Photo by Alex Knight on Unsplash

Challenges in designing AI products.

Artificial intelligence (AI) is rapidly changing the way we live and work. AI-powered products and services are becoming increasingly commonplace, and they are having a major impact on our lives. However, designing AI products is not without its challenges. While working on my project that involved designing an AI product, I realised we as designers need to evolve our craft for AI products.

Problem

AIs like ChatGPT and Bard are novel technologies. People in the IT industry, especially engineers, may find them easy to use or they can learn how to use this. However, the challenge for designers lies in helping the general users interact with these tools and still get accurate results.

Imagine, you are asking information about normal world to these AI such as chatGPT, Bard etc. For examples, let’s say you want an itinerary personalised to your travel style. It can only curate that, if you provide it with the right set of context and prompt (prompt engineering). An average non techie persona wouldn’t spend time learning new tech, hence for them to leverage this tech, designers need to find ways to design that does not require users to learn how to interact with this tech.

Here are the prominent challenges that can be seen :

1. Human interaction with AI — endless possibilities

Yeah, general user would be scared with this too. Source

The input itself is not bound and users can communicate anything they want. This flexibility and freedom itself comes with design challenges. People can easily be lost in endless conversations without meaningful outcome.

All click based sequential inputs can be given in one go by just talking to the AI. Since users are habitual on providing input one after another through clicks and flows, providing complex input is not easy.

Without proper context, results become inaccurate and remember, they are not here to learn tech, hence if they don’t get satisfied with AI response, you can easily see them churn.

2. Language barrier

These tools use English as their form of input. Now, it becomes really important to write a prompt that is correctly structured and is unambiguous. For a normal user, without in depth knowledge of English grammar, it will need multiple tries before they can effectively communicate what they need. Till AI products come in native languages and grammar, it’s going to be a difficult task.

I believe this is the next set of challenge UX writers can take up and build logic around.

3. Long response duration

Hmmm…..is it working or just stuck at loading?

After the input is given to the AI tools, it takes time to generate a response. The larger the context and prompt is, more time it will take to generate a response. F
or example, an AI system that is used to answer customer service questions may take several minutes to respond to a question, which can be frustrating for customers who are waiting for help.

Finding a balance of accuracy and speed is going to be critical. Sometimes, delays are unavoidable

For my travel product, I designed some engagement solution for users to act while they are waiting.

Informing users what is happening and letting them spend time with travel facts.

4. Safety and security of data

Recently, we saw in news that a company’s data was leaked because an employee added the input to chatGPT. As designers, we need to educate users to not provide confidential data to AI tools which are under continuous training. We cannot stop AI tools from reinforced learning, else their output may remain in accurate.

Some best practices

Despite these challenges, there are a number of best practices that designers can follow to create AI products that are easy to use and accessible to non-tech users. These best practices include:

1. Designing for simplicity:

Fundamental design thinking still applies. AI systems should be designed with the user in mind. When to add constraints, when to let them interact more freely, these are all decisions we need to be more careful about. This also means using simple, easy-to-understand language and interfaces.

2. Forgive human errors

AI systems should also be designed to be forgiving of user errors. For eg: since AI responses are not same each time, it becomes easy that AI can go in a totally different direction than desired outcomes. Maybe we could let users restart conversation from a mid checkpoint.

This is relatively easy in chatbot context, but what about complex products, such as a travel site.

3. Providing clear instructions upfront

AI systems should come with clear instructions that explain how to use the system and what it is capable of. These instructions should be written in clear, concise language that is easy to understand.

4. Offering help and support:

AI systems should offer help and support to users who are having trouble using the system. This can be done through a help center, a chatbot, or live customer support.

What we are doing for our AI product

Since our target audience involve general audience, for us we took an approach to design which brings in familiar interactions such as clicks for input, combined with normal text based input for inputs that are not key for the end outcome. We combine it at the backend in form of a prompt for the AI’s and display result back in familiar consumable format of travel sites.

Our approach has been a combination of traditional input for key inputs combined with conversation that user can do, as if like talking to an assistant for extra bit of personalisation. We will soon be testing out our product, and I’ll share the pros and cons of this approach.

If you have more thoughts on how one can design better AI products, feel free to drop your thoughts in comments.

Thank you for your time.