The nature of an AI product
The definition of Machine Learning (ML), which is going to be part of your AI product goes like this:
Machine learning (ML) is the science of helping computers discover patterns and relationships in data instead of being manually programmed. It’s a powerful tool for creating personalized and dynamic experiences.
This means however that unlike programmed experiences, there will always be variable in the outcome of such a product. You can optimize for it and I wrote an entire post focusing around Precision versus Recall in AI features, however it will never be foolproof.
And depending on the risk of the application your product is for, user trust can be of the utmost importance.
The risk matrix for chatbots where a medical bot on the top left will be high risk while a bot that only says Hodor will be very low risk. (source)
Having a good looking, professionally designed application in today’s world is a must to even get user trust in the first place. GUI for chatbots is pretty limited, so there your trust will be built by the quality of your copy. However, any kind of product relying on AI, be it a chatbot having an NLP algorithm, image recognition and search, speech to text, you name it, will face the above mentioned problem with variability. And this means you won’t be able to fully control the user experience. Which in turn will lead to the risk of users to lose trust in your product and lose it fast.
I would argue there are two ways you can lose user trust and one way where you don’t lose it but you should still pay attention to the use case.
When the user knows
A good example here is when a sub par off the shelf NLP algorithm doesn’t normalize for typos.
This immediately tells the user that you don’t know what you’re doing and they should not waste their time with your product. Another example I would say is a utility bill reading and payment feature in the banking app of OTP (a leading bank in Hungary). This feature promises to read the information of a utility bill.
This feature however didn’t work for me 9 out of 10 times I tried to use it. Granted, there are a lot of variables here, lighting, quality of printing and such, but when a banking app fails this bad, it is not a good sign. No wonder the local post office switched to a QR code based system by getting the utility companies to print a QR code on their bills.
This also reinforces a point many (including me) have made about AI tech. Don’t use it for the sake of it, if there is a simpler solution, opt for that. Character recognition on photos using neural nets or a tried and true QR code? Guess which will give a better user experience.
When the user doesn’t know (immediately)
There are trickier situations that can have much greater backlash from users. I’m talking about a flaw when the user doesn’t immediately realize that they didn’t get what they were expecting.
An example for this could be when you ask an AI assistant for the weather in Birmingham. There are multiple Birminghams in the world. One in the UK, one in Alabama and another one in Australia, but I’m confident there are more. Now an AI system can ask for you to specify, but not all of them do this. Some default to Birmingham, Alabama because they are US focused. This can result in you not taking your umbrella with you in Birmingham, UK because the AI said it’ll be sunny. By the time the user finds out, it is too late to do anything about it. The user experience is spoiled, user trust will be gone.
When the user doesn’t know at all
Take the Google Photos image search, something I wrote quite extensively about when discussing Precision versus Recall. Let’s say I want to find a specific food I took a picture of. At least I think I did. I’d type in food, go through the results and eventually either find it or give up. Or I could try searching for the type of food, let’s say a sundae. Google might be able to retrieve it or not. However, since I wasn’t sure if I took the picture in the first place, I’m less likely to blame the algorithm.
What do you think about user trust when it comes to AI driven products or features? Let’s discuss.