Chatbots and agents like Muse are getting a face — but does that make us trust them more? Here's what this AI expert had to say

Making people feel understood is the most important thing. People already have very personal conversations with AI tools like ChatGPT and Claude without any human face on the screen. What matters is whether the AI understands what you’re saying and responds in a way that feels relevant to you.

A humanlike persona can help, especially when you’re practicing a conversation with an AI avatar in a platform like Second Nature. But bear in mind that the more human the AI avatar looks, the more you expect it to behave naturally. If its expressions, timing or responses feel off, you’re entering uncanny valley territory, which can be distracting or uncomfortable. The face and video can help draw someone into a conversation, but the quality of the interaction is what keeps them engaged.

The form of the AI avatar should suit its audience and purpose. At Second Nature, a humanlike avatar makes sense because people are practicing sales conversations that they’ll eventually have with real people. For an everyday assistant to check your email or make dinner reservations, a friendly animated character may be enough. A realistic avatar sets a high bar: people expect its expressions, timing, and responses to match its appearance.

Trust goes much deeper than appearance. In order to trust the AI, people want to see that the AI knows them and what they are looking for. It should have deep context so the user does not have to repeat all the information in every interaction. It’s the difference between speaking with a friend you’ve known for many years and a stranger. This familiarity wins trust.

We’ve seen people laugh at an AI avatar’s jokes and respond to it in ways you’d recognize from a human conversation. We’ve tested this extensively, and when the AI avatar is more humanlike, users interact with it more openly, with more empathy, using phrases like “I hear you.” They also build more rapport with questions like “How are you today?” and “Nice to meet you,” and they say thank you more often.

People’s willingness to share also depends on the quality of the AI conversation itself: whether the AI asks relevant questions, responds thoughtfully, and makes them comfortable admitting a mistake or asking for help. A humanlike face can help set the scene for practicing a conversation, but feeling comfortable opening up rests on much more than the avatar itself.

The most unexpected thing is how quickly talking to an AI avatar can feel ordinary. You know you’re speaking to software, but once you get into the conversation, you can become less conscious of the technology.

Everyone using the Second Nature platform knows they're talking to an AI. Still, in more than four out of five practice conversations, the person thanks the AI avatar. A third of them apologize to it. Nearly half say goodbye or wish it a good day before they hang up. And almost nobody steps outside the scene: across tens of thousands of conversations, the number of times someone asked "are you a bot?" is practically zero.

We also see some people become perfectionists when they know they have unlimited tries. About a quarter of Second Nature users who score 90 or above go straight back and practice the role play again.

Our social instincts switch on whether or not we believe the other side is human. That's why this kind of training works: you rehearse the real emotional muscle. It's also why the AI industry owes people transparency. An AI should always be clearly labelled, and it should be built for the user's benefit and not to exploit that instinct. Being comfortable with it should depend on what the AI is for, not on how human it feels.

Yes, especially in situations where people feel vulnerable. It can be difficult to tell your manager you’re struggling with something or to ask the same question for the third time. An AI coach gives people a place to ask for help, practice, and get reassurance without worrying about how their manager will judge them.

One sales manager turned to our AI coach on the last day of the quarter, anxiously trying to get a customer’s signature. The AI gave him sound, practical advice about what to do next, which helped calm him down. Sales can be stressful, and sometimes people need reassurance as much as they need advice.

Consistency also matters. A manager reviewing 25 practice conversations may be tired by the last one, while an AI can apply the same criteria throughout. That can build confidence in its feedback. But being more comfortable with an AI, or finding its feedback consistent, doesn’t mean it’s always right.

When the AI gets an assessment wrong, people need a clear way to challenge it. At Second Nature, we have a “human in the loop” workflow that allows users to contest AI feedback and a human to review and override it. That also gives us an opportunity to understand what went wrong and improve the system. Trust grows when people know an assessment can be questioned and corrected.

All of the above. People will see AI differently depending on what they need from it. The same person might use AI as a tool to complete a task, a coworker to work through an idea, a coach to practice a difficult conversation, or a companion to talk to.

The biggest change I think people are underestimating is how much decision-making they’ll hand over to AI. Today, you might ask AI to compare products. Increasingly, you’ll ask it to choose and buy one. If you need toilet paper, you may care about the price and when it arrives, but be fine with letting your AI decide between Whole Foods and Walmart.

The same thing will happen in business. An AI agent that needs a service to send email or text messages can compare providers, check how easily their software connects, choose and purchase the service. That’s a buying decision a person used to make.

Businesses need to prepare for AI becoming part, or even most, of their audience: their products will need to be easy for agents to find, try, and use. For consumers, the question becomes which decisions you’re comfortable delegating and where you want to stay in control. Trusting AI enough to have a conversation with it is one thing; trusting it to make decisions and spend your money means you really need to trust the system and the company behind it.

If you stop practicing a skill because AI does it for you, of course there’s a risk of getting rusty. But I wouldn’t jump from that to calling it cognitive decline, and I don’t have any alarming examples to share that would justify that claim.

What I’m seeing is a shift in the skills people use. Calculators changed how we do math, and AI is changing how we approach problems. There’s more value in asking good questions, evaluating the answers, synthesizing information, and deciding what to do with it. Those activities still require critical thinking.

The most important skill may be knowing what you’re trying to achieve. AI can suggest a route and carry out tasks, but you need to decide where you want to go and judge whether it’s helping you get there. I’m more optimistic about AI’s ability to develop people’s skills when people use it that way.

With our customers, AI gives sales and customer support teams more opportunities to practice conversations. When you rehearse a sales conversation with an AI customer, you still have to listen, think on your feet, handle objections, and communicate your response. The AI creates the opportunity; you do the practicing. And you get measurably better at your job as a result.