The importance of humans when building AI Products
Artificial Intelligence has come a very long way in a short period. In less than fifty years humans have gone from barely discussing philosophical and mathematical implications about machine’s intelligence to having the technology for automatic image recognition (and generation) or having computers as our personal assistants. Our society is living a new revolution with the development of Machine Learning (ML) technology and we are changing habits fast.
But, may it be one day when humans are no longer required in the equation? No way! Actually, Machine Learning in a supervised approach needs humans to correct the predictions and so help the machine detect the correct patterns to enrich the results (what we call a Human-In-The-Loop). Even in the case of technology as solid as GPT-3 we end up needing humans to work with it. Through this technology the machine offers wonderful results of texts that seem to have been written by humans, but in the end, the ones who choose what to use among the results or if they are good enough are we, mere mortals. How about when we talk about unsupervised AI? Then, the machine finds patterns by its own through statistical algorithms, right? Yes. But think about this: whose problem should be solving your AI product? A human one, so humans will always be needed to construct the technology around and for them. Don’t expect ML to figure out what problems to solve!
The Importance of Humans When Building AI Products
This is something that also happens with non-technical products. When you design a solution it should be focused on relieving your client’s pains. One of the main causes why a product fails to be successful in the market is because that product solved a problem that none of its target users had, or that product’s usage experience was not positive enough for them.
Now imagine this applied to a technical product company, where the complexity increases. You can have an awesome team working in the core of the technology, data scientists working on complex neural networks or engineers with amazing skills to build software that can process thousands of records of data ultra fast. But what problems are we solving with all these resources? And how are we presenting the product that will help the user? Will they understand it at all?
What we need in our team is an MLUX (Machine Learning User Experience) designer who can work with Human-Centered Design. That is what I do!
What is Human-Centered Design? What Can it Offer to an AI Product?
Human-Centered Design is an approach for problem-solving that focuses on empathizing with humans in every step of the process. It is the perfect bridge between the possibilities that technology can offer and the real needs of the users.
A UX designer in your team will help you empathize with users by
- Reducing noise
Technological products can get awfully complex when they work with huge amounts of data, cumbersome analytics, integrations with many platforms, etc. But if we think of the users during the design process, the data displayed can be restructured thus reducing the friction and eliminating unnecessary turns to guide the users through an intuitive flow.
- Providing useful data for your clients and using their same language
Likewise, many of the results and insights that our product is offering are not always going to turn out useful to our target clients. Know your users, know their market, and give them just the data and functionalities they need, labeled with the terms they use in their work.
- Fine-tuning the product or your algorithms based on users feedback
We should not only research our users before releasing a product, the improvements must be continuous. Users can give you insights about removing unnecessary interactions that you first though interesting, adding new features that would help them even further, or focusing the technology in business areas that were initially out of the scope.
- Giving results in a comfortable way based on patterns that users have learnt
It is similar to talking the same language as your users. They have learnt established usage patterns in many other products and they are unconsciously expecting similar behavior in yours. There is no need to reinvent the wheel, don’t make the user learn a strange interaction when the established are the most intuitive ones.
- Helping users to tell predictions from real data
In AI we work a lot with predictions, insights that machines have calculated based on historical data. In many cases, those recommendations or predictions need more accuracy and users should know it before taking actions based on them.
- Giving the user control
Users don’t easily trust fully automatic processes. Not only automation with AI but even smaller interactions that we may think are helping by reducing noise. Imagine a form that autocompletes based on the user’s preferences. We should always give the user control before submitting automatically that form, or in other cases editing options to change results.
- Identifying weird edge cases and potential errors.
Testing is key to avoid disasters and it can make you correct errors that are yet to come. Also by testing you may identify difficulties in processes (like people not being able to upload a file because their format is completely different from the requested one). Test continuously, because weird edge cases are always there and they may appear when less expected to ruin your product.
OK, Designers Rule. But, What is the Workflow to Design the Product Inside a Tech Team?
At Lang.ai we are continuously building new functionalities or improving the existing ones. These changes may come from either our core technology team or feedback obtained through testing with users.
When the idea comes from innovation in our technology, we study this new feature having in mind what it is going to offer to our users, taking into account their needs. If the idea is a request from the client, we research it to check if that is the real problem or it comes from a different friction point. Also, we study the scalability of the product to be useful to other clients inside the same market.
Then, we start working horizontally by involving every team. From Product Design, we speak with our data scientists, engineers, and even the sales team to understand every aspect of the product life and get different points of view about the impact, results, and costs of the new functionality.
With the feedback of each team, we work in user’s research exercises to identify key gains for them and to map the stages of the work and issues to come.
Through an iterative prototyping process, involving user’s tests and the participation of the technical team, the designers develop a solid architecture and propose a wireframe of an intuitive interaction flow. By involving all the teams in the design process we ensure consistency and work concurrently in the same direction.
In summary, Human-Centered Design is the key to reach a great user experience. AI technological products should be built thinking about human’s pains. So, involve designers in each one of the stages to build your product. Make sure your data scientists, engineers, and product designers work together as a team throughout the process, you won’t regret it!
