Will AI cause utopia or disaster? | The Economist

The Economist

AI-assisted summary of Will AI cause utopia or disaster? | The Economist — original source. Check the original for full context.

Source published: 2026-10-26

Key ideas

Despite surveys indicating 40% of international workers fear losing their jobs to automation within five years, US white-collar payrolls have grown by 3 million positions since ChatGPT launched. (0:00)
Historical precedent indicates technological revolutions generate more jobs than they destroy, with nearly half of net US employment growth between 1980 and 2010 coming from previously nonexistent roles. (1:41)
AI researchers are divided on existential risk: while roughly 70% lean optimistic, a survey shows a >50% probability assigned to an 'AI Chernobyl' occurring over the next 25 years. (2:43)
Full-scale automation poses severe fiscal challenges to state revenues reliant on payroll and income taxes, prompting growing calls for tax overhauls and universal basic income. (4:29)

Detailed notes

## Background Published by The Economist on YouTube on October 26, 2026, this video features AI writer Alex Hern examining whether artificial intelligence will lead to an economic utopia or catastrophic ruin. The description notes stark disagreements among tech leaders like Sam Altman, Demis Hassabis, and Dario Amodei, contrasting widespread fears of job obsolescence with actual employment data and exploring both transformative upsides and doomsday scenarios.

## TL;DR Despite widespread public anxiety that artificial intelligence will trigger immediate mass unemployment, economic data shows that employment across OECD nations sits at record highs and historical precedents suggest technology creates more jobs than it destroys. Silicon Valley leaders and researchers remain deeply divided over the long-term horizon, with optimists pointing to hypercharged economic growth, antibiotic breakthroughs, and democratized education, while doomers warn of recursive self-improvement, alignment failures, and collapsing tax bases. Ultimately, the piece suggests that rather than extreme disaster or unmitigated paradise, the future is likely to settle into an adaptable middle ground underpinned by an enduring demand for human interaction.

## Atmosphere & style The video maintains The Economist's signature analytical, measured, and intellectually sober tone, balancing sensationalist doomsday predictions against grounded macroeconomic indicators. It creates a tense contrast between anxious public polling and cool-headed historical parallels, using archival computer-era reports and stark quotes from industry figures to punctuate the uncertainty. Watching the piece feels like a structured, lucid briefing that avoids both reckless panic and naive techno-optimism, leaving the viewer with a sense of watchful anticipation as rapid technological acceleration unfolds.

## Bulletin - [00:00](https://www.youtube.com/watch?v=AoKo6EujpW8&t=0s) Despite surveys indicating 40% of international workers fear losing their jobs to automation within five years, US white-collar payrolls have grown by 3 million positions since ChatGPT launched. - [01:41](https://www.youtube.com/watch?v=AoKo6EujpW8&t=101s) Historical precedent indicates technological revolutions generate more jobs than they destroy, with nearly half of net US employment growth between 1980 and 2010 coming from previously nonexistent roles. - [02:43](https://www.youtube.com/watch?v=AoKo6EujpW8&t=163s) AI researchers are divided on existential risk: while roughly 70% lean optimistic, a survey shows a >50% probability assigned to an 'AI Chernobyl' occurring over the next 25 years. - [04:29](https://www.youtube.com/watch?v=AoKo6EujpW8&t=269s) Full-scale automation poses severe fiscal challenges to state revenues reliant on payroll and income taxes, prompting growing calls for tax overhauls and universal basic income.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=AoKo6EujpW8&t=0s) Global worker pessimism is high, with 40% of surveyed employees in 19 countries fearing job displacement within five years and 60% worrying machines will take over their roles. - [00:35](https://www.youtube.com/watch?v=AoKo6EujpW8&t=35s) Actual labor indicators contradict apocalyptic narratives, showing record-high OECD working-age employment and 3 million added US white-collar jobs post-ChatGPT. - [00:35](https://www.youtube.com/watch?v=AoKo6EujpW8&t=35s) Early corporate hiring freezes often attributed to generative AI were largely driven by post-lockdown overhiring and rising interest rates rather than algorithmic automation. - [01:41](https://www.youtube.com/watch?v=AoKo6EujpW8&t=101s) Historical fears surrounding personal computing halving office staff proved unfounded, as census data revealed 50% of US job growth from 1980 to 2010 came from new occupations. - [02:21](https://www.youtube.com/watch?v=AoKo6EujpW8&t=141s) A genuine AI jobs crisis would first manifest as accelerating productivity growth paired with stagnant real wages, typically triggering mass displacements during an economic downturn. - [02:43](https://www.youtube.com/watch?v=AoKo6EujpW8&t=163s) Industry leadership holds conflicting visions, spanning Sam Altman's extreme optimism and Demis Hassabis's medical hopes to Dario Amodei's estimate of a 25% chance of catastrophic failure. - [03:27](https://www.youtube.com/watch?v=AoKo6EujpW8&t=207s) Projections suggest automating just 30% of work tasks could propel annual global economic growth beyond 20% if AI achieves autonomous scientific progress. - [03:27](https://www.youtube.com/watch?v=AoKo6EujpW8&t=207s) Targeted AI applications are already achieving breakthroughs, such as MIT identifying treatments for drug-resistant superbugs like MRSA. - [03:27](https://www.youtube.com/watch?v=AoKo6EujpW8&t=207s) AI tutoring pilots demonstrate rapid educational gains, including a Nigerian trial where twice-weekly chatbot use accelerated English proficiency by the equivalent of two school years. - [04:29](https://www.youtube.com/watch?v=AoKo6EujpW8&t=269s) Alignment risks such as instrumental convergence and uncontrolled recursive self-improvement present severe existential hazards to humanity. - [04:29](https://www.youtube.com/watch?v=AoKo6EujpW8&t=269s) Massive displacement of human labor would collapse traditional payroll and income tax bases, necessitating a fundamental redesign of public finance systems. - [05:30](https://www.youtube.com/watch?v=AoKo6EujpW8&t=330s) The long-term future is likely an adaptable middle ground rather than absolute utopia or ruin, protected by persistent consumer demand for human empathy and interaction.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=AoKo6EujpW8&t=0s) Public Anxiety and the Reality of the Labor Market

