This chapter will discuss why building the right level of trust with users is critical when designing AI products. We will look at different strategies for building and maintaining trust and opportunities for trust-building in your user experience. We will also look at the advantages, disadvantages, and considerations for imbuing personality and emotion in your AI.
I’m a stickler about having chai and biscuits in the morning. Every week I go to the neighborhood grocery store to buy a regular packet of Britannia Marie Gold biscuits. When paying at the counter, I don’t check prices; I simply hand out my card. The store owner keeps these biscuits stocked for me and sometimes even gives me a discount. Over the past few years, I have come to trust this little store.
Almost all successful human interactions are built on trust. You often spend more money on brands that you trust over cheaper alternatives, which is why companies spend a lot of money on building their brand. When you go to a doctor for a health checkup, the underlying assumption is that you can trust their diagnosis. Most of us prefer to deposit our savings with a trusted bank. We don’t make large purchases on websites we don’t trust. We tend to leave our kids to play in safe places with trustworthy people.
A successful team of people is built on trust, so is a team of people and AI.
Trust in AI
We can use AI systems in organizations to generate predictions, provide insights, suggest actions, and sometimes even make decisions. The output of AI systems can affect different types of stakeholders directly or indirectly. AI systems are not perfect. They are probabilistic and learn from past data. Sometimes they make mistakes, which is why we require humans to oversee them. For AI products to be successful, people who use them or are affected by them need to trust these systems.
Components of User Trust
- 1.
Competence
- 2.
Reliability
- 3.
Predictability
- 4.
Benevolence
Competence
Competence is the ability of the product to get the job done. Does it improve the experience or address the user’s needs satisfactorily? A good-looking product or one with many features that do not fulfill user needs is not competent. Strive for a product that provides meaningful value that is easy to recognize.2 Google Search is an example of a competent product. It generally offers satisfactory results for your questions.
Reliability
Reliability indicates how consistently your product delivers on its abilities.3 A reliable product provides a consistent, predictable experience that is communicated clearly. A product that performs exceptionally well one time and breaks down the next time is not reliable. Apple’s iPhone is an example of a reliable product. It might not carry all the features its competitors have, but you can reasonably trust the ones it does have.
Predictability
A predictable interface is necessary, especially when the stakes are high. If the user comes to your product to perform critical, time-sensitive tasks, like quickly updating a spreadsheet before a client presentation, don’t include anything in your UI that puts habituation at risk.4 A probabilistic AI-based solution that can break the user’s habit is not ideal in such cases. However, suppose you think that users have open-ended goals like exploration. In that case, you can consider a dynamic AI-based solution that sometimes breaks user habits, for example, selecting a movie from dynamic AI-based suggestions.
Benevolence
Benevolence is the belief that the trusted party wants to do good for the user.5 Be honest and upfront about the value your users and your product will get out of the relationship. Patagonia is a clothing brand that makes jackets that does a great job with benevolence. While their products can be expensive, they encourage people to reuse and repair their Patagonia clothes, giving a percentage of their sales for environmental causes. The company is upfront about its value to the customer.
Trust Calibration
Users can sometimes overtrust or distrust your AI system, which results in a mismatch of expectations. Users may distrust your AI when trust is less than the system’s capabilities. They may not be confident about your AI’s recommendations and decide not to use them. Users rejecting its capabilities is a failure of the AI system. Overtrust happens when user trust exceeds the system’s capabilities, leading to users trusting an AI’s recommendation when they should be using their own judgment. For example, users overtrusting the suggestions of a stock prediction service can lead to a financial loss. Overtrust can quickly turn into distrust the next time this person uses the service.
Product teams need to calibrate user trust in the AI regularly. The process to earn user trust is slow, and it’ll require proper calibration of the user’s expectations and understanding of what the product can and can’t do.6
How to Build Trust?
AI systems are probabilistic and can make mistakes. It is the job of product creators to build trustworthy relationships between the AI and its users. Building trust is not about being right all the time; it is about integrity, accepting mistakes, and actively correcting them. Users should be able to judge how much they can trust your AI’s outputs, when it is appropriate to defer to AI, and when they need to make their own judgments. There are two essential parts to building user trust for AI systems, namely, explainability and control.
Explainability
If we don’t understand how AI systems work, we can’t really trust them or predict the circumstances under which they will make errors.7 Explainability means ensuring that users of your AI system understand how it works and how well it works. Your explanations allow product creators to set the right expectations and users to calibrate their trust in the AI’s recommendations. While providing detailed explanations can be very complicated, we need to optimize our explanations for user understanding and clarity.
Control
Building trust is not about being right all the time; it is about integrity, accepting mistakes, and actively correcting them.
Explainability
We generally don’t trust those who can’t explain their decisions and reasoning. The same is true for AI systems that show recommendations, provide insights, or make decisions on behalf of people. Explaining their reason is especially important as AI systems significantly impact critical decisions like sanctioning loans, predicting diseases, or recommending jobs. When AI systems are used to help make decisions that impact people’s lives, it is particularly important that people understand how those decisions were made.9
But even people can’t always explain how they made certain decisions. You can’t look “under the hood” into other people’s brains. However, humans tend to trust that other humans have correctly mastered basic cognitive tasks. You trust other people when you believe that their thinking is like your own. In short, where other people are concerned, you have what psychologists call a theory of mind—a model of the other person’s knowledge and goals in particular situations.10 We don’t yet have a theory of mind for AI systems.
When AI systems are used to help make decisions that impact people’s lives, it is particularly important that people understand how those decisions were made.11
Who Needs an Explanation?
- 1.
Decision-makers
- 2.
Affected users
- 3.
Regulators
- 4.
Internal stakeholders
Decision-Makers
Decision-makers are people who use the AI system to make decisions. A decision-maker can be a bank officer deciding whether to sanction a loan or a radiologist diagnosing an ailment based on an AI’s recommendation. Decision-makers need explanations to build trust and confidence in the AI’s recommendations. They need to understand how the system works and sometimes require insights to improve their future decisions. Most decision-makers need simplified descriptions and have a low tolerance for complex explanations.
Affected Users
These are people impacted by recommendations that your AI system makes, like a loan applicant or a patient. Sometimes the decision-maker and the affected user are the same people. For example, giving Netflix your preferences is a decision, and getting movie recommendations is the effect. Affected users seek explanations that can help them understand if they were treated fairly and what factors could be changed to get a different result.14 They have a low tolerance for complex explanations, and outcomes need to be communicated clearly and directly.
Regulators
These are people who check your AI system. Regulators can be within the organization in the form of internal auditing committees or external in the form of government agencies. Regulators can enforce policies like the European Union’s General Data Protection Regulation (GDPR) and may require AI creators to provide explanations for decisions. Regulators need explanations that enable them to ensure that decisions are made in a fair and safe manner. They may sometimes use explanations to investigate a problem. Explanations for regulators can show the overall process, the training data used, and the level of confidence in the algorithm. These stakeholders may or may not have a high tolerance for complex explanations.
Internal Stakeholders
They are people who build the AI system, like ML engineers, product managers, designers, data scientists, and developers. Internal stakeholders need explanations to check if the system is working as expected, diagnose problems, and improve it using feedback. They generally have a high tolerance for complexity and need detailed explanations of the system’s inner workings.
Explainability and trust are inherently linked.
Guidelines for Designing AI Explanations
A mental model is a user’s understanding of a system and how it works. Once users have clear mental models of the system’s capabilities and limits, they can understand how and when to trust it to help accomplish their goals.15 A mismatch between the user’s understanding of how the system works and how it actually works can lead to frustration, misuse, and even product abandonment. Building the right mental models is key to user trust in your AI, and you can use explanations to calibrate it.
- 1.
Make clear what the system can do.
- 2.
Make clear how well the system does its job.
- 3.
Set expectations for adaptation.
- 4.
Plan for calibrating trust.
