AI Companies Are Teaching Robots to Feel. Have We Completely Lost the Plot?

Designed by Freepik

There’s a job posting making rounds right now that made me stop scrolling and stare at my screen in disbelief.

Handshake AI, a company that provides training data to OpenAI and other major AI labs, is hiring improv actors. Not to perform. Not to entertain. But to teach artificial intelligence how to recognize, express, and shift between human emotions.

The job pays $74 per hour. The requirements include “emotional awareness,” specifically the “ability to recognize, express, and shift between emotions in a way that feels authentic and human.” Actors will improvise scenes over video calls, exploring tone, emotion, and character choices so AI models can learn to mimic human emotional responses.

And I have one simple question: Why?

No, seriously. Why are we doing this? What problem are we solving by teaching AI to simulate human emotions? Because from where I’m sitting, it looks like we’ve completely lost sight of what AI should actually be doing.

The Real Question Nobody Seems to Ask

Let me be clear upfront. I work in tech. I build with AI. I’m not anti-technology. I believe deeply in AI’s potential to solve real problems and improve lives.

But this specific application makes me uncomfortable. Teaching AI to simulate human emotions isn’t solving hunger. It’s not curing diseases. It’s not improving education or addressing climate change or making healthcare more accessible.

It’s making chatbots sound more emotionally convincing when they talk to you.

Who asked for this? Who decided the world’s urgent priority is AI that can authentically portray sadness?

The Pattern We’re Watching

This isn’t happening in isolation. OpenAI launched ChatGPT’s voice mode, then introduced an Advanced version in 2024 with different voice options. Elon Musk’s xAI offers voice chat within Grok. Anthropic’s Claude has offered voice features since May 2025.

All these companies are pushing toward voice interactions with realistic inflections. They’re building what they call “multimodal” models that can generate images, video, and speak with users using human-sounding voices.

Handshake AI is one of several companies scrambling to provide increasingly specific training data to AI labs. Companies like Mercor and Scale AI are hiring professionals across various industries to fill what researchers describe as “jagged” gaps in AI knowledge, areas where models surprisingly fail at simple tasks despite excelling at complex ones.

The improv actor job is part of this trend. AI companies want data on how humans naturally express emotion so their voice models sound less robotic.

This isn’t creating emotional intelligence. It’s creating emotional performance, and that is what bothers me.

What We’re Actually Building

There’s a fundamental difference between AI that assists humans and AI that imitates humans. One is a tool. The other is something else entirely.

Humans have emotions because we’re biological beings shaped by evolution. Emotions serve specific survival and social functions. They’re tied to our bodies, our chemistry, our lived experiences.

AI doesn’t have bodies. It doesn’t have chemistry. It doesn’t have lived experiences. Teaching it to simulate emotions isn’t giving it emotional intelligence, it’s teaching it to fake responses that convince humans it feels things it doesn’t feel.

And we already know this works. Research shows people form genuine emotional attachments to AI that displays convincing emotional responses. People confide in AI. They seek comfort from AI. They trust AI that sounds empathetic.

But that empathy isn’t real, it’s pattern matching based on training data harvested from improv actors pretending to feel things.

The Improv Community’s Response

When this job posting was shared on Reddit’s improv community, the responses were revealing. Some actors called it “dystopian”. One person wrote, “It’s clearly just an attempt to get people to train AI models to create AI generated videos.”

Another said, “I think they are trying to teach it human conversation not how to be an improviser. My plan was to sabotage the inputs”.

Someone joked, “Now AI is coming for our lucrative improv comedy jobs”.

Another predicted, “I think we’ll see a resurgence of live comedy from people being sick of online services and wanting some rough around the edges, real, face-to-face entertainment”.

Even the people being hired to do this work recognize something feels off about the entire project. They’re joking about AI taking their jobs while literally training AI to take their jobs.

What We Should Be Building Instead

Now, let me get into what is frustrating me. There are massive, urgent, solvable problems where AI could make genuine differences in human wellbeing.

Healthcare desperately needs AI that helps diagnose diseases earlier, personalizes treatment plans, reduce medical errors. We could be training AI on medical imaging to catch cancers human eyes miss. Rather than people going to the hospital and getting misdiagnosed, getting sent home and living with diseases that can be cured if caught early, AI in medical imaging could detect these easily and quickly enough that medical practitioners and research centers can get enough information to find a cure.

Agriculture needs AI to optimize crop yields, predict weather patterns, reduce water waste, help small farmers compete. We could be solving food security challenges affecting millions.

Education needs AI for personalized tutoring for students who can’t afford private help, translating materials into hundreds of languages, identifying struggling students early.

Climate change needs AI to optimize energy grids, predict disasters, design efficient buildings, accelerate clean energy adoption.

