Why We Should Stop Making AI Look Human

Making AI visually indistinguishable from humans might seem compelling, but it’s a misguided goal that introduces friction, unmet expectations, and poor design.

As I write this (mid-2025), AI has rapidly transitioned from behind-the-scenes systems to daily interactive and conversational engagements. Text and chat remain dominant modes — straightforward, effective, and essential for everything from coding to email management.

Audio interfaces are also evolving: so far, it has been used for very transactional commands (Alexa, Siri, etc.), but we’re starting to see more nuanced interactions with voice assistants that can handle complex AI conversations (ChatGPT voice chat).

And now, we’re on the verge of another shift: AI interfaces that mimic human appearance and behavior.

At first glance, human-like interfaces feel natural and advanced. But do they improve how we interact with AI, or do they create confusion, distraction, and a false sense of capability that the tech can’t deliver?

Visual Anthropomorphization: An Unnecessary Frontier

It’s easy to imagine that the next wave of AI interfaces will include avatars that will be visually indistinguishable from humans.

Imagine a conversational AI represented by a lifelike avatar, complete with realistic expressions, subtle eye movements, and natural gestures. It could smile warmly to express agreement, furrow its brows slightly in confusion, or nod encouragingly during conversation. The avatar’s eyes might track your movement or gaze direction, reacting with appropriate emotional responses, such as empathy, surprise, or attentiveness. Essentially, interacting with it would visually feel like speaking directly with another human being.

Movies and futuristic narratives often present AI this way. The allure is understandable: familiar faces and humanized interactions feel natural, relatable, and advanced. But do we need AI to replicate human interaction, or are we chasing a mirage of optimal communication?

Designing AI to be indistinguishable from humans is a path fraught with distractions, unmet expectations, and misplaced effort. We should aim for interaction methods uniquely suited to AI and digital mediums.

Interface Evolution Proves Simpler is Better

The evolution of interface design has already taught us valuable lessons. I consider the example of the transition from skeuomorphic to flat design on computers and phones, which highlighted that focusing only on essential visual signals greatly improves usability.

Skeuomorphic design was the norm in early computing and was very prominent in the early days of mobile apps. Think of the first versions of iOS. It included plenty of shadows, textures, gradients, lighting effects, 3D buttons, and other visual elements that mimicked real-world objects. The idea was to make digital interfaces feel familiar and intuitive by replicating physical objects like books, buttons, and switches.

Flat design removed unnecessary detail, reducing cognitive load and confusion. Similarly, multi-touch interfaces on the iPhone succeeded by inventing gestures explicitly optimized for fingers, and not by mimicking older desktop paradigms. In a way, it proposed the removal of realism to create a more efficient human-computer interaction.

These lessons strongly suggest AI interfaces should evolve by identifying their strengths and limitations, rather than trying to replicate human interaction.

Designing AI interfaces to look and act human introduces a core mismatch between appearance and capability. Realistic avatars raise user expectations of empathy, context awareness, or human-like reasoning that current AI can’t meet. This creates friction: small lapses become jarring, interactions get bogged down by unnecessary social cues, and trust erodes in the uncanny space between “almost human” and “clearly not.” Instead of enhancing the experience, realism often distracts from it.

Consider “Emma”, which the City of Amarillo added to its website to help residents find information. As a test, I once asked her, “What are trash pickup days?”. The answer was vague and unhelpful: “It happens during weekdays but can change depending on holidays. For more information, you can call the waste management company.” This completely missed the mark. A lifelike AI face creates the expectation that it has immediate access to local knowledge, or at the very least, the awareness to ask for my address or neighborhood and retrieve a specific answer. Instead, the interaction felt shallow and frustrating. The realism set a bar that the system couldn’t meet, and the result was a worse experience than a simple, well-designed FAQ page.

Visual Cues, Not Visual Realism

While full realism has clear drawbacks, visual communication itself remains vital. Simple visual signals can convey AI’s status (listening, processing, confusion, agreement) and significantly streamline interactions. In some cases, expressive characters that can convey emotions or reactions can enhance engagement without mimicking human complexity.

Here are some examples, which can be thought of as a spectrum — from purely functional cues to more expressive visual personas — that enhance interaction without crossing into human mimicry:

• Functional status indicators: Loading spinners, color-coded feedback, subtle animations. These are purely functional visuals, clearly indicating system states without attempting to convey any personality.
• Animated icons with persona: A good example is Siri’s colorful waveform. Simple visuals are designed to subtly convey a sense of presence or persona, making interactions feel slightly more personal, yet non-human.
• Expressive non-human characters: Think Wall-E’s eye expressions, R2-D2’s beeps and head tilts, or Duolingo’s playful owl. These characters succeed because they signal intent, emotion, and presence clearly, without ever pretending to be human. They avoid the uncanny valley entirely while still engaging users in emotionally resonant ways. This design choice builds trust and personality without overpromising intelligence or empathy.

If realism isn’t the right path forward, what should guide effective AI interface design? Here are four core design principles that put usability and trust first:

1. Authenticity

The interface should never pretend the AI is more capable than it is. Avoid visual cues that imply human-level perception, empathy, or intelligence unless those capabilities exist.

Example: An avatar that maintains eye contact should be backed by actual camera input and gaze tracking, or it risks creating a false sense of awareness and trust.

2. Efficiency

Convey status or emotion using the simplest visual or interaction method available. Don’t replicate human dialogue if a click or tap is faster.

Example: Filling out a shipping address is easier through a form than by speaking it line by line.

3. Contextual Appropriateness

Match the interface design to the task and setting. Use minimal, efficient visuals for transactional tasks, and richer, more expressive ones for emotional or immersive experiences. Speed isn’t always the priority: some interactions benefit more from tone, presence, or emotional resonance.

Example: A children’s storytime app may benefit from a warm, animated character. But a smart home AI is better served by simple visual confirmations or audio cues that don’t distract from the task.

4. Accessibility and User Control

Allow diverse users to interact comfortably and maintain control over their experience and preferred method of interaction modes or interface complexity.

Example: Some people prefer typing over speaking; others love using voice input. A good AI interface doesn’t force one method — it gives users control over how they interact, whether through text, voice, or visual prompts.

The best AI interfaces won’t fool us into thinking they’re human. They’ll help us become more human by clearly communicating through minimal yet expressive signals, complementing our strengths rather than mimicking them.