- 5.
Be transparent.
- 6.
Build cause-and-effect relationships.
- 7.
Optimize for understanding.
Make Clear What the System Can Do16
Make sure your users understand the capabilities of the system. Users need to be aware of the full breadth of functionality of your feature. Providing contextual information on how the AI system works and interacts with data can help build and reinforce user trust. For example, if there is a search functionality within a grocery application, make sure all search possibilities are enumerated, like “search for fruits, vegetables, and groceries.” Clarify how input influences the results.
Make Clear How Well the System Does Its Job
AI systems are probabilistic. Help users understand when your AI system performs well and when and how often it makes mistakes. Set expectations of performance out of the box and clarify mistakes that will happen. In some cases, it is better to use uncertain language like “We think you’ll like this book” or indicate a level of confidence. When it is not feasible to provide this information directly, consider providing a help context that is easy to access or a universal help command like “Say help for more details.”
Don’t leave out setting expectations of performance and updating them when things change. For example, an increase in traffic can lead to higher wait times for a taxi service. AI systems operate with uncertainty. If your users expect deterministic behavior from a probabilistic system, their experience will be degraded.17
Set Expectations for Adaptation
Plan for Calibrating Trust
AI systems will not perform perfectly all the time. Users shouldn’t implicitly trust your AI system in all circumstances but rather calibrate their trust correctly.18 People may sometimes overtrust your system for something it can’t do, or they may distrust the AI’s output. Either of these cases causes a mismatch in expectations and can lead to the failure of the product. You can help users calibrate their trust in the AI by explaining what it can and can’t do or displaying its output confidence. This uncertainty can lead to some level of distrust initially. For example, telling users that the prediction may be wrong can lead to them trusting the system less, but over time as the AI improves, they can start relying on your product more with the right level of trust.
The key to success in these human-AI partnerships is calibrating trust on a case-by-case basis, requiring the person to know when to trust the AI prediction and when to use their own judgment in order to improve decision outcomes in cases where the model is likely to perform poorly.19 It takes time to calibrate trust with users. Plan for trust calibration throughout the user journey over a long time. AI changes and adapts over time, and so should the user’s relationship with the product. For example, a music recommendation service may not be very accurate for new users. But over time, it improves by learning user preferences. Explaining that the service learns your preferences over time and will get better at recommendations can help users calibrate their trust at different stages of using the product. Planning for calibrating trust can help you in designing better explanations and build more effective human-in-loop workflows.
- 1.
Communicating what data is used to train the AI. A speech recognition system built on American speaker data might not perform well for Indian users.
- 2.
Allowing users to specify language and genre preferences in a movie recommendation service.
- 3.
Allowing users to try the product in a “sandbox” environment can help calibrate trust.
- 4.
Displaying accuracy levels or a change in accuracy when recognizing product defects in an assembly line.
- 5.
An ecommerce website showing reasons for product recommendations like “Customers who bought this also bought…” or “Similar products.”
Plan for trust calibration throughout the user journey over a long time. AI changes and adapts over time, and so should the user’s relationship with the product.
Be Transparent
It is difficult to trust those who appear to be hiding something. Transparency means operating in ways that make it easy for others to understand the actions of a system. It implies acting out in the open. There are legitimate reasons for organizations to have trade secrets or proprietary information. But when it comes to pertinent information of customers, privacy, data use, bias, other stakeholders, or the efficacy of the algorithm, transparency is central to earning trust. This is especially true for AI operating in high-stakes environments like medicine or driverless cars. A lack of transparency increases the risk and magnitude of harm when users do not understand the systems they are using or there is a failure to fix faults and improve systems following accidents.20
Many AI systems will have to comply with privacy laws that require transparency about the collection, use, and storage of data and mandate that consumers have appropriate controls so that they can choose how their data is used.21 Product teams need to ensure that users are made aware of any data that is collected, tracked, or monitored and that it’s easy for them to find out how the data is collected, whether via sensors, user-entered data, or other sources.22 In necessary cases, use non–black box models so intermediate steps are interpretable and outcomes are clear, providing transparency to the process.23
The level of transparency will be different for different stakeholders. Your explanations can help users understand why the AI made certain predictions through transparency of data, its use, and the underlying algorithm. Good explanations can help people understand these systems better and establish the right level of trust.
Transparency means operating in ways that make it easy for others to understand the actions of a system.
Build Cause-and-Effect Relationships
People understand faster when they can identify a cause-and-effect relationship between their actions and the system’s response. Sometimes the perfect time to show an explanation is in response to a user action. Users might be confused if an AI system completes an action but does not respond about its completion. If it reacts unexpectedly, an explanation can help in calibrating or recovering trust.
On the other hand, when the system is working well, responding to users’ actions is a great time to tell the user what they can do to help the system continue to be reliable. For example, a user looks for personalized recommendations for breakfast places on a restaurant recommendation service like Zomato. If they only see recommendations for places they rarely visit or that don’t match their earlier preferences, they might be disappointed and trust the service a bit less. However, suppose the app’s recommendation includes an explanation that the system only recommends restaurants within a 2 km driving distance and that the user is standing in the heart of Cubbon Park in Bangalore. In that case, trust is likely to be maintained. The user can see how their actions affect the suggestion.
Building trust is a long-term process. A user’s relationship with your product can evolve over time through back-and-forth interactions that reveal the AI’s strengths, weaknesses, and behaviors.
Building trust is a long-term process. A user’s relationship with your product can evolve over time through back-and-forth interactions that reveal the AI’s strengths, weaknesses, and behaviors.
Optimize for Understanding
Different stakeholders in your AI system need different levels of explanations. Your AI explanations need to be understandable. Simply publishing algorithms underlying AI systems or providing a dump of data used is not meaningful. A list of a billion operations is not an explanation that a human can understand.24 The complexity of your explanations needs to be tailored for your users.
It can be challenging to explain how your AI system works. Providing a detailed explanation can sometimes confuse users. In such cases, the best approach is not to attempt to explain everything—just the aspects that impact user trust and decision-making.25 Sometimes, it is required by law in certain regions like the EU for this information to be communicated “in a concise, transparent, intelligible and easily accessible form, using clear and plain language.”26
The ability to explain can also determine the fate of your AI product. Product teams are responsible for making important judgment calls about which AI technologies might best be deployed for specific applications. A huge consideration here is accuracy vs. “explainability.”27 Deploying a deep learning system in high-stakes environments like medicine or driverless cars that we can’t explain is risky. Such algorithms may require periodic regulatory scrutiny and sometimes might not be worth the cost of maintenance.
Explainability is critical for building user trust. Explaining your AI system so people can actually understand it is a fundamental human-centered AI design challenge.
Types of Explanations
- 1.
What did the system do?
- 2.
Why did the system do it?
- 3.
Why did the system not do this?
- 4.
What would the system do if this happened?
- 5.
How does it do it?
- 6.
What is the overall model of how the system works?
- 7.
What data does the system learn from?
- 8.
How confident is the system about a prediction or an outcome?
- 9.
What can I do to get a different prediction?
- 10.
What changes are permitted to keep the same prediction?
- 1.
Data use explanations
- 2.
Descriptions
- 3.
Confidence-based explanations
- 4.
Explaining through experimentation
- 5.
No explanation
Data Use Explanations
- 1.
What data does the system learn from?
- 2.
Why did the system do it?
- 3.
How does it do it?
Guidelines for Designing Data Use Explanations
- 1.
Explain what data is used.
Ensure that users are made aware of any data that is collected, tracked, or monitored and that it’s easy for them to find out how the data is collected, whether via sensors, user-entered data, or other sources.30 It can also help to tell users the source of data so that they don’t overtrust or distrust the system. For example, a navigation service suggesting the time to leave for a meeting can tell users that it derived this insight using data from their calendar, GPS, and traffic information.
- 2.
Explain how the data is used.