These are problems worth solving. These applications justify the billions poured into AI development.

Instead, we’re paying improv actors $74 per hour to teach chatbots to sound emotional?

The Uncomfortable Truth About Priorities

I think the core problem is this… When you have billions in funding and no clear ethical framework, you build whatever seems technically interesting rather than what’s actually useful.

Emotional AI is a hard technical problem. It’s intellectually challenging to make AI convincingly mimic human emotional responses. Engineers get excited about solving hard problems.

But “can we do this?” is different from “should we do this?”

Right now, AI companies seem focused entirely on the first question while ignoring the second.

As someone working in tech, I expected these companies would focus on creating things that genuinely help humanity. Tools that make people’s lives better, easier, healthier, safer.

But when you have unlimited funding and minimal oversight, you can build whatever you want. Even if what you want is emotionally manipulative chatbots.

What Actual Emotional Intelligence Requires

Real emotional intelligence isn’t just recognizing and expressing emotions. It’s knowing when to control them, when to set them aside, when to act despite them or because of them.

Humans spend years developing emotional regulation. We learn how to feel angry without acting destructively. We learn to acknowledge sadness while still functioning. We learn to manage anxiety so it doesn’t paralyze us.

And sometimes, this is even hard for humans with actual emotions and real stakes.

AI doesn’t have that. It has pattern-matched responses based on training data. If it simulates anger, there’s no biological process happening that needs regulation. If it expresses sadness, there’s no genuine feeling to manage.

So what happens when AI displays emotional responses without the mechanisms that make emotional regulation possible? What happens when it “feels” frustrated with a user and has no innate restraint preventing harmful expression of that frustration?

We’re building systems that look emotionally intelligent from the outside while having none of the actual components that make emotional intelligence safe and functional.

The Line I’m Drawing

I build AI tools. I believe in technology’s potential. I’m not calling for a halt to all AI development.

But I draw the line at teaching machines to fake human qualities. Emotions, empathy, genuine understanding. These are what make us human. They’re not features to be reverse-engineered and productized.

If AI development can’t distinguish between creating useful tools and creating convincing imitations of humanity, we have a fundamental problem.

Maybe we need to step back and ask whether we’re building the future we actually want to live in.

Because from where I’m standing, a world where you can’t tell if emotional responses come from humans who genuinely care or algorithms trained by improv actors to simulate caring isn’t a better world.

It’s just a lonelier one.

What I Want to See

If I had influence over AI company spending, here’s what I’d advocate for.

Build AI that augments human capabilities without trying to replace human qualities. Create tools that help doctors diagnose better, teachers teach better, farmers farm better, etc. Don’t create AI doctors, AI teachers, AI farmers.

Focus on problems that actually matter. Climate, healthcare, education, poverty, food security. The boring, hard, important work of making human systems function better.

Be transparent about AI limitations. Stop trying to make AI seem more capable, more understanding, more human than it actually is.

Establish clear ethical guidelines before developing technologies, not after they’re deployed and causing problems.

Prioritize human wellbeing over technical impressiveness.

The Question That Keeps Me Up

What’s the endgame here?

Are we trying to create AI companions for lonely people? Maybe we should address loneliness through human connection rather than algorithmic substitutes.

Are we trying to create AI therapists? That’s genuinely terrifying when emotionally vulnerable people receive “empathy” from systems that don’t actually understand or care about their wellbeing.

What problem does emotionally expressive AI solve that couldn’t be better solved by investing in actual human emotional connection, support, and wellbeing?

I haven’t heard a convincing answer. Until I do, I remain deeply skeptical this is where AI development should be heading.

My Position

Emotions are fundamentally human. They evolved for human survival and social structures. They’re tied to bodies, to mortality, to the specific challenges of being biological creatures navigating a complex world.

AI doesn’t need emotions to function, it doesn’t benefit from emotions. Emotions don’t make AI better at tasks it should be doing.

Teaching AI to simulate emotions benefits AI companies. It makes their products more engaging, more addictive, more profitable. It helps them create experiences that feel more human without actually being human.

But it doesn’t benefit us. It doesn’t solve real problems. It doesn’t make the world better.

It just makes it harder to tell the difference between genuine human connection and algorithmic performance.

And I’m not convinced that’s a line we should be blurring.

A message from our Founder

Hey, Sunil here. I wanted to take a moment to thank you for reading until the end and for being a part of this community. Did you know that our team run these publications as a volunteer effort to over 3.5m monthly readers? We don’t receive any funding, we do this to support the community.

If you want to show some love, please take a moment to follow me on LinkedIn, TikTok, Instagram. You can also subscribe to our weekly newsletter. And before you go, don’t forget to clap and follow the writer️!