AI毁灭人类的末日预警
主持人: Recently, a researcher at Company A announced his resignation, stating online that he left because, if development continues as it is now, there is a 10% chance AI could kill all of humanity within 10 years. "You believe that AI could kill us all by the end of the decade?" "Yeah." "So the precise scenario sounds a little bit like science fiction." "It sounds like something real, but I think it's frighteningly real."
This estimate was also confirmed by AI godfather Geoffrey Hinton, sparking a wave of panic online. "A 10% chance seems not an unreasonable estimate... nobody really knows how to give a sensible estimate, right? But you just said 10% doesn't seem an unreasonable estimate... that could kill all humans."
嘉宾: Yes.
主持人: Oh my god.
嘉宾: Yes.
人类最后的不可替代工作
主持人: In today’s video, we want to take a step back and see how the latest social science research understands the impact brought by AI. The book The Last Work of Humans won the 2025 American Sociological Association's Distinguished Scholarly Book Award. It tells us: after AI automation, what will be the last work left for humans?
Once various industries are automated by AI, what will be the remaining work for humans? The author believes there is an activity that cannot be replaced by AI, called connective labor — labor that builds emotional connections with others and provides emotional value. Any job that involves interaction with others involves connective labor, such as service work, psychotherapists, coaches, managers, nurses, teachers, journalists, and field workers. These jobs all have one thing in common: you must put in the effort to make another self-aware person feel seen and validated. This is fundamentally something AI cannot replace.
A 2015 study in Finland asked two groups of people to tell each other stories while measuring their skin conductance. The study found that if the listener deliberately showed nodding or approving reactions, the storyteller would have more stable emotions. This means that when we share personal experiences, having an empathetic, attentive listener who nods and affirms us makes us feel cared for and validated. In contrast, if the other person shows no reaction, ignores us, or is visibly distracted, we instantly feel uncomfortable.
This dynamic exists everywhere in our lives. When we talk with friends and they suddenly pull out their phone, our urge to keep sharing instantly drops, because deep down we crave mutual emotional connection. We need the other person to actively see and understand us. This psychological experience of mutual recognition is the core definition of connective labor. It cannot be generated purely from one's own imagination. It requires the effort of another conscious person to help create that connection.
Connective labor differs fundamentally from general emotional labor. Emotional labor only asks employees to present a specific emotional state to customers during interactions, like a flight attendant maintaining a smile or a restaurant server acting enthusiastically. Emotional labor is a one-way performance to satisfy customers. Connective labor, however, is a reciprocal emotional exchange. Its purpose is not merely to please the other person, but to make both parties feel seen and understood in the interaction.
For example, when a psychotherapist works with a client, the counselor must actively feel the other person’s emotions and use tone, eyes, and body language to convey "I hear you," making the client feel they are being listened to. Or take teachers: a good teacher not only imparts knowledge, but also discovers a student's strengths and potential. Germany has a special pedagogical term called Bildung, which refers to cultivating students' critical thinking and potential, helping them understand how to grow into a self-aware, independent individual, rather than just an exam machine that scores well.
人机互动的去人性化危机
嘉宾: But if AI becomes widespread and these jobs are dehumanized, can the interaction between humans and machines create a similar experience of connection? If we are affirmed or praised by a machine, we don't feel like we've gained emotional value, because emotional experiences are generated through interactions between people. Therefore, the real question isn't which jobs AI will replace. It's that when AI takes over connective labor, humans may face a dehumanization crisis, because automation is inherently a process of dehumanization.
Even if AI's computing power improves and its responses become more articulate, it cannot replace the genuine connection between people, because algorithms will never truly understand what it feels like to be human. However, someone might counter: "Not necessarily; human connection can be handed over to AI too."
拥抱AI派的三种反驳观点
主持人: Now, let's hear the perspective of the opposing side. One group believes AI can outperform humans, and that even relational labor can be replaced. They argue we shouldn't think of humans as so special, as if we have something irreplaceable. They see it as mere human arrogance; for instance, AI can be designed to be even more empathetic than a real person. I think as the personalization loop kicks in and the AI just starts to get to know you better and better, I think that will just be really compelling.
Their second argument is that AI won't judge you like a real person does. For example, if you ask a "stupid" question in class, you might be mocked by teachers or peers. That’s why many students prefer asking AI in private rather than speaking up in class. Many language-learning AI apps have risen in popularity precisely because people are afraid of making mistakes. AI language platforms avoid that awkwardness because you no longer have to interact with real people. Language learners don't have to fear making mistakes, and psychological counseling is the same. People often lie or withhold their thoughts during real counseling, whether out of shame, fear of being judged, or fear of societal stigma. They worry the counselor might view them differently, so they dare not speak truthfully. But if it's an AI counselor, you don't have to worry about being judged or your secrets leaking out, allowing people to speak their minds more freely.