Your AI system will use data to provide a recommendation or make a decision. Users need to understand how the AI system uses the data. For example, a marathon training app that asks users to provide their location data explains that the AI will use this information to generate routes and calculate distance. Explaining how data is used can help build user trust.
- 3.
Explain what’s important.
Providing a list of data sources or a dump of data is not understandable. Your data use explanations need to highlight the relevant data during an interaction. Sometimes, you can surface which data sources had the greatest influence on the system output. Identifying influential data sources for complex models is still a growing area of active research but can sometimes be done. In cases where it can, the influential feature(s) can then be described for the user in a simple sentence or illustration.31 For example, an app that identifies cat breeds can show similar instances where it correctly identified the breed. In this case, it need not show you all samples across all breeds.
- 4.
Highlight privacy implications.
Your AI systems should respect user privacy. People will be wary of sharing data about themselves if they are not confident of how their data is used, stored, and protected. Sometimes users can be surprised by their own information when they see it in a new context. These moments often occur when someone sees their data used in a way that appears as if it weren’t private or when they see data they didn’t know the system had access to, both of which can erode trust.32 For example, a service that uses a person’s financial data to predict loan eligibility without explaining how it has access to this data is not trustworthy. Privacy is a key pillar of building trust with users. Communicate privacy and security settings on user data. Explicitly share which data is shared and which data isn’t.33 For example, a social music streaming service that shares what you are listening to with your friends needs to communicate explicitly that your friends can see your listening activity. Additionally, users should also be able to opt out of sharing this information.
Types of Data Use Explanations
- 1.
Scope of data use
- 2.
Reach of data use
- 3.
Examples-based explanations
Scope of Data Use
Scope refers to the type and range of data used to provide a result. When using scope-based explanations, show an overview of the data collected about the user and which aspects of the data are being used for what purpose.34 Sometimes users may be surprised to see their information shown in the product; this can erode trust. To avoid this, explain to users where their data is coming from and how it is used. Your data sources need to be part of the explanation. For example, a personal assistant may offer to book a cab for your appointment. In this case, the personal assistant needs to communicate that it knows about your appointment because your calendar is linked and you have configured your preferred mode of transport as an Uber. However, remember that there may be legal, fairness, and ethical considerations for collecting data and communicating about data sources used in AI.35 Sometimes this may be legally required in certain regions.
Reach of Data Use
Reach-based explanations tell users if the system is personalized to them or a device or if it is using aggregated data across all users.36 Your sources of data need to be part of reach-based explanations. For example, when Spotify creates personalized playlists for users, the description often explains how it was generated: “Your top songs 2020—the songs you loved most this year, all wrapped up.” When Amazon’s recommendation system on a product page shows aggregated suggestions, the explanation also explains the logic for the recommendation in a user-friendly manner: “Customers who read this book also read…”
Examples-Based Explanations
Sometimes it is tricky to explain the reasons behind an AI’s output. In examples-based explanations, we show users examples of similar results or results from the training data relevant to the current interaction. Examples can help users build mental models about the AI’s behavior intuitively. These explanations rely on human intelligence to analyze the examples and decide how much to trust the classification.37 Examples-based explanations can be generic or specific.
Descriptions
- 1.
What did the system do?
- 2.
Why did the system do it?
- 3.
What is the overall model of how the system works?
- 4.
How does it do it?
Guidelines for Designing Better Descriptions
- 1.
Explain the benefit, not the technology.38
As product creators, we are often excited about the underlying technologies, especially if we’ve solved a particularly hard problem. But most users don’t care about the technology; they just want to get a job done. They don’t necessarily need to understand the math behind an algorithm in order to trust it. Within the product experience, it is much better to help users understand how the technology benefits them. For example, in a spam filtering application, it is better to say “We help you remove spam from your inbox” than saying “We’ve created a sophisticated email filtering technique that categorizes email into spam and non-spam by using extensive training data.”
If you think your users would be interested in the underlying technology, you can always provide more details with tooltips and progressive disclosure. You can also provide this information in your marketing communications. If you talk about the AI system, focus on how it benefits the user and not how you built it.
- 2.
Make it understandable.
Make sure that your descriptions are understandable to your users. Try to make your descriptions less like a user manual with technical jargon and more like an aid to decision-making. For example, typing “area of Poland” into the search engine Bing returns the literal answer (120,728 square miles) along with the note “About equal to the size of Nevada.” The numeric answer is the more accurate, but the intuitive answer conveys the approximate size of Poland to far more people.39
- 3.
Account for situational tasks.
It is essential to consider the risks of a user trusting an inaccurate suggestion. You can tailor the level of explanation by accounting for the situation and potential consequences. For example, in a logistics system, if the items being delivered are low stakes, like clothes or toys, you can get away with explaining the tentative time of delivery. However, suppose the items being delivered are high stakes, like surgical equipment or ventilators. In that case, you should let users know the tentative delivery time as well as limitations of your system, like the data refreshing every hour or the possibility of a delay. Additionally, you can give users more control by enabling them to contact the delivery agent or shipping provider.
- 4.
Explain what’s important.
While interacting with your AI product, users may not need all the information. In fact, providing all the information might sometimes be detrimental. Explain only what’s important in a given scenario, intentionally leaving out parts of the system’s function that are highly complex or simply not useful.40 For example, a user trying to select a movie to watch on an AI-based streaming service might find it irritating if the system overexplains its working. However, you can explain the system’s entire workings through progressive disclosure or other channels like blog posts or marketing communications.
- 5.
Use counterfactuals (sometimes).
Using counterfactuals is a way of telling users why the AI did not make a certain prediction. They are the answers to questions like “What would happen if ____?” or “Why did the system not do this?” Counterfactuals are not always helpful, but sometimes they can significantly improve your explanations. For example, an AI service that predicts the selling price of your car can provide insights like “You would get 5% more if the paint was not chipped” or “You would get 10% less if you wait another three months.” You need to determine if you should use counterfactuals through trial and error.
Types of Descriptions
Designing user-centered explanations for your AI can significantly improve your product’s user experience. In general, we can categorize descriptions into two types.
Partial Explanations
- 1.
Generic explanations
General system explanations talk about how the whole system behaves, regardless of the specific input. They can explain the types of data used, what the system is optimizing for, and how the system was trained.41 For example, a marathon training app can say “This app uses your height, weight, and past runs to find a workout” when suggesting exercises.
- 2.
Specific explanations
They explain the rationale behind a specific AI output in a manner that is understandable. For example, a recipe recommendation system can say, “We recommended this recipe because you wanted to make a light meal with broccoli, pasta, and mushrooms,” or a dog recognition system can say, “This dog is most likely a German shepherd because of XYZ features.” Specific explanations are useful because they connect explanations directly to actions and can help resolve confusion in the context of user tasks.42
Full Explanations
These are detailed explanations of how your AI system works or why it made a particular prediction. These can be in the form of research papers, blog posts, open source code, etc. In most cases, it is not advisable to give full explanations within your product’s user experience. Such explanations might be irrelevant for the task and can also confuse users. If you really want to provide full explanations, you can do it through progressive disclosure, tooltips with external links, etc. The best place for full explanations can be on your product’s blog, marketing collaterals, the company’s landing page, or even the careers page to attract the right candidates.
Confidence-Based Explanations
Confidence-based explanations are unique to probabilistic systems like data science reports or AI products. A confidence level is a statistical measurement that indicates how confident the AI system is about a particular outcome.43 Confidence-based explanations provide answers to the following question: “How confident is the system about a prediction or an outcome?”
Confidence is a readily available output from an AI system, the value of which ranges from 0 (no confidence) to 1 (complete confidence). Because these systems are based on probability, in most real-world scenarios, you will never get a definite 0 or 1. Sometimes, we convert these values into percentages. Displaying confidence levels can help users gauge how much to trust an AI’s output. Rather than describing why the AI came to a particular decision, confidence-based explanations tell users how certain the AI is about the output. Sometimes it shows the alternatives the AI model considered.
