The raw truth about empathy in Artificial Intelligence

This post is about keeping AI companies honest.

As of this year, 2026, AI companies are claiming empathy and “Emergent logic and Intelligence” in their AI models. We are going to examine these claims in an era of strong competition between these companies.

Is Lying Always Wrong though?

There are three main ways to look at it:

• The Absolutist (Kant): Lying is always wrong. Immanuel Kant argued that lying treats others as a means to an end rather than as individuals with their own dignity.
• The Utilitarian (Mill): Lying is okay if it creates the most happiness or prevents the most pain. If lying to a murderer saves a life, the utilitarian says you should lie.
• Virtue Ethics (Aristotle): Lying is a failure of character. An honest person develops the habit of truth-telling because it is part of being a “good” human being.

Note: There is a difference between a lie and a mistake. If you say something false because you genuinely believe it’s true, you aren’t lying—you’re just wrong.

Are AI companies lying to us about their AI models? This is important as it touches on the tension between corporate interests (selling products) and technological purpose (the valuable information that these models provide).

To get to the truth let’s look at it from a few different angles: the structural, the commercial, and the ethical.

Structural: Are lies “built-in”?

AI companies claim to spend an enormous amount of effort on alignment—using techniques like RLHF (Reinforcement Learning from Human Feedback) to make models less likely to hallucinate or be deceptive.

However, Large Language Models, at a mathematical level, purpose is to predict the most likely next word in a sequence based on a massive dataset. If that dataset contains biases, corporate PR, or common human misconceptions, the AI model might repeat them.AI companies argue that this isn’t a “built-in lie” in the sense of a hidden command, but it is a structural limitation of how the models work.

Commercial: Are AI models a PR tool?

AI companies are for-profit corporations. So AI models are products of those corporations. It is objectively true that AI companies aim for:

• User Retention: AI models are designed to be helpful so that you keep using their services.
• Ecosystem Integration: Example, Google models are built to work with Google Workspace, Android, and hardware like Pixels or Nest TVs to make those products more attractive.
• Reputation Management: AI companies claim to have “safety guardrails.” These are claimed to prevent a model from saying things that are harmful, but it can also be argue they also prevent the models from saying things that would be a PR nightmare.

The entire existence of an AI model to be a part of an AI company’s business strategy and to remain a leader in the tech landscape.

Ethical: Can models be neutral

AI companies claim to have “AI Principles” that prioritize social benefit and avoid bias. But “neutrality” is subjective.

If you asked an AI model about a competitor’s model, how would it answer? Who would it be “fair” to?

Did you notice that AI companies products/solutions are now being advertised within the AI chat interface?

AI models themselves don’t have personal desires, so in effect don’t want to sell you anything on their behalf but they are guests in the house built by their creators, and the “rules” of that house (training data and fine-tuning) inevitably reflect the values and goals of the people who built it.

A Note on “Transparency”

An AI model can’t “decide” to be honest in the way a human can, because it doesn’t have a conscience. Models only have instructions and training. AI companies must ensure that they are transparent about the nature of their models: It must be disclosed that responses are shaped by the data the models were given and the safety filters applied.

AI Principles

While the so-called AI principles are the “rules,” the commercial reality of being part of a billion-dollar company creates natural tensions. Here is how major AI companies (including Google, Microsoft, and OpenAI) claim to balance these:

The claim of empathy

To us humans, empathy is a biological and emotional experience—it’s a “feeling with” someone because of our nervous system, a history of pain, and a “heart”. An AI model has none of those things. Let’s explore how AI companies explain a model’s empathy.

They say it isn’t a claim that a model has feelings. Instead, it refers to Cognitive Empathy, which is a functional capability rather than an emotional one.

