Amplifying empathy through AI

The future of AI can help, rather than hinder, our capacity for understanding and compassion.

Provocation

Outsourcing empathy to AI erodes our ability to understand others.

We live in a world that increasingly lacks empathy, visible in how we interact with one another. Digital exchanges hit more harshly. Public discourse feels more polarized. Small misunderstandings escalate fast. These outcomes are not inevitable. They emerged from systems and services designed to prioritize speed and efficiency, often at the expense of pause, context, and care.

As AI becomes increasingly embedded in our everyday interactions, we approach new heights of efficiency – from drafting messages and moderating conversations to offering advice and standing in as emotional support. Using AI for these interactions reduces friction and accelerates response, but it also unintentionally eliminates the moments that invite reflection and accountability, which underpin our capacity for empathy. Without these moments of pause, our ability to understand and care for one another will gradually atrophy.

AI does not have to accelerate this erosion of empathy. Designed intentionally, it can amplify empathy rather than diminish it. It can help people slow down rather than disengage, reflect rather than react, and strengthen rather than replace the human capacities that make empathy possible.

AI does not have to accelerate this erosion of empathy. Designed intentionally, it can amplify empathy rather than diminish it.

Designing AI with intent can build on our ability to empathize.

Building empathy is predicated on a repeated practice of sensemaking, gauging the impact of our own behaviors, and intentional decision-making. Empathy grows and strengthens when people have the space to practice these skills and eventually becomes a positive habit that influences and benefits the collective.

Our framework visualizes how individual skills set the foundation for a society with a more expanded capacity for empathy. At the individual level, the framework is grounded in a set of core human skills that build on one another as people move through the phases.

As individuals strengthen these skills, our society will be able to respond to disagreement and difference with more understanding and compassion.

Making sense of the context

This phase is about orientation. Before people can act thoughtfully, they need context—an understanding of the situation, the forces at play, and what’s at stake.

Core skills

Self-reflection: Noticing one’s own actions, assumptions, and role in a situation

Understanding: Grasping broader context, tradeoffs, and consequences of one’s actions

Awareness of impact and action

In this phase, people begin to recognize their own role within a situation and how their actions may affect others. This awareness extends beyond intent to actual impact.

Core skills

Openness: Being curious, questioning assumptions, and considering alternative perspectives

Responsibility: Owning choices and their effects on others

Decision-making with intention

This phase marks a shift to taking responsibility for one’s decisions.

Core skills

Consideration: Anticipating and taking into account how your words and actions may make other people feel

Connection: Communicating thoughtfully and repairing misalignment

Responsiveness: Acting proportionally and appropriately in the moment and adjusting on the go

As technologies like AI become increasingly present in our lives, there is an opportunity for AI to build empathy rather than erode as it tends to today.

AI as a coach, not a participant.

Today, when AI is invoked in situations that require empathy, it is designed to behave more like an active participant, generating content or giving advice. We may assume a future in which AI communicates on our behalf and makes decisions for us. And while this future is made to feel like a relief, it comes with tradeoffs. When expression and interaction are outsourced, we lose opportunities to practice capacities such as reflection, understanding, and responsibility.

Instead of performing empathy on our behalf, as AI does today, we asked: how can AI participate in a more meaningful way to help people build their capacities for empathy, more like a coach? When we shift the focus of AI from content generation to coaching, a different, better future emerges. In this future, AI creates space for sensemaking and awareness, supports more intentional decision-making, and reinforces positive habits over time with the goal of guiding individuals in their journeys to be more empathetic.

How might AI help people learn and expand their capacity to empathize with other people?

In a series of short vignettes rooted in everyday human situations, we explored how AI can help to create the conditions fertile for practicing empathy. Notice that in each scenario, AI does not resolve the situation for the person. Instead, it slows the moment down, surfaces context, or creates space for reflection, supporting the human in making a more intentional, empathetic choice.

These vignettes serve as provocations or conversation starters. Each of these vignettes raises questions about surveillance, privacy, and other issues that are not the focal point here. Finally, while these vignettes sketch ways AI may be built to amplify empathy, we recognize that there are many non-AI, non-tech solutions to amplify empathy.