Let’s assume that we build an AI system that detects donuts from images. We categorize results into high, medium, and low levels of confidence. A result with a value less than 40% has low confidence, while one with a value of more than 80% is categorized as high confidence, and anything between 40% and 80% has a medium level of confidence. The team building the AI decides these threshold values, which can vary across systems. Using these confidence levels and categories, we can customize the system’s output. For example, if you show an image of a donut that results in a high level of confidence, we can design it to say, “That’s a donut.” If the confidence about the image is medium, then it can say, “Maybe that’s a donut.” When you show the donut detector a picture of pasta that results in a low confidence, it can say, “I don’t think it’s a donut.”
Guidelines for Designing Confidence-Based Explanations
- 1.
Determine if you should show confidence.44
Displaying confidence values can sometimes help users calibrate their trust in the AI’s output. But confidence values are not always actionable. In many cases, it is not easy to make confidence values intuitive. Even if you’re sure that your user has enough knowledge to properly interpret your confidence displays, consider how it will improve usability and comprehension of the system—if at all.45 To avoid this, test with users if showing confidence is beneficial. To assess if showing confidence improves user trust and helps people make decisions, you can conduct user research on your target stakeholders. There is always a risk wherein showing confidence values can be irrelevant, distracting, or confusing and even have negative consequences.
Here are some cases when you should not indicate confidence:
- a.
Consider not showing the confidence value if it doesn’t help with decision-making.
- b.
Don’t show the confidence value if it misleads users, causes confusion, or creates distrust. An inaccurate high confidence result can cause users to trust an AI decision blindly. If the confidence value is misleading for certain users, reconsider how it is displayed, explain it further, or consider not showing it at all.
- 2.
Optimize for understanding.
Showing a confidence value should help users in making a decision and calibrate the right level of trust. Users can’t always gauge if a 75% confidence is high or low or enough to make a decision. Using friendly terms like high or low can be more beneficial than saying 95% confidence. Sometimes, showing too much granularity can be confusing. Which option should the user choose between 84.3% and 85% confidence? Is the 0.7% difference significant?
Statistically, information like confidence scores can be challenging for different users to understand. Because different users may be more or less familiar with what confidence means, it is essential to test your confidence displays early in the product development process. We can display confidence levels in many ways. You can choose to reinterpret confidence into statements like “I’m not sure if this is a donut,” “I’m pretty sure this is a donut,” or “Maybe this is a donut” for a low, high, or medium confidence, respectively. Sometimes you can modify the user interface based on the level of confidence.
- 3.
Balance the influence of confidence levels on user decisions.
Product teams need to understand how much influence displaying confidence in the AI suggestion has on the user’s decision. If showing confidence does not significantly impact the user’s decision-making, presenting it might not make sense. On the other hand, if confidence has a disproportionate influence, it can lead to overtrust. Your confidence-based explanations should help users calibrate the right level of trust in the AI.
Types of Confidence-Based Explanations
- 1.
Categorical
- 2.
N-best results
- 3.
Numeric
- 4.
Data visualizations
Categorical
N-Best Results
N-best means showing a specific or “n” number of best results that your AI outputs. We display multiple results in this method rather than presenting only one result with a particular confidence level. The confidence values of those results may or may not be indicated. This type of explanation is advantageous in low-confidence situations or in cases where it is valuable to provide users with alternatives. Showing multiple options prompts the user to rely on their own judgment. It also helps people build a mental model of how the system relates to different options.46 Determining the number of alternatives will require testing and iteration.
Numeric
In this method, numeric values of confidence are shown, often in the form of percentages. However, using numeric confidence explanations is risky since it assumes that users have a baseline knowledge of probability and an understanding of threshold values. Showing numeric explanations can even confuse users. It is difficult to gauge if an 87% confidence value is low or high for a particular context. Users might trust an AI that predicts a song they like with 87% confidence, but they would be wary of trusting it to drive their kids to school.
- 1.
Specific explanations provide users with a prediction along with the AI’s confidence value. For example, when you select a movie on Netflix, it shows a percentage match indicating your likelihood of enjoying it.
- 2.
General explanations present an average confidence of the system, for example, this app recognizes dogs with a 90% accuracy.
Data Visualizations
When it comes to indicating confidence values, data visualizations are graphical indications of certainty over a span of time, types of results, or any other metric. For example, a stock price or a sales forecast could include elements indicating a range of possible outcomes based on your AI’s confidence. Your data visualizations can be static or interactive. Keep in mind that many data visualizations are best understood by expert users in specific domains. However, it is safe to assume that many people understand common visualizations like pie charts, bar graphs, trend lines, etc. Using a visualization and choosing the right type will require user research, testing, and iterations.
Explaining Through Experimentation
- 1.
What would the system do if this happened?
- 2.
What can I do to get a different prediction?
- 3.
What changes are permitted to keep the same prediction?
You need to be intentional about letting users interact with the AI and test its limits. Building an experimentation-based explanation can be time-consuming and require a lot of effort. Developing such experiences will need multiple rounds of user testing.
Guidelines to Design Better Experimentation Experiences
- 1.
Allow experimentation with specific features.
Asking users to try out the entire system as a way of experimentation is not explainable. Doing this can even confuse users about your product’s value. You need to tie your explanations to a specific output. For example, changing the lighting in a document scanning application can give a better outcome, or in a search engine, modifying your search query can provide better results. Point users to specific features where they can quickly understand the value of your product. Otherwise, they may find the system’s boundaries by experimenting with it in ways it isn’t prepared to respond. This can lead to errors, failure states, and potentially erosion of trust in your product.47
- 2.
Consider the type of user.
A user’s willingness to experiment will depend on their goals. For example, an average consumer might enjoy trying out different AI-enabled features on a smart TV as part of the explanation, but a person buying hundreds of TVs for an enterprise might find this frustrating.
- 3.
Build cause-and-effect relationships.
People learn faster when they can build a cause-and-effect relationship between their actions and AI’s outputs. Within an experimentation environment, a specific user action should generate a response. The AI can respond to the success or even failure of the system in the form of errors. Providing no response can keep users waiting and frustrated and can erode trust in your system. For example, a smart speaker that offers no response and keeps the user hanging despite an error is worse than providing an error response.
No Explanation
In some cases, there is no benefit in explaining the AI’s function. If the way an AI works fits a common mental model and matches user expectations for function and reliability, then there may not be anything to explain in the interaction.48 For example, in a camera application that automatically adjusts to external lighting, it might be irrelevant to explain every time it adjusts the image. Your users might already have a mental model of when and how that happens. Explanations can sometimes get in the way of a user’s actions. In the case of an AI writing assistant, it can be distracting if the system explains every time it corrects a word while you are drafting an email. It would also be wise to avoid explanations if they reveal private information or proprietary techniques.
Explaining your AI system so people can understand it is a fundamental design challenge.
Summary: Types of explanations
Data use explanations | Scope of data use | |
Reach of data use | ||
Examples-based explanations | Generic explanation | |
Specific explanation | ||
Descriptions | Partial explanation | Generic explanation |
Specific explanation | ||
Full explanation | ||
Confidence-based explanations | Categorical | |
N-best results | ||
Numeric | ||
Data visualizations | ||
Explaining through experimentation | ||
No explanation |
Evaluating Explanations
When designing for AI explainability, you need to assess if your explanations increase user trust or make it easier for users to make decisions. When embarking on a new project, you can evaluate your explanations internally within your team and later with users.
Internal Assessment
- 1.
Consider if your type of explanation is suitable for the user and the kind of product.
- 2.
Observe how your team members interact with the explanation. Ask them what they understand from the explanation and what parts are confusing.
- 3.