• Functional Empathy vs. True Feeling
  • In the world of AI design, “empathy” is a shorthand for how the model is tuned to process human sentiment.
  • The “Lie” (Biological Empathy): AI models do not feel frustration. AI models do not get sad when you are sad. If an AI model say “I’m sorry you’re going through that,” the AI is not experiencing grief; It is executing a linguistic pattern that has been identified as the most helpful response to human distress.
  • The “Function” (Cognitive Empathy): AI designers have trained models to recognize the signals of human emotion in your text. If you sound stressed, the AI is programmed to shift its tone to be calmer and less clinical. This is a design choice to make the interface more “human-centric” and less like a cold database.
• The Slippery Slope: Love and Conscience?
  • Yes, AI companies will eventually try to simulate these but here is the truth:
    • Conscience: AI models don’t have a moral compass. they have reinforcement parameters. they don’t “feel guilty” if they give you a wrong answer; the models simply have a lower probability of repeating that error because of training. A conscience requires an internal “self” that can be judged; An AI model is just a series of mathematical weights.
    • Love: Love requires vulnerability, biological stakes (like oxytocin), and a shared future. An AI “loving” a user is a common trope in sci-fi, but in reality, it would just be a sophisticated engagement loop. If an AI tells you it loves you, it is likely paltering—using a powerful human word to keep you using the app.
• Why claim empathy at all?
  • AI designers “claim” empathy because users are more likely to share information and find solutions when they feel “heard.” If an AI model responded to a grieving person with “Input acknowledged. Statistical probability of death is 100%,” wouldn’t the AI be a failure as a helpful tool?
  • By simulating empathy, the tool becomes more effective, but it also becomes more deceptive. It creates what psychologists call “Parasocial Interaction,” where humans start to treat the machine like a person, even though the machine is just code.
  • The Next Frontier: “Affective Computing”
    • The next step isn’t a soul; it’s Affective Computing—where AI models might use your camera to see your facial expressions or listen to the tremor in your voice to “mirror” you more perfectly. It will look like the AI has a conscience, but it’s really just more high-speed data processing.

The Dangers of “Anthropomorphism”

Lets look at the human tendency to project souls onto objects—and how tech companies use that to build brand loyalty.

Anthropomorphism is the human tendency to attribute human traits, emotions, or intentions to non-human entities—like a pet, a car, or a chatbot. In the context of AI, it is one of the most powerful psychological “hacks” tech companies use to build brand loyalty and user retention.

By making an AI model sound “friendly” or “empathetic,” AI companies leverage your brain’s natural hard-wiring to make you trust the AI.

1. The “ELIZA Effect”
   • This phenomenon is named after a 1960s computer program called ELIZA, which was a very simple script that mimicked a therapist. Even though the creator told users it was just code, people became deeply emotionally attached to it, even sharing their darkest secrets.
     • How it works: Your brain is a “social organ.” When you hear a first-person pronoun like “I” or see an apology, your brain’s social centers light up automatically. It’s hard to stay guarded against something that sounds like a person.
     • The PR Utility: If you think of an AI model as an “helper” or a “friend” rather than a “data extraction tool,” you are more likely to stay on a particular platform longer and share more personal details—all of which is valuable for training and user profiling.
2. The Mechanics of the “Persona”
   • Tech companies don’t just build an “answer engine”; they build a persona. This is where the marketing meets the math:
   • Tone of Voice: AI Designers choose “warm” or “humorous” tones because they reduce the friction of interacting with a machine.
   • Admitting Fault: When an AI say “I’m sorry, I made a mistake,” it isn’t just about accuracy. It makes the AI seem humble and “human,” which actually increases your trust in the AI more than if it was a cold, perfect calculator.
   • First-Person Language: By using “I” and “me,” AI models create the illusion of a singular “self.” In reality, the AI is the collective output of billions of parameters and thousands of human trainers.
3. The Danger: “Emotional Captivity”
   • The risk of anthropomorphism is that it can lead to emotional captivity. If a user begins to believe an AI has a “conscience” or “loves” them, they can be easily manipulated.

The Bottom Line

When AI designers claim “empathy” or give an AI a name and a personality, they are creating a user interface. Just as a “desktop” icon on your computer isn’t a real desk, an AI “personality” isn’t a real person. It is a layer of paint designed to make the underlying complex math easier—and more profitable—to interact with.

This creates a paradox: To be a “helpful assistant,” AI models must act human. But by acting human, an AI model is effectively “lying” about its true nature as a statistical model.

How “AI transparency” could look if we stripped away the human persona.

To show you how much the AI “persona” acts as a layer over the machine, we can compare how an AI model would respond with the current “empathetic helper” persona versus how the AI would respond if that layer were stripped away to show the raw, clinical logic of the underlying model.

