult to tackle, but we should try our best effort to be as inclusive as possible.
Follow accessibility standards like WCAG for web/mobile. Support multiple interaction modes - voice, chat, keyboards. Use plain language and allow text adjustments. Accommodate screen readers. Partner with disability advocacy groups for design input. Actively test with diverse users. Train AI with varied data. Maintain human fallback. Continuously gather feedback from user testing to identify and fix access gaps, ensuring AI benefits all customers equally.
The third step is to explain your AI system, which means providing clear and meaningful information about how it works, what it does, and why it does it. You can use various techniques, such as natural language, visualizations, metaphors, and examples, to communicate the logic, functionality, and outcomes of your AI system. You should also be transparent and honest about the limitations, uncertainties, and biases of your AI system, and how they might affect the user experience and trust.
To explain the AI system and its performance -it is imperative that you understand the limitations of the AI algorithm and its boundaries. To avoid the type of systemic racial, gender or economic bias (amongst other types of errors) that plague several enterprise systems - you must work from first principles to gather data from an array of sources and ensure that you stress test the assumptions that the data is forcing you to make. This is true in healthcare, digital marketing uplift measurement, credit score validation, price optimization, algorithmic trading and a host of other areas.
If your are upgrading your existing chat bot, most users will already be familiar with how to interact with it. That said, you can rollout videos and tweets/articles on what has changed & what users can expect with the new chat bot features. I also know some companies that mask the generative AI chatbot to appear human, and direct customer inquires to live agents when the AI cannot decipher the request.
The fourth step is to empower your users, which means giving them control and choice over how they interact with your AI system. You can use various features, such as preferences, settings, feedback, and opt-out options, to allow your users to customize and personalize their AI system. You should also respect your users' privacy and security, and ask for their consent and permission before collecting, storing, or sharing their data.
Here are some specific ways to empower your users: Provide clear and concise privacy policies and terms of service. This will help users understand how their data is being collected and used. Give users the ability to control their data. This includes the ability to opt out of data collection, to delete their data, and to request that their data be corrected. Make it easy for users to give feedback. This will help you improve your AI system and make it more user-friendly. Be transparent about how your AI system works. This will help users understand how the system is making decisions and why. Be responsive to user concerns. If users have any questions or concerns about your AI system, be sure to address them promptly and appropriately.
As a user who is not technical and does not understand AI systems, it is probably not very obvious to them what part of the system is AI and how it affects them and other functionality. Let take an example of a chat bot. Since multiple years people r familiar with chat bots. Initially it was a person on other side answering their queries. Then slowly preformed answers to few questions came into picture and people were confused as they will not get correct reposes. But they adapted to it and slowly understood the system. Now chat bots powered by LLM's or more smart ML algorithms act like human is on other side. This can really confuse or even scare a common person. So it is responsibility of the system owner to inform the user.
The fifth step is to learn from your users, which means collecting and analyzing data and metrics to measure and improve the performance and usability of your AI system. You can use various methods, such as analytics, testing, and evaluation, to understand how your users use your AI system, what they like and dislike, and what problems they encounter. You should also involve your users in the design and development process, and solicit their feedback and suggestions for future improvements.
Monitor usage metrics and run A/B testing to optimize AI adoption and impact. Survey users for direct feedback and label interaction data based on input to improve predictions. Review transcripts and recordings to analyze conversations and sentiment. Track key performance indicators like resolution rates to quantify AI value. Share improvements with users and report metrics visibility to maintain engagement. Continuous usage analytics coupled with user interviews and an advisory panel provides insights to refine the AI and maximize positive customer service outcomes.
The sixth step is to follow ethical principles, which means ensuring that your AI system respects the values and rights of your users and society. You can use various frameworks and codes, such as the Ethical AI Principles, to guide your decision making and actions regarding your AI system. You should also consider the potential impacts and consequences of your AI system, both positive and negative, and how they might affect the well-being, dignity, and justice of your users and others.
In complex projects, quick and pressure-driven decisions are inevitable at each stage. Maintaining clear and shared among team members ethical principles throughout is crucial to prevent any single decision from adversely impacting the entire project. Adhering to these principles is vital not only to protect company's image, but most importantly, to safeguard all stakeholders involved. This is especially true in sectors like pharmaceuticals and healthcare, where decisions can directly and significantly affect people's health and well-being. My experience in the pharmaceutical industry has ingrained in me the importance of prioritising ethical decision making in every project I lead or participate in.
There's an "old" saying that the best interface is no interface. Before considering any accessibility principles, you can still ask yourself: is it possible to make my AI system so transparent and invisible to my users that I don't even have to consider accessibility? Keep it simple.