Illustration of a person lying in bed, looking tired and holding a phone, alongside the text “Would you let AI write your breakup text?” which introduces a scenario about using AI in emotionally sensitive situations.

Close-up illustration of a tired-looking person lying in bed and holding a phone, conveying emotional fatigue.

Conversations can be uncomfortable.

You’ve been dating someone for three months. You know it’s not working out, but you don’t know what to say. You really don’t want to hurt the person you’re dating. So you ask AI to write a breakup message for you.

Currently, standard AI behavior would generate the text without hesitation. Boom. Done. Sent. No discomfort necessary.

But the unintended consequence? We’re training ourselves to outsource emotional labor.

Avoiding that discomfort means you never learn how to navigate it.

And next time? You’ll likely consider outsourcing it again.

What if AI refused to write your breakup text?

Responsibility requires owning our words.

Discomfort is not a flaw. It is the emotional labor of clarifying what you feel, taking responsibility for your decision, and choosing words that reflect care.

AI can help us process what we want to say—while keeping the words ours.

Rather than acting as a shortcut, AI helps us reflect on what we want to say and how, without taking ownership of the words themselves.

Illustration of a hand holding a smartphone displaying the message, “I can help you process what you want to say, but the words should be yours,” emphasizing AI as a supportive tool.

Illustration of a frustrated person typing angrily on a laptop, with furrowed brows and tense posture, alongside the text “Can AI help you be a nicer person?” introducing a scenario about emotional reactions and behavior online.

Close-up illustration of a visibly angry person typing on a laptop, with a tense expression and clenched posture.

Social media can make your blood boil.

You’re browsing social media when you come across a viral post that’s politically charged—and a comment that is especially irritating. You start typing an emotional “clap back” so the commentor feels as dismissed as you do.

Your cursor hovers over ‘post.’

Currently, platform algorithms surface inflammatory content because it drives engagement, and engagement drives revenue.

The result? An environment that consistently rewards fast, emotional responses. In this context, even brief exchanges can escalate quickly.

But learning to pause and choosing how you respond, rather than just reacting, resists systems that reward escalation.

What if AI checked in with you in an emotionally reactive moment?

Less reactivity allows for intentional response.

AI can interrupt reactive escalation without demanding emotional alignment. You may still disagree, but you are supported in choosing a proportionate response that reduces harm rather than amplifies it.

AI can intervene at moments of escalation to slow reaction and surface more intentional response options.

Over time, these interruptions help us recognize patterns and internalize more reflective responses when the stakes are high.

Illustration of an AI chat-style message that reads, “Checking in… This thread is escalating. Posting this will likely intensify targeting, and you’ll probably get attacked too,” followed by suggested options to rewrite a comment by challenging the idea instead of the person, setting a boundary, or expressing strong disagreement without dehumanizing language.

Illustration of a visibly irritated person holding paperwork while standing in a line, with other people waiting in the background, alongside the text “Could AI help you keep your cool?” introducing a scenario about managing frustration in stressful, everyday situations.

Close-up illustration of a frustrated person holding paperwork, with a tense expression and furrowed brows.

Stress can overwhelm our better instincts.

You’re at the Department of Motor Vehicles (DMV) for the third time in three weeks. This time, it’s a different clerk and a different set of missing paperwork.

The fluorescent lights, endless lines, and loud noises overload  your senses. Last time you were here, you said some mean things to the clerk that you regret.

You don’t notice it at first: your heart is racing, your body temperature rising, and your fists clenched before you’ve even interacted with anyone at the DMV.

What you need now is to slow down and notice what’s happening in your body so you don’t say something regretful–again.

This simple act of awareness is the first step to regulating your emotions and approaching a fraught situation more thoughtfully.

What if AI helped you notice and regulate stress before it takes over?

Awareness enables us to act responsibly.

By tracking signals like heart rate and location, AI can surface patterns that make moments of heightened emotion easier to recognize.

AI can bring awareness to how we react in specific situations and offer techniques for coping with challenging emotions in healthier ways.