Determine if the components of your explanation are relevant to the user. Are we highlighting the right parts in the explanation?
- 4.
Determine if your explanation has any implications on user privacy, proprietary information, or the product’s security.
User Validation
You should also validate your explanations with the users of your product. Your user group should reflect the diversity of your audience. You can use qualitative or quantitative methods to validate your explanations.
Qualitative Methods
- 1.
User interviews
You can conduct one-on-one interviews with users and ask them what they think about your explanation. Here, you need to check if their mental model matches your product’s model of how the AI works.
- 2.
Surveys and customer feedback
You can float survey forms to your customers or ask for feedback inside the product while interacting with it. You might sometimes be able to validate your explanations by listening to customer service calls, reviews and feedback of your product on external websites and app stores, and public conversations about your product on social media.
- 3.
Task completion
You can ask your users to complete a predefined task on your product or a prototype and observe if your explanation helps them accomplish the task. You can also have a completion time defined as a metric of success. For example, an explanation is a success if the user is able to scan a document for the first time within one minute.
- 4.
Fly on the wall
You can be a fly on the wall, that is, you can ask to be a silent observer on your user interactions. In many cases, your product might already be tracking user interactions on the product. Make sure to get the appropriate permissions when shadowing users. This method can help you uncover confusing parts of your product, where users hesitate, where you need to explain better, and which parts need further explanations.
Quantitative Methods
You can sometimes validate your explanations using quantitative methods like product logs, funnel diagrams, usage metrics, etc. However, quantitative methods will most likely provide weak signals and ambiguous information about explainability. For example, a search result page that indicates the confidence of results may have many drop-offs. But it is difficult to pinpoint with only quantitative data if the drop-off happened due to poor explainability or something else. Quantitative methods are suitable for finding broad problems in the product. They are a good starting point, but they need to be coupled with qualitative assessments.
Control
Users should have the ability to exercise their own judgments and second-guess your AI’s predictions. They should not blindly trust the AI’s results.
Guidelines for Providing User Control
Providing control assures users that if things go wrong, they can always correct them. Incorporating control mechanisms like the ability to edit data, choose or dismiss results, and provide feedback allows users to assess the results and make their own judgments. Control mechanisms can also help your team improve the AI through feedback. Here are some guidelines that can help you design better control mechanisms.
Balance Control and Automation
- 1.
Stakes of the situation
The level of control you provide will depend on the stakes of the situation. You would need to provide stakeholders with greater control in high-stakes scenarios like surgery, healthcare, finance, or criminal justice. A higher level of control can include greater visibility into the algorithms, the ability to edit the AI algorithm or underlying training data, or correcting AI’s mistakes. On the other hand, users might be okay with lesser control in low-stakes situations like music recommendations, applying filters on photos, etc.
- 2.
Time required for AI to learn
Consider how long it would take for your AI to reach the target level of accuracy and usefulness. In many cases, your AI might learn user preferences from scratch. In the beginning, when it might not be as accurate or helpful, you can provide a greater level of user control by putting higher weightage on control mechanisms in the interface. Over time as the AI improves, you can shift the weightage to predictions. For example, in a movie recommendation service, in the beginning, a user may be asked to choose most of the titles from predefined categories like action, comedy, horror, etc., thereby providing more control over the results. As the system learns the user’s preferences and reaches a level of acceptable accuracy, you can start showing AI recommendations—slowly increasing the weightage from predefined categories to personalized ones.
- 3.
Time required for users to learn
In cases where you’re introducing new mental models, it might take time for users to learn the system’s workings. Users might need time to understand how the AI system generated its results and the relationship between their input and AI’s output. Your users may be wary of trusting the AI’s recommendations from the start; they might tinker and experiment with the product before establishing the right level of trust. Even if your AI is highly accurate, it is a good idea to provide users with a greater level of control while they learn the system’s mental model. For example, workers in a factory using a robot may choose to control it manually in the beginning and slowly hand off various tasks as they gain more trust.
Hand Off Gracefully
When your AI makes mistakes, the easiest way is to return control to the user. It should be easy for users to take over the system. The level of control and override would depend on the type of situation. Sometimes, it is risky to hand off to a user immediately. For example, it can be dangerous if a self-driving car suddenly asks the driver to take over at high speed. You need to ensure that the handoffs are graceful. When this AI-to-manual control transition happens, it’s your responsibility to make it easy and intuitive for users to pick up where the system leaves off quickly. That means the user must have all the information they need in order to take the reins: awareness of the situation, what they need to do next, and how to do it.50
Types of Control Mechanisms
Control mechanisms allow users to take over the system when it makes mistakes and use their own judgment. Allowing users to ignore your AI’s results, edit them, or give feedback makes it more likely for them to trust the system and continue using it. You can build different types of control mechanisms within your product. These may or may not exist together.
- 1.
Data control
This refers to control over input data that you or your users give the AI.
- 2.
Control over AI output
This refers to control over the AI’s results and how they are displayed.
Data Control
Your users should have the ability to view, access, edit, and share their data that the AI system uses privately and securely. The product should empower them through an explicit opt-in system and explainable information of how the AI stores and uses their data. You can allow users to select the data that the system uses and disable parts of the data they don’t want to be used. In some cases, this might result in limited functionality, inaccurate results, or even the AI not working. Communicate these limitations, but don’t force users into providing data if they wish to opt out. Not allowing users control over their data erodes their trust in the system.
Not allowing users control over their data erodes their trust in the system.
Data control is different from digital consent.
Here are some considerations when designing data control mechanisms.
Global Controls
- 1.
Let users enable or disable external inputs. For example, enabling or disabling location input in a shopping application can enable or disable a “trending near you” feature.
- 2.
Allow users to indicate preferences. For example, a running app can allow users to indicate if they want to avoid slopes, rocky trails, or traffic.
- 3.
Continue to clearly communicate permissions and settings throughout the usage of your AI. Ask permissions early in the user journey. Think about when the user might want to review preferences they’ve set in the past and consider reminding them of these settings when they shift into different contexts and may have different needs. They may also forget what they’re sharing and why, so explain the reasons and benefits.54
- 4.
Allow users to deny service or data by having the AI ask for permission before an interaction or providing the option during an interaction. Privacy settings and permissions should be clear, findable, and adjustable.55
- 5.
Make sure that your controls are not too broad. Simply enabling or disabling an AI feature is a very broad control that leads to your users either using the product or not, that is, they need to fully trust your AI or not at all. For example, an AI-based music recommendation service with controls that only allow enabling or disabling the service would not be useful for many users. Users who want to use the service will be forced to accept all data inputs even if they aren’t comfortable sharing parts of their data. Forced data sharing can erode your user’s trust in the AI and easily make them susceptible to switching to a competitor. However, if users can control which parts of their data are shared with the service, they can calibrate their level of control and trust in the system. For example, enabling users to disable location data in the music recommendation service can disable location-based playlists but allow them to use the rest of the application. Keep in mind that there is a risk of being too granular with the level of control you provide. Users can be confused if they have to choose granular controls like the precision and recall of your ML models.
- 6.
Don’t stop users from deleting or modifying information that the system has already collected. For example, a search engine should allow users to delete previous searches, or a running app should allow users to remove past runs. Users should always maintain control over what data is being used and in what context. They can deny access to personal data that they may find compromising or unfit for an AI to know or use.56 In many cases, this might be a legal requirement.
Editability
Removal and Reset
Opting Out
Your system should allow users to opt out of sharing certain aspects of their data and sometimes all of their data. Opting out of sharing their entire data can be equivalent to a reset. Consider allowing users to turn off a feature and respect the user’s decision not to use it. However, keep in mind that they might decide to use it later and make sure switching it on is easy. While users may not be able to use the AI to perform a task, consider providing a manual, nonautomated way to complete it.
Empower users to adapt your AI output to their needs.