嘉宾: You know, that "QQ sister" thing you just mentioned.
主持人: Go ahead and say it.
嘉宾: So, being replaced by AI won't necessarily be worse; it might even be better. Back in 2009, an experiment at Boston Medical Center found that during discharge procedures, a virtual nurse could explain things more clearly than a human and even provide patients with emotional support. Surprisingly, up to 74% of patients said they preferred the virtual nurse over a human one, because medical staff are too busy, with only an average of 7 minutes to care for each patient. However, an AI system can give each patient an hour of detailed explanation, meaning AI can provide better relational labor than humans, and at a lower cost too. For example, many disadvantaged or lower-income families can access AI counseling services at an affordable price. In many U.S. states, due to a lack of educational funding, many preschools have started offering AI online courses to low-income families.
The third perspective is the one favored by AI companies: the complementary theory. AI can augment human work and make our work much more efficient. For instance, research shows that AI social robots providing emotional support to children in hospital wards can significantly reduce pain and anxiety during vaccinations or medical procedures, while also helping to lighten the nurses' workload. So, you don't need to worry about being replaced; AI isn't here to steal your job, but to lighten your burden.
Researchers conducted a field study in San Francisco at a school that was among the first in the U.S. to introduce AI. They used online platforms combined with AI resources to provide personalized, customized instruction, since every student has a different aptitude and requires different material difficulty. If a human teacher had to customize dozens of different sets of materials, it would be too time-consuming. It’s better to let AI do it, which saves the teacher time. For those with better aptitude, AI will directly help you skip ahead to higher-grade coursework. It also helps monitor whether you have completed your daily homework. Once the teachers' original instructional duties are replaced, they can shift their focus specifically to relational labor — building relationships with children, fostering human connections, and helping them develop a sense of belonging. Leave morning class time to AI, allowing students to learn autonomously. In the afternoon, teachers can interact with students, providing the relational labor that AI cannot. This creates a win-win situation and is likely the new model for AI education in the future.
After reviewing these three schools of thought, we find they are all overwhelmingly in favor of AI, but the author argues that no matter how much AI advances, relational labor remains irreplaceable by AI. Let’s take one example: real people have the ability to "read the room," sensing the atmosphere and detecting the meaning behind subtle body movements. For instance, counselors need to use their physical senses and eye contact to capture the atmosphere. In family therapy, counselors often experience a sense of tension in the air when two factions of a family are speaking past each other. Even if the two sides don’t say it explicitly, the counselor can feel that they want the counselor to take a side. This ability to sense that "smell of gunpowder" is a form of relational labor that AI cannot perform. AI can pull up all sorts of counseling theories from big data to provide explanations. Yet, it cannot perceive the atmosphere of the counseling session. The success of the relationship often depends on whether the counselor can detect these subtle signals and make the client feel heard and seen.
The same applies to school teachers; many have developed an ability to read the room, noticing when students aren't paying attention — merely pretending to listen while their minds are already elsewhere. It is hard to imagine how AI could achieve this. How could AI distinguish between someone pretending to listen while being distracted and someone who is genuinely listening but just looks vacant? Visually, these two scenarios look the same. Because AI lacks a "being-present" chip.
主持人: I apologize, that was my husband trying to be funny.
嘉宾: Finally someone to laugh at my jokes, besides you.
40年管理异化铺平AI替代之路
主持人: This kind of atmosphere perception requires a self-aware person, combined with years of accumulated experience, to be able to detect it. Now, let’s broaden our perspective and look at the evolution of work over the past 40 years. We will find that the reason AI is currently so easily replacing our jobs is that, long before AI appeared, much of this relational labor had already been in decline.
Why is that? Because for the past 40 years, the neoliberal management system has swept across the globe, and countries everywhere have begun introducing standardized SOPs in the workplace to increase efficiency. It has shifted to focus solely on KPIs and meeting monthly sales targets. This transformation actually paved the way for AI, making human jobs even easier to replace.
For example, many industries have seen the "scripting" of relationships, giving employees a manual to follow so they don't have to worry about saying or doing the wrong thing. They claim this reduces work burden and pressure; for instance, online customer service agents are required to follow a scripted process, making their work highly mechanical, monotonous, and boring. From a company's perspective, standardized scripts are meant to avoid wasting call time and to save on labor costs. However, since the emergence of AI, this type of robotic work is easily replaced, because once you feed the same script to an AI, it learns how to be an online customer service agent. So why bother hiring human agents? When relational labor is broken down into a rigid SOP, humans are merely acting out a pre-written script. The clearer and more replicable this script is, the easier it is for AI to learn, thereby replacing the relational labor once performed only by humans.
Yet, the relational labor provided by humans remains valuable; a major 2016 study analyzing 15,000 crisis hotline conversations found that when counselors avoided canned responses and replied more creatively, the caller's suicidal ideation was more likely to improve. If you're already feeling hopeless and call for help, only to receive canned responses, you would probably feel even worse.