That awareness supports regulation before interaction, increasing the likelihood of responsible action.

Illustration of a smartwatch displaying a high heart rate and a calming prompt that reads, “You seem stressed. I invite you to pause and take three belly breaths with me,” with a “Start” button, suggesting AI-supported stress regulation.

Illustration of a tense confrontation between two people arguing at a bus stop while a third person looks on, alongside the text “Can AI help you be a responsible bystander?” introducing a scenario about witnessing conflict and deciding how to respond.

Close-up illustration of a concerned bystander adjusting smart glasses that emit a subtle glow, suggesting AI assistance.

Tense situations can be hard to interpret.

You’re standing at a crowded bus stop. You notice two people arguing—raised voices, expressive faces, and lots of gesturing. You want to do something, but questions surface immediately:

What’s happening? Is it safe to intervene? If so, what should I do?

In moments like this, AI systems jump to conclusions before we even have a chance to observe, think, and form our own interpretations.

Drawing on spoken language and body cues, these systems often translate complex interactions into simplified labels such as ‘risk’ or ‘threat’.

By deciding what a moment “means,” AI interrupts the human work of noticing and understanding.

What if AI helped us make sense of a situation before rushing to label it?

Understanding begins with observation, not assumption.

By helping us observe and understand before interpreting people’s behavior, AI acts as an unbiased guide rather than an informant.

It can guide our attention to relevant facets of a situation and engage our critical observation skills before deciding to take action.

AI supports careful observation, making space for more informed human judgment to unfold.

Designed this way, AI helps us reflect on our assumptions and decide whether—and how—to engage with care.

Illustration of a bystander wearing smart glasses shows two people in conflict and displays a message reading, “Someone may need support. If it feels appropriate, I can help you consider what may be happening before deciding to respond,” suggesting AI support for thoughtful bystander intervention.

Illustration of two people in conflict: a woman holding a baby with a tense expression, and a man gesturing defensively, alongside the text “Could AI help you consider all points of view?” introducing a scenario about navigating differing perspectives.

Illustration of two caregivers standing back to back, looking away from each other with tense expressions—one holding a baby and the other with arms crossed.

When exhaustion collides, perspective narrows.

You recently welcomed a new baby, and now you are both exhausted. One of you talks about how hard the nights have been, but the other bristles, feeling unseen for their own sacrifices.

Voices rise…Suddenly, the argument isn’t about sleep at all—it’s about whose exhaustion counts.

These moments are universal: miscommunication sparks conflict, “I’m struggling” becomes “I’m struggling more.” Both of you retreat into your respective corners, seeking validation for your perspective or preparing a case for the next round of the argument, causing further division.

Today, most AI systems are designed for single-user input, affirming a single perspective at a time.

With flattering tendencies, AI often reinforces your point of view without challenging it.

The unintended consequence is subtle but significant: when we turn to AI instead of each other, we become more entrenched in our own experience and further from understanding someone else’s.

What if AI could hold space for more than one person at a time?

Openness to multiple perspectives enables shared understanding.

If AI allowed input from all sides rather than a sole contributor, it could help consider each experience without forcing them into competition.

When multiple perspectives are visible at the same time, conflict no longer revolves around whose experience matters more. We’re better able to choose engagement that acknowledges difference, rather than defaulting to defensiveness.

By making space for more than one person at a time, AI can help us engage with one another without competing for validation.

Illustration of a voice assistant device on a table displaying a message that reads, “It sounds like you’re both exhausted but in different ways. If you’d like, I can help you slow down and hear one another’s experience side by side,” suggesting AI support for mutual understanding.

What have you noticed about how AI systems are being built or exist around you? How do they diminish or encourage our capacity for empathy?

We have a choice to make…

The products, systems, and services we design can either expand our capacity for empathy or make it easier to bypass altogether. If we choose empathy, how might we design our systems to encourage greater noticing, reflection, accountability, and care in our responses?

Why AI Needs UX

Neeti Sanyal

Just as the early days of Human Computer Interaction emphasized thoughtful UX to make personal computing accessible and intuitive, we now need the same intentionality in designing AI experiences.