Control over AI Output
Your AI won’t be perfect for every user all the time. Empower users to adapt your AI output to their needs, edit it, ignore it, or turn it off. Here are some considerations when enabling users to control the AI’s outputs.
Provide a Choice of Results
Allow Users to Correct Mistakes
Support Efficient Dismissal
Make it easy to dismiss any irrelevant or undesired AI recommendations. Make sure to clarify how to dismiss a result, for example, swipe left on a dating app, saying “Cancel” in the case of a voice assistant, hiding or reporting ads in a search engine, etc. Don’t obfuscate the method of dismissal, for example, a dismiss button for a text autocomplete feature that is too small or not obvious enough.58
Make It Easy to Ignore
Make it easy to ignore irrelevant or undesired AI recommendations. The ability to ignore a recommendation is especially useful in cases where the AI has a low level of confidence. You can do this by providing your recommendations unobtrusively. For example, a shopping application can show recommended products below the fold of the main product page that are easy to ignore.
People trust things that other people have trusted.
Borrowing Trust
People trust things that other people have trusted. You might trust buying a car from Tesla if you know your friends have bought it. Through social proof, you can borrow trust from your customers, peers, service providers, reputable brands, or trustworthy organizations. For example, The Currency Shop, an Australian currency comparison site, strategically displays other well-known and recognizable brands in Australia such as ANZ, OFX, and Commonwealth Bank within their landing page, which helps build trust among their customers who may not be familiar with them.59 Many organizations add seals from security providers like McAfee and Norton or show SSL badges on their website to borrow assurances for the product’s overall security and management of customer information.
When building trust with your user, go beyond the product experience. When in-product information is not sufficient to build trust, you can borrow it. You can also support it with a variety of additional resources, such as marketing campaigns to raise awareness and educational materials and literacy campaigns to develop mental models.60
The process of trust-building starts at the beginning of the user journey and continues throughout your product’s use.
Opportunities for Building Trust
To build trust, you need to explain how your AI works and allow your users to be in control of their relationship with the system. Some products are used every day, so their mental model gets formed by ongoing use. But some products are only meant to be used occasionally. For these products, mental models might erode over time, so it’s helpful to consider ways to reinforce them or to remind users of the basics.61 You can strengthen the user’s mental model by using consistent messaging, tone of voice, or an identity for AI-based features. Over time users can start recognizing what is AI-powered or not and calibrate their expectations accordingly.
Building and calibrating the right level of user trust is a slow and ongoing process. The process of trust-building starts at the beginning of the user journey and continues throughout your product’s use. Every touchpoint where the user interacts with the AI is an opportunity for building trust. While trust-building can start even before the user is onboarded, like product landing pages, marketing, and sales communication, in this section, we will focus mainly on in-product opportunities.
In the previous sections of this chapter, we discussed explainability methods, control mechanisms, and their types. This section will look at how we can apply those in your AI interfaces to build and calibrate user trust.
- 1.
Onboarding
- 2.
User interactions
- 3.
Loading states and updates
- 4.
Settings and preferences
- 5.
Errors
Onboarding
Onboarding is the process of introducing a new user to a product or service. Your onboarding experience begins even before a user purchases or downloads the product. Users can start forming mental models of your product from your website, marketing communications, or even word of mouth. Onboarding should not only introduce the system but also set the users’ expectations of how it will work, what it can do, and how accurate it is.62 Here are some guidelines that can help you design better onboarding experiences.
Set the Right Expectations
- 1.
Be upfront and make clear what your system can and can’t do. You can also collaborate with your marketing team to define this messaging in-product and within marketing collaterals.
- 2.
Make clear how well the system does its job. Help users understand when the AI might make mistakes. For example, a service that recognizes cats from images might not perform too well on jungle cats.
- 3.
Set expectations for AI adaptation. Indicate whether the product learns over time. Clarify that mistakes will happen and that user input will teach the product to perform better.64 Let users know that the AI may need their feedback to improve over time. Communicate the value of providing feedback.
- 4.
Show examples of how it works to clarify the value of your product. Explain the benefit, not the technology.
Introduce Features Only When Needed
Clarify Data Use
Allow Users to Control Preferences
Give users control over their data and preferences as they get started. Give them the ability to specify their preferences and make corrections when the system doesn’t behave as expected, and give them opportunities to provide feedback.67 This can also help set expectations that the AI will adapt to preferences over time.
Design for Experimentation
Reboarding
Consider reboarding, that is, onboarding again, if there are noteworthy improvements in how a feature works or the data it uses. Ensure that these changes are significant enough for users to notice. You can also reboard existing users who are interacting with your system after a long time. In such cases, it is good to ask users if they want to be onboarded again and provide the option of skipping. Reboarding can also be used when a user forgets how to use your service and needs a refresher in the form of tutorials, wikis, etc.
User Interactions
Once your users are onboarded and have started using the system, you need to ensure that their trust is maintained and calibrated regularly. The user’s interactions with your AI should strengthen established mental models. You need to set the right expectations of the AI, explain how the system works, and provide user control over preferences. You can help strengthen the mental models of your AI by maintaining a consistent messaging of its value, confidence, improvements over time, and data use. Here are some guidelines that can help you maintain and calibrate trust across the user journey.
Set the Right Expectations
- 1.
Explain how your system works through partial descriptions.
- 2.
Make sure that you explain what your AI feature can and can’t do. Ensure that your users are aware of the full breadth of functionality. For example, a search functionality within a pharmacy website outlines the types of searches you can make.
- 3.
Make it clear how well your system performs. Make sure your users understand when your AI makes mistakes. For example, a translation service built for American English speakers might not perform perfectly with a British accent.
- 4.
You can use the AI’s confidence levels to indicate performance and accuracy. This can enable users to calibrate the right level of trust.
- 5.
Set expectations for adaptation. Let users know that the system improves over time through user feedback. Encourage users to provide feedback when appropriate.
Clarify Data Use
Make sure your users understand what data is collected, tracked, and monitored by the system and how it is used. You should clarify which parts of their data are used for what purpose, what is private, and what is shared. You can also let your users know if the AI’s results are personalized for them or aggregated over lots of users.
Build Cause-and-Effect Relationships
You can strengthen your AI’s mental models by building cause-and-effect relationships between your user and the AI system. In many cases, the perfect time to show an explanation is in response to a user’s action. People understand better if they can establish a relationship between their response and the AI’s output. A user’s relationship with your product can evolve over time through back-and-forth interactions that reveal the AI’s strengths, weaknesses, and behaviors.
Allow Users to Choose, Dismiss, and Ignore AI Results
Users can build a more trustworthy relationship with your AI system if they can use their own judgment. Whenever possible, your AI system should provide a choice of results. If your AI’s results are not helpful, users should be able to ignore or dismiss them easily.
Loading States and Updates
- 1.
Provide users with information about updates to the system, for example, a “What’s new?” section to inform users about the AI system additions or capability updates.69
- 2.
Inform users about any changes to the privacy policy or legal regulations.
- 3.
Inform users about any changes to how data is used and shared.
- 4.
Don’t obfuscate updates in the algorithm, especially when they lead to direct changes in the AI’s behavior, for example, ranking of results in a stock prediction service.
Settings and Preferences
Your AI product’s settings and preferences section is an excellent destination for providing transparency and explaining how the system uses user data. It is also a great place to provide users control over their data and preferences. You can give users the ability to view, access, edit, and share their data that the AI system uses in a private and secure manner. Here are some guidelines that can help you design better settings interfaces for your AI products.
Provide Global Data Controls
Provide global controls by allowing users to customize any data that is collected, tracked, or monitored by the system. Let users enable or disable any external inputs like location, call information, contacts, etc. Clearly communicate permissions and privacy settings. Make sure they are findable and adjustable.
Clarify Data Use
In your system settings, make sure to communicate what data is collected, tracked, and monitored by the system and how it is used. You should clarify which parts of their data are used for what purpose, what is private, and what is shared.
