Can AI Replace Humans? Exploring AI's True Capability Boundaries Through an Absurd Viral Video

The Real Question Behind an Absurd Video

A video recently went viral on Bilibili (China's major video platform) titled "Tell me, how is AI supposed to replace humans?" On the surface, it appears to be a random collage of illogical, jumpy snippets of everyday life — tying hair, smearing cake, late-night crying, an old man cooking braised intestines… At first glance, it seems completely chaotic, even absurd. But it's precisely this "chaos" that reflects a proposition that's been endlessly debated yet frequently misunderstood: Can AI truly replace humans?

The video uses what amounts to performance art to showcase the trivial, random, emotionally charged, and context-dependent moments of everyday human life. And these are precisely the things that current AI systems struggle most to truly understand and replicate.

The "Unpredictability" of Human Daily Life: A Chasm AI Can't Cross

The conversations in the video are wildly erratic: one moment a couple is making up after a fight, the next a parent is having a late-night meltdown with a baby, and then suddenly a street vendor is stir-frying in a wok. These scenes follow no unified script — they're full of improvisation, regional slang, emotional swings, and subtle social dynamics.

AI's Enormous Challenge in Contextual Understanding

When someone in the video asks, "Do you think I look like a green tea bitch?" — the sentence carries a complex web of social context, the probing dynamics of an intimate relationship, and anxieties about self-perception. AI can recognize the literal meaning of "green tea" as internet slang, but truly understanding the subtext of this sentence within a specific relationship and emotional state remains an enormous challenge.

The core mechanism of current large language models (such as GPT-4, Claude, etc.) is autoregressive prediction based on the Transformer architecture — in simple terms, predicting the next most likely word based on statistical probabilities from preceding text. This means they are fundamentally performing extremely sophisticated "pattern matching" rather than genuine "understanding." In the field of Natural Language Processing (NLP), this is known as the challenge of "Pragmatics": the meaning of language isn't determined solely by its literal content but is also highly dependent on the speaker's intent, the listener's expectations, the physical and social context of the conversation, and the shared knowledge between both parties. Linguist H.P. Grice's theory of "conversational implicature" points out that a vast amount of information in human dialogue is conveyed implicitly through "violations of the cooperative principle" — such as irony, innuendo, and polite platitudes. While current AI models deliver impressive results on benchmark tests, they still frequently stumble when faced with real-life, highly context-dependent, non-literal expressions.

Human communication relies heavily on shared cultural backgrounds, non-verbal signals, and contextual memory. Those seemingly meaningless conversations in the video are actually built upon long-established rapport between the speakers — and this is precisely the kind of "tacit knowledge" that current large language models struggle to truly master.

The "Electronic Pet" Metaphor: Can AI Understand Humor?

One amusing segment in the video features a guy who discovers his younger brother lying on the floor and jokingly calls him an "electronic pet that's gone offline." He then proceeds to pile things on top of him — a comforter, a suitcase, a bicycle, a stool, a fire extinguisher… This absurd escalation is brimming with a uniquely human sense of humor and mischief.

Humor and Creativity Remain a High Bar for AI

The reason this segment is funny lies in the expectation of "surprise" and the exaggeration of "overreaction." AI can generate jokes, but truly understanding why something is funny, nailing just the right degree of absurdity, and spontaneously creating this effect in real-time interaction remains extremely difficult.

There is an academic field dedicated to studying this problem called "Computational Humor." One of its core theories is the "Incongruity-Resolution Theory": humor arises when an expectation is first established, then broken in an unexpected way, and finally "resolved" by the audience within a new frame of reference. While AI can mimic the form of humor by learning from vast joke datasets, it lacks the ability to gauge "social timing" and "emotional atmosphere" — the same sentence lands completely differently at a funeral versus at a party. Furthermore, research from the MIT Media Lab has shown that human improvisational humor largely depends on "Embodied Cognition" — our bodily experiences and physical-world perceptions directly participate in cognitive processes — which is precisely what pure text-based models lack.

Humor is fundamentally a clever subversion of established expectations, requiring precise awareness of social norms, the other person's psychology, and the rhythm of timing. This ability is rooted in rich life experience and emotional depth, not merely statistical data.