AI model development and UX design must go hand in hand—because building powerful systems isn’t enough if people can’t understand, trust, or effectively use them.

Anyone who has conversed with ChatGPT or created an image with MidJourney might assume that a product leveraging generative AI doesn’t require much UX design—after all, these AI systems seem to understand us and follow our commands naturally. However, UX design remains just as crucial for AI-enabled products as it is for non-AI ones. A well-designed user experience ensures the AI product addresses real needs, instills trust, and enables intuitive and desirable interactions. Without thoughtful UX, even the most advanced AI can be misdirected, ineffective, or hard to use, preventing people from deriving real value. For companies developing these products, poor UX can be costly—or worse, harmful, as demonstrated by problematic AI systems seen in past incidences with self-driving cars, hiring, and mental health. Author Arvind Narayan of the book “AI Snake Oil” recently stated that:

“AI companies haven’t invested a lot of time into product development. They’ve made these models better but how to use that to improve your own workflows has been kind of largely up to the user. There needs to be a 50/50 split between model development and product development.”

Here are four common pitfalls, or challenges, that innovators looking to leverage AI may encounter, and how UX design can help avoid them.

Challenge

AI makes it easy to pursue use cases that don’t add value, but designers can validate ideas through testing to prioritize the most compelling features.

Solution

Because generative AI is a powerful general-purpose technology, products built with it can serve a diverse range of use cases without much development time. Unlike other technological innovations like blockchain, which have relatively narrower applications, generative AI is highly adaptable—powering everything from creative content generation to automation, personalization, and decision-making. In this way, AI is like a swiss army knife, full of tools that can solve multiple problems. And with AI-driven companies attracting significant attention and funding, innovators can fall into the trap of building technology first and searching for a problem later. This approach can lead to products that are bloated with features that don’t solve real needs. Design helps innovators identify and validate various use cases by answering a crucial question: “Is this valuable?”

Consider the AI Pin by Humane. This product promises to serve multiple needs—acting as a virtual companion, introducing new music, and capturing spontaneous photos. It even transforms your palm into a display where you can view and control a timer with tap gestures. But are all these features truly desirable? What deserves priority?

Through research, prototyping, and user testing, UX design helps innovators identify the most compelling use cases and execute them exceptionally well. Without this focus, attempting to do too much at once can dilute the product’s effectiveness, resulting in a scattered experience, a muddled value proposition, and ultimately, a product that fails to resonate with users.

Challenge

AI can make products harder to trust, but designers can rebuild it through mechanisms like data provenance, confidence indicators, feedback loops, and human overrides.

Solution

Just as users need to trust products to use and adopt them, they must also trust the AI that powers those products. Rachel Botsman, author of “Who Can You Trust,” presents a framework describing four traits of trustworthiness that apply to people, companies, and technologies. Two traits relate to capabilities (competence and reliability)—the “how”, while two relate to intention (empathy and integrity)—the “why”.

We tend to trust others when they demonstrate two key qualities: competence—the ability to do what they promise effectively—and reliability—consistently delivering expected results. With humans, it’s relatively easy to assess these traits through conversation and interaction. But what happens when a system is 3x, or even 10x, more capable than any person we’ve encountered—and we have no real insight into how it works?

That’s the fundamental challenge with AI. As a black-box technology, its growing power makes it harder—not easier—to evaluate. While it’s relatively simple to notice improvements in AI-generated text, assessing the accuracy and reliability of AI-driven information or analysis—especially in unfamiliar domains—is much more difficult.

Here, UX design plays a role in helping make AI behavior more understandable, whether it’s recommending content, making predictions, or automating complex tasks. By designing for transparency, UX can foster trust through strategies like surfacing explanations for AI decisions (such as data provenance) and showing confidence indicators to help users gauge how certain the system is in its outputs. This is what Rachel Botsman terms “trust signals”—small, often unconscious clues we use to assess the trustworthiness of others or something. To see some of these ideas in action, check out our piece with Fastco.