Allow Editing Preferences
Your users might have provided their preferences during onboarding. Sometimes, your AI can learn user preferences over time. However, user choices can change over time. Allow users to specify or edit their preferences at any point in time. For example, a music recommendation service might adapt to a user’s likings over time, but it might not be as expected. As users’ tastes in music change, they might want to change the preferences that your AI captured previously.
Allow Users to Remove or Reset Data
Whenever possible, allow users to remove their entire data or a part of it. You can also provide users with the ability to reset their information and start from scratch. This is especially useful if the system has learned incorrect preferences leading to irrelevant predictions. In such cases, it is best to reset and start from scratch.
Allow Opting Out
Errors
When your AI makes mistakes, your users might trust it less. However, errors are great opportunities to explain how the system works and why the AI made a mistake, collect feedback, and recalibrate user trust. Here are some suggestions for using errors as trust-building opportunities.
Adjust User Expectations
- 1.
Repair broken trust by making clear why the system made a mistake. Enable users to understand why the system behaved as it did.
- 2.
Allow users to know what data was used to make the incorrect prediction.
- 3.
Set expectations for adaptation. While it might not be accurate right now, tell users that the system is learning and will improve over time.
Hand Off Gracefully
When your AI makes a mistake, the easiest method is to give users a path forward by giving them control. You can allow them to take over the system and manually complete the action. Address the error in the moment and let users know how you will resolve the problem. However, you need to design the AI-to-manual transition carefully to ease your users into taking control of the system.
Allow Users to Correct AI Mistakes
You can enable users to correct the AI’s mistakes through feedback. Providing feedback can also help set expectations for adaptation. Prevent the error from recurring: give users the opportunity to teach the system the prediction that they were expecting, or in the case of high-risk outcomes, completely shift away from automation to manual control.70 Ability to correct mistakes helps users calibrate their trust in your AI system.
Allow Users to Choose, Dismiss, and Ignore AI Results
Opportunities for building trust in your user journey
Opportunities | Considerations |
---|---|
Onboarding | Set the right expectations. |
Introduce features only when needed. | |
Clarify data use. | |
Allow users to control preferences. | |
Design for experimentation. | |
Consider reboarding existing users periodically. | |
User interactions | Set the right expectations. |
Clarify data use. | |
Build cause-and-effect relationships. | |
Allow users to choose, dismiss, and ignore AI results. | |
Loading states and updates | Provide information about updates to the system. |
Inform users about any changes to the privacy policy or legal regulations. | |
Inform users about any changes to how data is used and shared. | |
Settings and preferences | Provide global data controls. |
Clarify data use. | |
Allow editing preferences. | |
Allow users to remove or reset data. | |
Allow opting out. | |
Errors | Adjust user expectations. |
Hand off gracefully. | |
Allow users to correct AI mistakes. | |
Allow users to choose, dismiss, and ignore AI results. |
Personality and Emotion
Our emotions give cues to our mental states.
While an AI that appears human-like might feel more trustworthy, your users might overtrust the system.
We’re forming these tight relationships with our cars, our phones, and our smart-enabled devices.72 Many of these bonds are not intentional. Some argue that we’re building a lot of smartness into our technologies but not a lot of emotional intelligence.73 Affect is a core aspect of intelligence. Our emotions give cues to our mental states. Emotions are one mechanism that humans evolved to accomplish what needs to be done in the time available with the information at hand—to satisfice. Emotions are not an impediment to rationality; arguably, they are integral to rationality in humans.74 We are designing AI systems that simulate emotions in their interactions. According to Rana El Kaliouby, the founder of Affectiva, this kind of interface between humans and machines is going to become ubiquitous, that it will just be ingrained in the future human-machine interfaces, whether it’s our car, our phone, or our smart device at our home or in the office. We will just be coexisting and collaborating with these new devices and new kinds of interfaces.75 The goal of disclosing the agent’s “personality” is to allow a person without any knowledge of AI technology to have a meaningful understanding of the likely behavior of the agent.76
- 1.
Avatars in games, chatbots, and voice assistants.
- 2.
Collaborative settings where humans and machines partner up, collaborate, and help each other. For example, cobots in factories might use emotional cues to motivate or signal errors. An AI assistant that collaborates and works alongside people may need to display empathy.
- 3.
If your AI is involved with caregiving activities like therapy, nursing, etc., it might make sense to display emotional cues.
- 4.
If AI is pervasive in your product or a suite of products and you want to communicate it under an umbrella term. Having a consistent brand, tone of voice, and personality would be important. For example, almost all Google Assistant capabilities have a consistent voice across different touchpoints like Google Lens, smart speakers, Google Assistant within Maps, etc.
- 5.
If building a tight relationship between your AI and the user is a core feature of your product.
Designing a personality for AI is complicated and needs to be done carefully.
Guidelines for Designing an AI Personality
Designing your AI’s personality is an opportunity for building trust. Sometimes it makes sense to imbue your AI features with a personality and simulate emotions. The job of designing a persona for your AI is complicated and needs to be done carefully. Here are some guidelines to help you design better AI personas.
Don’t Pretend to Be Human
Good design does not sacrifice transparency in creating a seamless experience. Imperceptible AI is not ethical AI.81
Clearly Communicate Boundaries
You should clearly communicate your AI’s limits and capabilities. When interacting with an AI with a personality and emotions, people can struggle to build accurate mental models of what’s possible and what’s not. While the idea of a general AI that can answer any questions can be easy to grasp and more inviting, it can set the wrong expectations and lead to mistrust. For example, an “Ask me anything” callout in a healthcare chatbot is misleading since you can’t actually ask it anything—it can’t get you groceries or call your mom. A better callout would be “Ask me about medicines, diseases, or doctors.” When users can’t accurately map the system’s abilities, they may overtrust the system at the wrong times or miss out on the greatest value-add of all: better ways to do a task they take for granted.82
Consider Your User
- 1.
Define your target audience and their preferences. Your user persona should consider their job profiles, backgrounds, characteristics, and goals.
- 2.
Understand your user’s purpose and expectations when interacting with your AI. Consider the reason they use your AI product. For example, an empathetic tone might be necessary if your user uses the AI for customer service, while your AI can take a more authoritative tone for delivering information.
Consider Cultural Norms
When deploying an AI solution with a personality, you should consider the social and cultural values of the community within which it operates. This can affect the type of language your AI uses, whether to include small-talk responses, the amount of personal space, the tone of voice, gestures, non-verbal communications, the amount of eye contact, the speed of speech, and other culture-specific interactions. For instance, although a “thumbs-up” sign is commonly used to indicate approval, in some countries this gesture can be considered an insult.83
Designing Responses
Leveraging human-like characteristics within your AI product can be helpful, especially if product interactions rely on emulating human-to-human behaviors like having conversations, delegation, etc. Here are some considerations when designing responses for your AI persona.
Grammatical Person
The grammatical person is the distinction between first-person (I, me, we, us), second-person (you), and third-person (he, she, they) perspectives. Using the first person is useful in chat and voice interactions. Users can intuitively understand a conversational system since it mimics human interactions. However, using first person can sometimes set wrong expectations of near-perfect natural language understanding, which your AI might not be able to pull off. In many cases, like providing movie recommendations, it is better to use second-person responses like “You may like…” or third-person responses like “People also watched…”
Tone of Voice
What we say is the message. How we say is our voice.84 When you go to the dentist, you expect a different tone than when you see your chartered accountant or your driving instructor. Like a person, your AI’s voice should express personality in a particular way; its tone should adjust based on the context. For example, you would want to express happiness in a different tone than an error. Having the right tone is critical to setting the right expectations and ease of use. It shows users that you understand their expectations and goals when interacting with your AI assistant. An AI assistant focused on the healthcare industry may require some compassion, whereas an AI assistant for an accountant may require a more authoritative/professional tone, and an AI assistant for a real estate agency should have some excitement and enthusiasm.85
Strive for Inclusivity
- 1.