The Weight of Real Life: The Existential Meaning AI Cannot Bear

In the latter half of the video, there's a remarkably realistic segment: a young person born in 2004 puts 150,000 yuan down on a house, takes out loans for renovations, a friend buys a car, prepares for marriage, his wife gets pregnant — all while shouldering a lifetime of debt and sending 3,000 yuan home to his parents every month. Narrated with a self-deprecating tone, it captures the real-life pressures facing today's young people.

The "Burden of Living" That AI Cannot Replace

This touches on the most overlooked dimension in the "AI replacing humans" debate: Humans aren't just tools for completing tasks — they are subjects who bear the consequences of life.

This relates to a classic philosophical question: whether AI possesses "Agency" and "Phenomenal Consciousness." Philosopher John Searle's 1980 "Chinese Room" thought experiment remains a central reference point for this discussion — a person who doesn't understand Chinese sits in a room and processes Chinese symbols according to a rule book. To the outside world, they appear to "understand" Chinese, but they are actually just performing symbol manipulation. The current state of AI is remarkably similar. Furthermore, existentialist philosopher Heidegger's concept of "Being-in-the-world" emphasizes that human meaning isn't externally assigned but is generated through concrete engagement with the world — through shouldering our finitude and responsibilities.

AI can help us write code, generate copy, and analyze data, but it will never truly experience the anxiety of buying a house, the responsibility of parenthood, or the pressure of paying off debt. It won't lose sleep over mortgage payments or feel anxious when a child runs a fever. These aren't questions of "capability" — they're questions of "existence." A great deal of human value lies in the fact that we truly live, bear burdens, and feel — and this very act of "enduring" is at the core of what it means to be human.

Where Exactly Are AI's Capability Boundaries?

From a technical perspective, current AI (especially large language models) excels at structured, quantifiable tasks with clear rules — coding, translation, summarization, image generation, and so on. But in the following areas, AI still falls significantly short of humans:

• Context-dependent commonsense reasoning: Understanding the subtext in complex social scenarios
• Genuine emotional experience: Not simulating emotions, but actually having them
• Spontaneous creativity and humor: Grasping subtle nuances in real-time interaction
• Bearing the consequences of life: The responsibility and meaning that come with being a conscious subject

The industry commonly divides AI into two tiers: "Narrow AI" and "Artificial General Intelligence" (AGI). Narrow AI can surpass humans in specific tasks — for example, AlphaGo defeating the world Go champion, or GitHub Copilot assisting programmers with code — but it cannot transfer capabilities from one domain to a completely different one. AGI aims to build systems with human-like general cognitive abilities that can flexibly adapt to any new environment. Currently, there is enormous disagreement in academia about when AGI will be achieved — or even whether it's possible at all. Optimists like DeepMind co-founder Demis Hassabis believe breakthroughs could come within a decade, while cognitive scientist Gary Marcus at NYU argues that the current path based on Scaling Laws may have fundamental ceilings, because "bigger models" don't equate to "deeper understanding." The core of this debate is precisely the question this article addresses: there is a fundamental chasm between task execution ability and genuine cognitive understanding.

This seemingly nonsensical video uses the most down-to-earth approach to remind us: AI replaces "tasks," not "people." The trivial, chaotic, emotionally rich, and meaning-laden moments of daily life are precisely what constitute the core of being human.

Instead of Worrying About AI Replacement, Think About Human-AI Collaboration

"Tell me, how is AI supposed to replace humans?" The question itself carries a hint of irony. When we see how rich, random, and warm human life is, perhaps the answer is already clear: AI is powerful, but what it excels at and what humans cherish are fundamentally different things.

Human-AI Collaboration has become the dominant paradigm for real-world AI deployment. Stanford University's "Human-Centered AI" (HAI) initiative emphasizes that AI systems should be designed to augment human capabilities rather than simply replace humans. In practice, this philosophy has already produced many success stories: in healthcare, AI-assisted diagnostic systems (like Google's MedPaLM) don't replace doctors but help them screen imaging and organize medical records more quickly, freeing up more time for patient communication. In creative industries, generative AI tools like Midjourney and Suno are used by designers and musicians as "inspiration catalysts," while the final aesthetic judgments and emotional expression remain in human hands. This collaborative model of "AI handles repetitive labor, humans make value judgments" is redefining the future of work.

The real question worth pondering isn't whether AI will replace humans, but rather how we can collaborate with AI — letting technology handle the repetitive and tedious parts so humans have more time to tend to the things that are truly human — tying hair, celebrating birthdays, cooking a plate of braised intestines, and bearing the weight of life itself.