Over the past 40 years, neoliberalism has promoted the use of quantitative performance metrics to measure work. This has caused relational labor to be increasingly undervalued, because relational labor is inherently difficult to convert into concrete metrics. In its place, we have customer satisfaction, student satisfaction, teaching evaluations, and academic output assessments, which are then used to rank universities globally. The medical system has introduced electronic health records, with more and more human data being digitized. These performance indicators were originally intended to help managers identify work efficiency. However, after decades of implementation, they have caused a decline in relational labor, as many universities prioritize rankings and force professors to focus on administrative busywork, leaving less time for teaching and interacting with students.
Many American physicians also complain that electronic health records have become a cause of professional burnout. For every hour American doctors spend with patients, they spend two hours in front of a computer writing electronic records. This time could have been used to build connections with patients. Therefore, administrative management actually hinders the relational labor of physicians.
Finally, the rise of the gig economy has implicitly disrupted relational labor, as platform orders are purely transactional. Previously, businesses had regular delivery drivers who could build connections with repeat customers, but now everyone uses platforms, the drivers change every time, making it impossible to establish any connection.
The anxiety that AI will replace human jobs stems from the fact that, for the last 40 years, capitalism has turned every profession into a set of fixed SOPs. They claimed this would increase efficiency, but it led to dehumanized interactions, making human labor increasingly redundant. We worry about AI becoming more human-like and replacing us, but the reality is the opposite: the last 40 years of capitalism have already dehumanized work, turning humans into robots.
Yet, even with this dehumanization, many personalized services still target specific consumer groups. Services like personal trainers, private financial advisors, and private doctors are all aimed at those at the top of the social pyramid. So, saying automation will cause human jobs to vanish is incorrect. A more accurate view is that relational labor will become a luxury in the future, with the elite enjoying the human-to-human service provided by those below them. However, the masses will only have access to AI-automated services. Only the wealthy will be able to afford relational labor and retain the humanity of being recognized by others. This is what AI is really bringing us.
If relational labor becomes a luxury, the question shouldn't be what jobs AI will replace, but how we can build a social structure around AI that allows relational labor to flourish instead of becoming a privilege for the rich.
The author mentions the principal effect in educational credentials. Principals rarely teach classes themselves, yet they profoundly influence how students learn. Research shows that a good principal can impact student achievement by as much as 25%. If a below-average principal is replaced by an above-average one, students gain an average of three extra months of learning per year. Why is that? The principal doesn't influence students directly; they influence the teachers' dedication. The key isn't the teaching method, but the organizational need for someone to actively care about the connections between staff. The author compared several US middle schools and found that when schools prioritized a teacher's ability to see students' needs over their academic credentials or teaching skills, it had a more positive impact on the students. This doesn't mean education and experience don't matter, but that the ability to build connections must also be on the checklist. AI may replace a teacher's instructional capacity in the future, but it cannot replace the ability to build connections with students.
If we look back at the history of supermarkets, we realize that people only start missing human interaction after relational labor disappears. A century ago, supermarkets were actually places where customers provided their shopping lists to clerks. Then, the convenience store clerk would find the items and hand them to the customer. During this process, you are bound to interact with the customer. Local merchants usually remember their regulars, making it easy to build connections. However, starting in 1916, the world’s first self-service supermarket, Piggly Wiggly in Memphis, allowed customers to find items themselves. This eliminated the need for interaction with clerks, saving companies a massive amount in labor costs. Store clerks only needed to stand at the counter to process payments. But this shift made the cashier's job increasingly robotic, repeating the same actions and phrases over and over in one spot.
嘉宾: Please open another register.
构建有温情连接的未来社会
主持人: A hundred years later, we have unmanned supermarkets, where even the cashier’s role has been replaced by machines. From the perspective of capitalist development, this labor cost saving is a major victory, but from the perspective of human development, it unconsciously causes human "relational labor" to vanish.
In 2019, the Dutch government launched a campaign to combat loneliness. Dutch supermarkets designed a dual system, allowing customers to choose between a "chat checkout" to talk with a clerk or a self-checkout. Will future schools, hospitals, and businesses share the same fate as supermarkets, becoming fully self-service?
As relational labor is automated piece by piece, we should ask ourselves what kind of society we want to build. Do we want to live in a world where most jobs are dehumanized? Where only the wealthy can afford human service, while everyone else is served by AI? The issue is no longer whether humans will lose their jobs, but that the endless pursuit of dehumanization ignores the fact that humans are a species that craves connection. If AI is unstoppable, we must think about how to build a humane, connected society within an AI-driven world.
Leave a comment and share this video post for a chance to win The Last Human Job. You can also use Super Thanks or join our channel membership to watch exclusive content.