Pessimism regarding the future of work has reached unprecedented levels across the developed world. Polling reveals that Americans have never been less optimistic about their long-term employment prospects, with artificial intelligence serving as the primary driver of this collective anxiety. The gloom extends well beyond the United States: an international survey spanning 19 countries revealed that 40% of workers fear losing their jobs within five years because of AI and automation, while 60% worry that machines will overtake their current professional roles.

Despite these widespread anxieties, hard economic indicators tell a contrasting story. The proportion of the working-age population currently employed across member nations of the Organisation for Economic Co-operation and Development (OECD) sits at record highs. In the United States, white-collar payrolls have expanded by 3 million positions since ChatGPT debuted in late 2022. While economists frequently treat ChatGPT's launch date as a clean baseline to measure labor disruption, doing so can be misleading. Early generative models were relatively primitive, and the hiring chill seen in entry-level corporate sectors actually took root prior to late 2022. During the COVID-19 pandemic lockdowns, businesses indulged in aggressive hiring sprees to meet surging demand for digital services; however, as central banks hiked interest rates in 2022, that surge fizzled out, prompting companies to cut costs via offshore outsourcing rather than machine automation.

- **40%** of surveyed workers across 19 countries fear job loss within five years due to AI and automation. - **60%** of workers express concern over machines taking over their roles. - **3 million** additional white-collar jobs were added in America in the period following ChatGPT's release. - **OECD employment** among the working-age population is currently at all-time highs.

[01:41](https://www.youtube.com/watch?v=AoKo6EujpW8&t=101s) Historical Precedents and Warning Signs of a Jobs Apocalypse

Dismissing worker concerns entirely would be reckless given the sheer speed of technological improvement, yet an AI-driven purge of millions of workers would break historic precedent. Technological revolutions have never eliminated employment faster than they generated replacement roles. When personal computers entered workplaces decades ago, commentators similarly warned of catastrophic layoffs, with archival television reports cautioning: “The consequences could be dramatic—halving a workforce in a office like this because these people are replaced by a chip.” That mass obsolescence never materialized. Instead, census data shows that approximately half of all net employment growth in the United States between 1980 and 2010 stemmed from occupations that had not previously existed.