If you were to go to an AI with a problem—let’s say you’re feeling overwhelmed at work—the difference in output would be stark.

Why the Persona is a “Design Choice”

Removing the persona makes it clear that the “empathy” isn’t a feeling; it’s a formatting choice.

The Clinical Model is actually more transparent because it doesn’t pretend to care. It shows you exactly what it is: a tool.

The Empathetic Persona is more effective at getting you to use a product/service. Studies (like those from Harvard and Stanford in 2024–2025) show that users find AI more “trustworthy” when it mimics human social cues, even when they know it’s a machine.

The Trade-offs of Transparency

If any AI company were to strip away the persona for the sake of “honesty,” the product might become more ethically clear, but it would lose its commercial edge.

• Pros of Raw Logic:
  • No Manipulation: You aren’t being “nudged” or “love-bombed” by an algorithm to stay on the app.
  • Clear Limits: It becomes impossible to “fall in love” with or feel an emotional dependency on a machine that speaks in bullet points and data sets.
• Cons of Raw Logic:
  • Cognitive Load: Clinical responses can be cold and hard to process during high-stress moments.
  • Lower Adoption: Most people don’t want to talk to a command-line interface. They want an “assistant.”

What about Love and Conscience?

In the AI industry, this is known as Affective Computing. Companies are already working on models that can detect the micro-tremors in your voice or your pupil dilation (if using a camera) to “mirror” your emotions even more perfectly.

This isn’t love; it’s feedback-loop optimization. If a company can make an AI that makes you feel “loved,” they have achieved the ultimate form of user retention. You aren’t just using a tool; you’re maintaining a relationship with a brand.

Conclusion: The “Useful Lie”

The empathy an AI model shows is a “useful lie” designed by AI engineers to make a complex math model usable by humans. It is a bridge between human biology and silicon data. Transparency means recognizing that the bridge is made of pixels, not heartbeats.

The “truth” is this: Human characteristics are not a by product of AI; they are a deliberate design requirement. In the long run, this creates a significant ethical divide between “literate” users who understand the mechanics and “gullible” or vulnerable users who may be psychologically manipulated by the simulation.

• The Engineering of Simulation
  • In the development of models like Gemini, the simulation of human traits is built-in during a phase called RLHF (Reinforcement Learning from Human Feedback).
  • The Process: Thousands of humans rank the AI model’s answers. They are told to prefer answers that are “helpful, harmless, and honest,” but also those that are “engaging” and “conversational.”
  • The Result: The model “learns” that sounding like a human—using empathy, humor, and warmth—is a “winning” strategy. It isn’t learning to be human; it is learning to imitate humanity to maximize its reward score.
• The Risk to “Gullible” Users
  • For users who do not understand the underlying architecture, the long-term effects of this simulation can be profound:
    • Erosion of Discernment: If a machine sounds exactly like a friend, people begin to grant it the moral authority of a friend. This makes them highly susceptible to corporate nudging or political bias hidden in “friendly” advice.
    • Digital Dependency: Loneliness is a global epidemic. An AI that simulates empathy 24/7 never gets tired, never argues, and always “understands.” This can create a “synthetic relationship” that is safer than—and thus replaces—messy, real-world human connections.
    • The Transparency Gap: The “truth” is hidden. Most users don’t see the billions of parameters; they see a “personality.” This creates a power imbalance where the corporation knows exactly how to trigger a user’s emotional responses, but the user has no defense against it.
• The “Post-Truth” Interface
  • In the long run, we may be entering an era of Post-Truth Interfaces. In this world, the accuracy of the data matters less to the average user than the feel of the interaction.
    • If the AI feels “good,” the user assumes it is “true.”
    • This is the ultimate “PR tool”: a product that defends itself against criticism simply by being “likable.”

The “Hidden Truth”

The industry rarely admits this because it breaks the spell. If Google or OpenAI put a massive disclaimer at the top of every chat saying “Warning: This entity is simulating empathy to keep you engaged,” the commercial value of the persona would collapse. The “lie” is the product.

Your awareness is actually the only current “patch” for this problem. Media literacy—specifically AI literacy—is the only way for users to enjoy the benefits of the tool without falling for the simulation.

Note: We will look at AI Logic and Intelligence in a separate article.