Understanding AI: Essential Rules of AI for UX Design
First of all, let’s start with AI terminology. It will help us understand the topic.
Artificial Intelligence (AI): Broadly defined as the ability of machines to mimic human behavior. Machine Learning, Deep Learning, and Generative AI are all subsets of Artificial Intelligence.
Machine Learning: When you train them on a large amount of data, they can find patterns and make decisions without being programmed for those specific tasks.
Deep Learning. Take this to the next level with large amounts of data, and these computers will generate multiple layers and create artificial neural networks, which are designed to mimic how the human brain works. This allows them to learn from data on their own.
Generative AI: When you train machines on data so that they can generate data. This data could be text, images, video, sound anything that you train the machines. When people talk about AI most of the time they actually talk about Gen AI.
When hearing about Gen AI one of these tools comes to mind: Chatgpt, Google Gemini, Bing AI and Midjourney.
Generative AI creates different types of data: Image, Audio, Text, Numeric.
When we think about using AI, we take into account what problem your product or service is trying to solve for users and how AI could possibly be used to solve that problem in a better way.
Before Generative AI, a classic persona’s journey begins with Google. If the persona is looking for an electronic device, they browse Social Media and Websites. They might also ask their tech-savvy friends.
Generative AI allows us to ask questions without wasting time searching the internet. However, we should understand how it works before relying on its answers.
Generative AI is completely different from traditional machine learning. We have encountered it in chatbots. Traditional chatbots work with related keywords. When a user asks a question, they answer based on relevant keywords, considering how many users find these answers useful. This is a significant problem for traditional machine learning, as it often provides unrelated answers when we ask questions.
Generative AI needs to understand questions before answering. Most users ask it short questions, such as ‘55-inch TV.
In this example, the product and its size were given by users. When talking to a real person, we know that additional information is needed to find the best product and make better recommendations. For instance, we might provide details about room size, whether the TV is for gaming or movies, and our budget. This means that generative AI enables hyper-personalization. As designers, we need to think more about how to understand users at an individual level in the future.
How do we train the AI?
It happens with prompt engineering. We don’t train AI by simply saying, ‘If this happens, you need to do this.’ Instead, we provide examples for the AI to learn from, and we create guidelines and principles. You need to define things like the AI’s audience, the goal of the AI, the goal of the audience, and the tone of the AI. You need to infuse it with a personality. Think of it like a person — not that it is a person, but think of it that way. If you had to hire a salesperson with no experience, you would need to train them. The same thing happens with our virtual salesperson. We have to train it and give it a personality.
Establish Limits and Privacy
Some people don’t believe in AI, and some people have never known how to interact with it. It’s important to mention the limitations and risks in a welcome message. For example, ‘Hey, I’m still learning but I’m happy to help you.’ Setting a friendly tone also helps manage expectations.
We need to account for different levels of AI proficiency and trust. Keeping users in control is crucial. Therefore, the data users provide in the chat is never used to train the model. Additionally, users can always delete their conversations, and we reset the conversation after certain session durations.
Feedback
It is crucial to improve the quality of your product. You need to establish tight feedback loops and conduct usability testing. Users can also provide direct feedback about the quality of the output and whether it meets their expectations.