Consider your AI’s gender or whether you should have one. By giving it a name, you are already creating an image of the persona. For example, Google Assistant is a digital helper that seems human without pretending to be one. That’s part of the reason that Google’s version doesn’t have a human-ish name like Siri or Alexa.86 Ascribing your AI a gender can sometimes perpetuate negative stereotypes and introduce bias. For example, an AI with a doctor’s persona with a male name and a nurse’s with a female name can contribute to harmful stereotypes.
- 2.
Consider how you would respond to abusive language. Don’t make a game of abusive language. Don’t ignore bad behavior. For example, if you say “Fuck you” to Apple’s Siri, it denies responding to you by saying “I won’t respond to that” in a firm, assertive tone.
- 3.
When users display inappropriate behavior like asking for a sexual relationship with your AI, respond with a firm no. Don’t shame people, but don’t encourage, allow, or perpetuate bad behavior. You can acknowledge the request and say that you don’t want to go there.
- 4.
While it can be tempting to make your AI’s personality fun and humorous, humor should only be applied selectively and in very small doses.87 Humor is hard. Don’t throw anyone under the bus, and consider if you are marginalizing anyone.
- 5.
You will run into tricky situations when your users will say that they are sad or depressed, need help, or are suicidal. In such cases, your users expect a response. Your AI’s ethics will guide the type of response you design.
Don’t Leave the User Hanging
Ensure that your users have a path forward when interacting with your AI. You should be able to take any conversation to its logical conclusion, even if it means not having the proper response. Never leave users feeling confused about the next steps when they’re given a response.
Risks of Personification
- 1.
We should think twice before allowing AI to take over interpersonal services. You need to ensure that your AI’s behavior doesn’t cross legal or ethical bounds. A human-like AI can appear to act as a trusted friend ready with sage or calming advice but might also be used to manipulate users. Should an AI system be used to nudge a user for the user’s benefit or the organization building it?
- 2.
When affective systems are deployed across cultures, they could adversely affect the cultural, social, or religious values of the community in which they interact.88 Consider the cultural and societal implications of deploying your AI.
- 3.
AI personas can perpetuate or contribute to negative stereotypes and gender or racial inequality, for example, suggesting that an engineer is male and a school teacher is female.
- 4.
AI systems that appear human-like might engage in psychological manipulation of users without their consent. Ensure that users are aware of this and consent to such behavior. Provide them an option to opt out.
- 5.
Privacy is a major concern. For example, ambient recordings from an Amazon Echo were submitted as evidence in an Arkansas murder trial, the first time data recorded by an artificial intelligence–powered gadget was used in a US courtroom.89 Some AI systems are constantly listening and monitoring user input and behavior. Users should be informed of their data being captured explicitly and provided an easy way to opt out of using the system.
- 6.
Anthropomorphized AI systems can have side effects such as interfering with the relationship dynamics between human partners and causing attachments between the user and the AI that are distinct from human partnership.90
A successful team of people is built on trust, so is a team of people and AI.
Summary
- 1.
Your AI system will work alongside people and will make decisions that impact them. People and AI can work alongside each other as partners in an organization. To collaborate efficiently with your AI system, your stakeholders need to have the right level of trust. Building trust is a critical consideration when designing AI products.
- 2.
Users can overtrust the AI when their trust exceeds the system’s capabilities. They can distrust the system if they are not confident of the AI’s performance. You need to calibrate user trust in the AI regularly.
- 3.
Users need to be able to judge how much they should trust your AI’s outputs, when it is appropriate to defer to AI, and when they need to make their own judgments. There are two key parts to building user trust for AI systems, namely, explainability and control.
- 4.
Explainability
Explainability means ensuring that users of your AI system understand how it works and how well it works. This allows product creators to set the right expectations and users to calibrate their trust in the AI’s recommendations. While providing detailed explanations can be extremely complicated, we need to optimize our explanations for user understanding and clarity.
- 5.
Different stakeholders will require different levels of explanation. Affected users and decision-makers often need simpler explanations, while regulators and internal stakeholders might not mind detailed or complex explanations.
- 6.The following are some guidelines to design better AI explanations:
- a.
Make clear what the system can do.
- b.
Make clear how well the system does its job.
- c.
Set expectations for adaptation.
- d.
Plan for calibrating trust.
- e.
Be transparent.
- f.
Optimize for understanding.
- 7.The following are the different types of AI explanations:
- a.
Data use explanations
- b.
Descriptions
- c.
Confidence-based explanations
- d.
Explaining through experimentation
- 8.
Sometimes, it makes sense to provide no explanation when the explanations get in the way of a user’s actions. Your users might already have a mental model of when and how that happens. It would also be wise to avoid explanations if they reveal private information or proprietary techniques.
- 9.
You can consider evaluating your explanations through internal assessment with your team; qualitative methods like user interviews, surveys, etc.; or quantitative methods like usage metrics, product logs, etc. Quantitative methods are a good starting point to find the broad problem, but they need to be coupled with qualitative assessments.
- 10.
Control
Users should be able to second-guess the AI’s predictions. You can do this by allowing users to edit data, choose the types of results, ignore recommendations, and correct mistakes through feedback. Users will trust your AI more if they feel in control of their relationship with it. Giving users some control over the algorithm makes them more likely feel the algorithm is superior and more likely continue to use the AI system in the future.
- 11.The following are some guidelines to design better control mechanisms:
- a.
Balance the level of control and automation by considering the stakes of the situation and the time required for the AI and user to learn. You would need to provide stakeholders with greater control in high-stakes scenarios like surgery, healthcare, finance, or criminal justice.
- b.
Hand off gracefully by returning control to the user when the AI fails or makes a mistake.
- 12.The following are the types of control mechanisms:
- a.
Data control
Your users should have the ability to view, access, edit, and share their data that the AI system uses in a private and secure manner. You should empower them through an explicit opt-in system and explainable information of how the AI stores and uses their data. You can allow users to select the data that the system uses and disable parts of the data they don’t want to be used.
- b.
Control over AI output
Empower users to adapt your AI output to their needs, edit it, ignore it, or turn it off. You can provide a choice of results and allow users to correct the AI’s mistakes through feedback. You can also make it easy for users to ignore the AI’s result or allow them to dismiss AI outputs easily.
- 13.
Apart from building trust through explanations and control mechanisms, you can also borrow trust from your customers, peers, service providers, reputable brands, or trustworthy organizations through social proof.
- 14.The following are some key trust-building opportunities for your AI:
- a.
Onboarding
- b.
User interactions
- c.
Loading states and updates
- d.
Settings and preferences
- e.
Error states
- 15.
You need to be intentional about whether to give your AI a personality. While an AI that appears human-like might feel more trustworthy, your users might overtrust the system or expose sensitive information because they think they are talking to a human. An AI with a personality can also set unrealistic expectations about its capabilities.
- 16.
It is generally not a good idea to imbue your AI with human-like characteristics, especially if it is meant to act as a tool like translating languages, recognizing objects from images, or calculating distances.
- 17.The following are some guidelines for designing an AI personality:
- a.
Don’t pretend to be human.
- b.
Clearly communicate your AI’s limits and capabilities.
- c.
Consider whom you are designing the personality for to help you make decisions about your AI’s brand, tone of voice, and appropriateness within the target user’s context.
- d.
Consider the social and cultural values of the community within which the AI operates.
- e.
Strive for inclusivity.
- 18.
Imbuing your AI with a personality can be dangerous if used to manipulate people, engage in psychological manipulation, or interfere with relationship dynamics between humans. It can sometimes perpetuate or contribute to negative stereotypes and gender or racial inequality. Privacy is also a major concern where your users might overtrust the system or expose sensitive information because they think they are talking to a human.