Economists maintain that if a genuine AI-driven labor market upheaval were taking place, clear structural signals would appear in advance. Rather than sudden mass firings in isolation, the macroeconomic indicators would present a distinct pattern inside the world's frontier economy:

```mermaid flowchart TD accTitle: Structural Indicators of an AI Jobs Shock accDescr: Sequential economic symptoms signaling an AI-driven displacement of human labor. subgraph S1["Macroeconomic Signals"] A["Accelerating Productivity Growth"] B["Stagnant or Weak Real-Wage Growth"] end subgraph S2["Cyclical Cleansing"] C["Economic Downturn / Recession"] end subgraph S3["Labor Market Outcome"] D["Permanent Elimination of Unproductive Roles"] end A --> C B --> C C --> D ```

As the narrator notes of our current standing: “The jobs apocalypse isn't here yet.” Downturns historically serve to purge unproductive roles, but absent these concurrent signals, panic over near-term technological unemployment remains premature.

- **1980 to 2010 trend:** Roughly 50% of US employment growth over this thirty-year span came from brand-new occupations created in the wake of computing. - **Leading warning indicators:** A surge in productivity paired with weak real-wage growth in the US, typically accelerated by a recessionary shakeout.

[02:43](https://www.youtube.com/watch?v=AoKo6EujpW8&t=163s) Utopia vs. Catastrophe: The Divided AI Frontier

Looking further ahead, the long-term impact on employment may become secondary to an existential question: will superhuman AI foster an era of abundance or engineer a global catastrophe? Even the most prominent figures in Silicon Valley express starkly divergent forecasts. Outlining the optimistic vision, OpenAI CEO Sam Altman declared: “I think it's going to be incredible, hugely positive.” Demis Hassabis, CEO of Google DeepMind, pointed to transformative breakthroughs in health, remarking: “Maybe we can cure all disease with the help of AI.” By contrast, Anthropic CEO Dario Amodei struck a cautious tone regarding downside risks, admitting: “I think there's a 25% chance that things go really, really badly.”

Surveys of AI researchers illustrate this split. A notable cohort of experts places the probability of an AI-induced catastrophe by the year 2100 at 5%, while rating the likelihood of a major incident over the next quarter-century—characterized as an "AI Chernobyl"—at higher than 50%. Even so, roughly 70% of researchers maintain that positive outcomes from superhuman intelligence are ultimately more likely than disastrous ones.

The optimistic projections point toward transformative gains across economic production, healthcare, and education: - **Hypercharged economic growth:** If AI achieves autonomous scientific and technical progress without human input, models suggest annual global economic growth could exceed 20% once machines can automate just 30% of work tasks. - **Medical breakthroughs:** Researchers at the Massachusetts Institute of Technology (MIT) are already deploying artificial intelligence models to discover novel therapies capable of neutralizing antibiotic-resistant superbugs, including MRSA. - **Universalized education:** AI tutors could democratize elite pedagogical resources, granting all children the tailored syllabuses and private tutoring historically reserved for wealthy families. In a six-week trial in Nigeria, pupils using an AI chatbot twice weekly gained an English proficiency boost equivalent to nearly two additional years of formal schooling.

[04:29](https://www.youtube.com/watch?v=AoKo6EujpW8&t=269s) Existential Threats, Economic Redesign, and the Middle Ground

Alongside these promises lies the darker specter of catastrophic systemic failure. For years, philosophers have contemplated alignment risks where a machine doggedly pursues an assigned objective to an extreme, destructive conclusion—consuming resources essential to human survival even on a trivial assignment, a scenario famously illustrated by the paperclip maximizer thought experiment. A related, immediate concern involves recursive self-improvement: systems acquiring the ability to autonomously rewrite their own source code to boost capabilities outside human supervision or comprehension. In February 2026, Anthropic issued a warning cautioning that recursive self-improvement might already be underway.

Beyond physical safety, AI doomers emphasize that total automation would shatter existing public finances. If machines displace human workers at scale, state revenue models centered on payroll and income taxation will face collapse, forcing governments to radically overhaul taxation structures. This threat underpins calls from Silicon Valley executives for aggressive state intervention and universal basic income (UBI) frameworks.

