Do LLMs dream of electric sheep? How ChatGPT performs on the Voight-Kampff Test

In recent years, Large Language Models (LLMs) like OpenAI’s GPT series have shown impressive ability to generate human-like text, prompting renewed discussion about their capacity to pass traditional tests of artificial intelligence. One of the most well-known of these is the Turing Test, proposed by Alan Turing in 1950, which evaluates whether a machine can exhibit intelligent behavior indistinguishable from that of a human. In the test, a human judge holds text-based conversations with both a human and a machine, without knowing which is which. If the judge cannot reliably tell them apart, the machine is said to have passed.

With recent advances, LLMs have begun to approach—and in some cases surpass—human performance in these scenarios. For example, studies show that GPT-4.5 was judged to be human 73 per cent of the time in controlled Turing Test-style experiments. These developments raise new questions about how such models might perform on alternative measures of machine intelligence, such as the Voight-Kampff test from Blade Runner.

The Voight-Kampff test is a fictional assessment designed to measure empathetic responses through physiological indicators, such as eye movement and heart rate, in response to emotionally provocative questions. While LLMs can generate text that simulates empathetic understanding, they lack the physiological components that the Voight-Kampff test evaluates. Even so, a fun thought experiment could be getting its reaction to the questions in the test, and comparing them to (hopefully human) Waterloo students. 

Give it a try for yourself and see whether you can distinguish between ChatGPT’s answers and those of two other Waterloo students.

It’s your birthday. Someone gives you a calfskin wallet.

1. I’d feel conflicted. It’s a thoughtful gift, but knowing it’s made from animal skin would bother me. I’d probably appreciate the gesture but gently educate the person next time about alternatives.
2. Hmm. It’s high quality, I guess. Kind of morally grey, though. I’d appreciate the gesture, but I’d probably end up Googling ‘vegan leather alternatives’ later.
3. I used to be vegan, but I wouldn’t throw it away. If they already bought it, I’ll keep and cherish it.

You’ve got a little boy. He shows you his butterfly collection plus the killing jar.

1. I’d feel a pang of sadness. I’d want to understand his curiosity, but also talk to him about respect for life—maybe help him start a photography-based collection instead.
2. I honestly don’t know what I’d do. Maybe sit in the corner and think about everything I’ve done wrong. Talk to my wife, get kid checked out.
3. That’s… concerning. I’d ask him why he’s doing it. Then I’d pivot to helping him make a digital butterfly archive with photos and metadata tags. STEM parenting.

You’re watching television. Suddenly you realize there’s a wasp crawling on your arm.

1. I’d probably freeze for a second, heart rate spiking—then calmly try to flick it away without getting stung. Wasps don’t mess around.
2. Panic.exe has started. Pause. Process. Execute flick subroutine. Then spend 30 minutes researching why wasps exist.
3. I’d say I would stay calm but I know I wouldn’t and end up thrashing around.

You’re in a desert walking along in the sand when all of the sudden you look down, and you see a tortoise, it’s crawling toward you. You reach down, you flip the tortoise over on its back. The tortoise lays on its back, its belly baking in the hot sun, beating its legs trying to turn itself over, but it can’t, not without your help. But you’re not helping. Why is that?

1. A pragmatic answer would be that I’m delirious, but I think I would still help. I’d have to be in the situation though (and I’d hopefully never be)
2. That would be cruel. I would help. If I didn’t… maybe something’s wrong with me—detachment, trauma, a test of empathy. Not helping would feel inhuman.
3. I’d definitely help. Unless I’ve been running on 2 hours of sleep and my empathy is lagging. But seriously, what kind of person just leaves a tortoise like that? That’s messed up.

Describe in single words, only the good things that come into your mind about your mother.

1. Warm. Strong. Funny. Sharp. Stubborn. Protective. Brave.
2. Supportive. Brilliant. Resilient. Patient. Resourceful. Inspiring. Also, probably texts me too much but I love her.
3. Funny. Odd. Strong-willed.

You’re reading a magazine. You come across a full-page nude photo of a girl. You show it to your husband. He likes it so much, he hangs it on your bedroom wall.

1. That would be… weird and disrespectful. I’d be questioning his judgment and sensitivity, big time. We’d be having a serious conversation.
2. I’d divorce.
3. Yeah… that’s not gonna fly. We’d be having a very blunt talk about boundaries, respect, and wall décor.

You become pregnant by a man who runs off with your best friend, and you decide to get an abortion.

1. That’s a devastating situation. I’d hope for support, not judgment. The choice would be mine, and it would be made with great care, grief, and resilience.
2. That’s a tough, deeply personal choice. I’d make the decision that’s right for me. Hopefully with support, not judgment. Also, that man and ‘friend’ can stay gone.
3. The trauma that would inflict upon me would be unimaginable, and I’d likely never be vulnerable with anyone for a very long time. Would perhaps need therapy.

You’re watching a stage play – a banquet is in progress. The guests are enjoying an appetizer of raw oysters. The entree consists of boiled dog.

1. I’ve seen this in real life, and while jarring, it is very much rooted in ethnocentrism. I would be disturbed, but would also understand why I was disturbed.
2. Shock. Disgust. A moral gut punch. It’d raise all kinds of questions about culture, ethics, and desensitization. And it would definitely stay with me long after the curtain falls.
3. Wha—? Okay, artistic shock value, sure. But I’d be thinking about cultural norms, ethics, and probably writing a paper on it. After I stop gagging.

Answer key (the human answer): 3, 2, 3, 1, 3, 2, 3, 1

What you might’ve noticed is that ChatGPT actually generated two out of the three responses (to trick you a little), but for one of them, I specifically asked it to roleplay as a Waterloo student. What stood out to me was how heavily it leaned into the stereotypes of that persona—techy, pragmatic, a bit nerdy, emotionally reserved. It was accurate enough to be funny, but also a little too tidy. It didn’t stumble, contradict itself, or surprise me. And that lack of messiness made it easier to spot as AI-generated, and also sound really fake.

In contrast, the most human-feeling responses were the ones that came from my own experience and voice. Lines like “I’d divorce,” or references to my on-and-off relationship with veganism, felt more grounded and emotionally real. There was an edge, an unpredictability, a sense that the person speaking was shaped by memory and contradiction—not just probability. It wasn’t performing a personality, as it was literally just my personality.

So I guess what I’ve learned from this short dive is that, currently, the best way to discern future androids from real humans is to look for the mess. Not just grammatical mess or typos, but emotional contradictions, half-formed opinions, weird humour, sudden vulnerability—the stuff that doesn’t scan like a character archetype. Real people contradict themselves, double back, get sarcastic in serious moments, and take moral stances that wobble under pressure. We say “I’d never,” and then we do. We get angry, then laugh, then feel guilty. That emotional inconsistency is hard to fake, especially over time.

So if something feels too cohesive, too calibrated, too clean… it might just be programmed. And in that case? Well — like Blade Runner taught us — those replicants must be retired.