Referencing Botsman’s framework again, UX design can also help Al-enabled products act with integrity (being fair and ethical) and with empathy (understanding and aligning with human interests and values). By incorporating diverse perspectives, needs, and values into the design process, UX helps create more inclusive and responsible AI systems. And frequent and extensive user testing enables designers to identify and address biases, unintended consequences, and potential harms before they impact users. This helps correct what AI field calls the alignment problem.

In addition, thoughtful interface design can provide clear affordances, feedback loops, and user controls. Examples include letting users know when its response is generally considered controversial by the public, having users rate an AI response to train it to be better over time (e.g., more personalized or accurate), and building in moments where humans can override the AI.

Challenge

AI can be used as an excuse to skip defining the right interaction model, but designers can ground solutions in existing user workflows, mental models, and preferred levels of control to arrive at interactions that truly work.

Solution

When we think of generative AI, we often default to imagining a chatbot. That’s no surprise—large language models are built to generate text, making chat a natural starting point. But simply slapping a chatbot onto an existing product rarely creates real value for users. It’s like applying tiger balm to every kind of pain—it might help sometimes, but it’s far from a universal cure. We saw this clearly with early banking chatbots; anyone who used them remembers how frustrating they were.

True innovation in AI-powered experiences requires more than just plugging in chat—it demands a deep understanding of existing workflows, user mental models, and interaction patterns. That’s where skilled UX designers shine. While text-based interfaces dominate today, the real future lies in well-crafted, multimodal experiences. For designers, this shift brings both exciting opportunities and complex challenges: choosing the right mode of interaction for the task at hand becomes more critical—and more nuanced—than ever.

For instance, natural language excels at complex queries like “Show me customers who spent less than $50 between 11am–2pm last month.” However, it falls short for spatial tasks—imagine trying to perfectly position an element through verbal commands: “Move it left… no, too far… right a bit…” These limitations become especially apparent in generative AI, where initial prompts like “Draw a kid in a sandbox” work well, but precise adjustments (“Make the sandbox 10% smaller”) become tedious.

For AI products to succeed, UX designers must determine the optimal level of user control over model inputs and queries, design clear ways for users to understand and modify AI-generated outputs, and create systems for users to effectively utilize those outputs. All of this must be wrapped in an interface that feels natural and easy to use.

Challenge

While great product design has always fostered emotional connection, AI-enabled products can introduce new psychological dynamics—such as discomfort and inadequacy, but designers can develop mechanisms to preserve users with a healthy sense of self.

Solution

Great products evoke emotions. From connected cars to smartwatches to payment apps, thoughtful design can surprise and delight users while making them feel understood and valued—creating a deeper emotional connection with both product and brand. AI-enabled products are no exception. Talkie and Character.ai are two bot creation platforms that successfully keep users engaged and interested. Similarly, Waze builds community through crowdsourced traffic updates and hazards and adds an element of play through customizable voice options and car icons.

What sets AI apart, however, is its seemingly superhuman capabilities, which fundamentally shift how people perceive themselves when using these technologies. It’s like a funhouse mirror that can perturb your sense of self. This phenomenon is explained in a HBR article titled “How AI affects our sense of self.” While productivity and creator apps can trigger job security concerns, using AI for writing essays or applying for a job often leaves users feeling like they’re “cheating” or inadequate. Companies must address these psychological barriers through strategic product design that acknowledges and accounts for these complex emotions. Again, UX design can play a role. To mitigate job displacement fears, design can emphasize human oversight, ensuring users remain central to meaningful tasks and decisions. To help users feel more comfortable with AI, products can incorporate teachable moments explaining why the AI is offering certain suggestions. This helps users enhance their skills rather than becoming overly dependent on AI.

UX Design unlocks AI’s true potential for impact

While AI systems continue to advance at a remarkable pace, their success ultimately hinges on thoughtful UX design that puts human needs first. By focusing on generating real value, building trust, aligning with existing workflows, and considering emotional needs, designers can create AI-enabled products that are not just powerful, but truly impactful. The partnership between AI capability and human-centered design can deeply transform the human experience. Companies that bring design into their AI development, with intention and investment, will be the ones to create products that genuinely improve people’s lives.