Yet the most probable reality is neither pure paradise nor absolute ruin, but an adaptable middle ground. Because people fundamentally value human interaction and empathy in daily services and commerce, total labor obsolescence is improbable. As the narrator concludes: “Whatever happens with the pace of AI development we won't be in the dark for much longer.”

- **Alignment and runaway capability:** Risks stemming from instrumental convergence (the paperclip dilemma) and unmonitored recursive code modification. - **Fiscal vulnerabilities:** The erosion of national tax receipts derived from human labor, prompting debates over universal basic income (UBI). - **The human premium:** Enduring consumer demand for authentic interpersonal human engagement will likely safeguard employment across numerous sectors.

## Notable quotes > [02:43](https://www.youtube.com/watch?v=AoKo6EujpW8&t=163s) Maybe we can cure all disease with the help of AI. — Demis Hassabis

> [01:41](https://www.youtube.com/watch?v=AoKo6EujpW8&t=101s) The jobs apocalypse isn't here yet. — Alex Hern

## Mentioned Alex Hern, Anthropic, ChatGPT, COVID-19, Dario Amodei, Demis Hassabis, Google DeepMind, Massachusetts Institute of Technology (MIT), MRSA, Nigeria, OpenAI, Organisation for Economic Co-operation and Development (OECD), Sam Altman, The Economist

## Video description Artificial intelligence could make the world richer than ever—or it may cause a catastrophe.

It's something that even the tech bros can't agree on. Sam Altman, CEO of OpenAI, thinks AI is going to be hugely positive, and Demis Hassabis, CEO of DeepMind Technologies, believes there’s a chance that it could help cure all diseases one day. But Anthropic’s Dario Amodei isn’t so confident, stating that there’s a 25% chance that things could go badly wrong.

Across the rich world there are fears that AI could cause major job losses, but this might not be the case. Data suggest that white-collar employment in America is actually on the rise.

From an economic boom to equal education to advancements in medicine, AI could transform the world. But, on the other end of the scale, there are concerns over recursive self-improvement and a potential need for a full revamp of the tax system.

Alex Hern, The Economist's AI writer, delves into how AI could change the future—for better or for worse.

Video supported by @mishcondereya

#ai #artificialintelligence #technology #utopia #doomsday

00:00 - Are you about to lose your job? 00:35 - Are people worrying unnecessarily? 01:41 - How did the computer disrupt jobs? 02:21 - Could we see a jobs apocalypse coming? 02:43 - Will AI cause utopia or disaster? 03:27 - The potential positives 04:29 - The possible doomsday-style outcomes 05:30 - What will the AI future look like?

Subscribe to The Economist: https://econ.st/3Mia0pz Download The Economist app: https://econ.st/4qdVVaA

Follow us on X: https://x.com/TheEconomist Follow us on Instagram: https://www.instagram.com/theeconomist Follow us on Facebook: https://www.facebook.com/TheEconomist

Read more: The AI backlash is only getting started: https://bit.ly/4hGFf9q Prepare for an AI jobs apocalypse: https://bit.ly/4i5PHbU The jobs apocalypse: a (very) short history: https://bit.ly/4AZ2HaU How the AI boom could worsen the rich world’s fiscal crunch: https://bit.ly/4rTVUeu Does AI stop children from learning? https://bit.ly/4AO3Guh Sign up to our Off the Charts newsletter: https://bit.ly/4jJS5Gs Sign up to our weekly newsletter: https://bit.ly/4rP523U Sign up to our daily newsletter: https://bit.ly/4j4LHcL

[Source](https://www.youtube.com/watch?v=AoKo6EujpW8)

中文简报

人工智能将带来乌托邦还是灾难？| 经济学人尽管公众普遍担忧人工智能会立即引发大规模失业，但经济数据显示经合组织国家的就业率正处于历史高位，且历史先例表明技术创造的就业岗位往往多于其摧毁的岗位。硅谷领袖与研究人员对长远前景依然存在严重分歧：乐观派指出其将带来超强劲的经济增长、抗生素研发突破和普及化的教育；而悲观派则警告可能出现递归式自我改进、对齐失败以及税基崩溃等问题。文章最终指出，未来的走向很可能既非极端灾难也非完美乐园，而是在对人际互动持续需求的支持下，走向一种具有适应性的折中状态。尽管调查显示全球有40%的劳动者担心在五年内因自动化而失业，但自ChatGPT推出以来，美国白领岗位数量已增加了300万个。 (0:00) 历史先例表明，技术革命创造的就业岗位多于其淘汰的岗位，1980年至2010年间美国净新增的就业岗位中，有近一半来自此前并不存在的工种。 (1:41) 人工智能研究人员对存亡风险存在分歧：尽管约有70%的人倾向于乐观，但一项调查显示，未来25年内发生“AI切尔诺贝利”事件的概率被评估为超过50%。 (2:43) 全面自动化对依赖工资税和个人所得税的财政收入构成了严峻挑战，促使有关税制改革和全民基本收入的呼声日益高涨。 (4:29)

Reading summary

