This article is transferred fromHow to build AI applications users can trust"

The original text is in English, so if you use Google machine translation, it will not be smooth to read.But it does not affect understanding.The purpose of the article has been achieved, so there is no need to worry too much about the quality of the translation.

In order to work effectively,Algorithm requires user data-Usually continuous to help improve and improve the experience.To get user data, you need users.In order to attract users, especially persistent users who trust you with your data, you need to provide options that suit their comfort levels while still allowing them to make changes in the future.Essentially, to gain user acceptance, you need to take a two-step approach: let users know what data you want to collect and why you want to collect data, and let them control the collection.

Step 1: Provide ongoing transparency

The first step in finding balance is to provide knowledge to your users.Users need to know what data is being collected and how to use it before deciding to interact with the application.Industry pressureHas been developed in this direction:Apple recently announcedAll its applicationsPrivacy label, Which will make users more aware of the data collected when using their applications.The Microsoft CaptionBot below is a good example of how to provide users with an easy-to-understand overview to understand the status of their data behind the scenes.

Microsoft's CaptionBot provides clear information about data storage, distribution, and use, as well as an easy-to-understand overview of the system that enables AI captioning tools to run in the background.

Microsoft's CaptionBot provides clear information about data storage, release, and use, and provides an easy-to-understand overview. These systems can make AI captioning tools work in the background.

Health app belowAdaAn example of how to prevent users from being confused about data collection choices.

Above: Health app ada first explains the logic behind its input selection, so users can understand how their input affects the application and its ability to perform the required actions.

Sharing this information in advance will not only give users a sense of empowerment and help you build trust in your experience over time, but also provide you with an opportunity to help them understand how sharing their data can improve their experience. And how to reduce their experience without reducing shared resources.The data.By providing users with information that helps them understand what happens when they share data, we also provide them with tools to understand how this communication benefits them, thereby enhancing users’ enthusiasm for using the app.advertise

In addition to these details, it is also important to provide users with information when they use the application.Sharing information about the effectiveness of the algorithm (the likelihood of the algorithm succeeding on the task) and the algorithm's confidence (how certain the algorithm is in the results produced) can make a big difference in how users comfortably use these technologies.As we all know, comfort plays an important role in adoption and participation.

Please consider the following Microsoft's trust in certain products.

Above: When the algorithm makes the "best guess", the confidence level is displayed (in the first image using Microsoft's Bing Image Search, the score is 0 to 1, and in the second image from Microsoft's Celebs Like Me In the picture, the percentage score) helps users understand how much trust they should have in the results of the algorithm.

Users should also be made aware of certain operations and mechanisms.Therefore, it is important to confirm when these mechanisms work or "think", when they are handed over from algorithms to humans, or when data will be shared with third-party systems or stored for potential future applications.Constantly providing opportunities to build awareness and understanding of your application will increase the trust in using it.The more users trust it, the more likely they will continue to use it over time.

Step 2: Transfer control

Even if the advantages of the application are enough to attract users to choose to use, users may not always want to use AI.In some cases, they want to withdraw or limit the number of uses of the technology.How do we empower them to choose the number of AIs they interact with at the most important moments?Again, the combination of pre- and semi-regular check-in works well here.advertise

When informing users what data you are collecting and how to use them, if the use case does not meet their needs, give them the opportunity to choose not to share certain types of data.Where possible, provide them with a series of hierarchical options (what you get when you enable all data sharing instead of partial data sharing instead of all data sharing) so that they can choose the option that suits them best.

Consider the following from the food ordering appRExample.

Above: Ritual, a popular food ordering app, allows users to choose to share certain data and tells users how opt-out will affect the functionality of the app.

Whenever you add a new product feature or users use the feature for the first time, prompt them to view or change their data sharing level.Content that previously seemed irrelevant to them may be very attractive to the proposed new use cases.If a new type of data is being collected, please prompt again.advertise

The last way to provide control: give users the opportunity to directly control the application.This means that you can simply contact your users from time to time to find out which features they like, which features they dislike, and what they want from the app.Or, more importantly, it can be used as part of the application itself.Can the user adjust the level of certain inputs to produce different results (for example, for a recommendation algorithm, weight one input over another)?Can they take a step back or manually cover certain aspects?Handing over controls as literally as possible can make users feel authorized by the application, rather than being intimidated by the application.

EatThe AI ​​Therapy app of the company is a good illustration of how to provide control to the user.

Above: Youper’s AI therapy app does not require users to set all the parameters at the beginning, but instead provides regular opportunities to refine their experience as they continue to interact with the app (and explains why it can help They do it).

Every application is different, so each method of authorizing users will be different.However, when you provide a transparent way to understand how and why the system uses existing information, and give consumers the opportunity to choose not to share certain information, you create a space of trust.When your users trust you, they will be more inclined to share the data needed to make your products and services active.