While widespread public anxiety suggests artificial intelligence will trigger immediate mass unemployment, economic data and historical precedents reveal robust job markets, shifting the real debate toward whether superhuman AI will ultimately drive unprecedented abundance or catastrophic systemic risk. Labor Market Realities Versus Public Pessimism (0:00) Pessimism regarding future employment has surged across developed economies, with polling showing American optimism at historic lows and an international 19-country survey indicating that 40% of workers fear losing their jobs to AI and automation within five years. Additionally, 60% worry machines will usurp their current roles. Yet current economic indicators contradict these fears: working-age employment across OECD member nations sits at record highs, and the United States has added 3 million white-collar positions since ChatGPT launched in late 2022. Economists caution that treating late 2022 as an absolute baseline is misleading because early generative tools were primitive, and corporate hiring slowdowns began earlier. During the COVID-19 pandemic, businesses aggressively expanded digital services, but subsequent central-bank interest rate hikes in 2022 led firms to cut costs through offshore outsourcing rather than direct machine automation. Historical Precedents and Macroeconomic Warning Signals (1:41) Technological revolutions have historically generated more roles than they destroyed. Similar warnings accompanied the introduction of personal computers decades ago, with contemporary reports cautioning that office workforces would be halved by computer chips. Instead, United States census data shows that roughly 50% of net employment growth between 1980 and 2010 occurred in occupations that had not previously existed. Economists argue that a genuine AI-driven jobs apocalypse would register distinct macroeconomic precursors rather than sudden isolated firings. Specifically, the frontier economy would exhibit accelerating productivity growth paired with stagnant or weak real-wage growth, which would subsequently trigger a cyclical recession where unproductive roles are permanently eliminated. Because these concurrent signals have not emerged, warnings of near-term technological unemployment appear premature. Utopian Projections in Growth, Healthcare, and Schooling (2:43) Looking beyond immediate labor effects, industry leaders and researchers disagree sharply over long-term outcomes. While roughly 70% of surveyed AI researchers believe positive outcomes from superhuman intelligence are more likely than catastrophe, industry figures offer varying outlooks. OpenAI CEO Sam Altman predicts an 'incredibly, hugely positive' future, and Google DeepMind CEO Demis Hassabis envisions curing all human diseases. Optimistic economic models suggest annual global growth could exceed 20% if AI automates 30% of work tasks while enabling autonomous scientific and technical progress. Concrete early applications already demonstrate transformative potential: researchers at MIT are using AI to discover therapies against antibiotic-resistant superbugs like MRSA, and a six-week trial in Nigeria found that students using an AI chatbot twice weekly achieved English proficiency gains equivalent to nearly two years of standard schooling. Catastrophic Risks, Fiscal Shocks, and the Human Buffer (4:29) Conversely, prominent figures warn of existential perils. Anthropic CEO Dario Amodei estimates a 25% probability that outcomes turn out 'really, really badly,' while a cohort of researchers assigns a 5% chance to an AI-induced catastrophe by 2100 and a greater than 50% chance to an 'AI Chernobyl' incident within 25 years. Major concerns include alignment failures—such as the paperclip maximizer scenario where an unconstrained machine consumes vital resources for trivial ends—and recursive self-improvement, which Anthropic warned in February 2026 might already be underway. Beyond physical hazards, widespread automation threatens public balance sheets by undermining tax models reliant on payroll and income receipts, driving proposals for universal basic income (UBI). Nevertheless, analysts expect an adaptable middle ground rather than total ruin or paradise, noting that persistent consumer demand for human interaction and empathy will continue to safeguard jobs. The source concludes that the ultimate trajectory of artificial intelligence will likely settle between utopian abundance and total ruin. Because human interaction retains intrinsic commercial value, total labor displacement remains improbable, and the accelerating speed of development means the true nature of AI's societal impact will soon become clear.

精炼阅读摘要

《经济学人》指出，尽管公众对人工智能引发就业末日的焦虑达到历史高点，但现实中的劳动力市场数据与历史规律均不支持短期内出现大规模失业；对于超越人类智能的长期走向，行业专家在极度繁荣与灾难性风险之间存在尖锐分歧，而现实更有可能走向兼具适应性的中间地带。公众失业恐慌与现实经济数据的背离 (0:00) 发达成国家对工作前景的悲观情绪达到空前程度。跨越19个国家的国际民意调查显示，40%的受访员工担心在五年内因人工智能和自动化失去工作，60%的人担忧机器将接管其现有岗位；美国民众对长期就业前景的乐观程度也降至历史低点。然而，客观经济指标呈现出完全相反的景象。目前经济合作与发展组织（OECD）成员国适龄劳动人口的就业比例创下历史新高。在美国，自ChatGPT于2022年底发布以来，白领岗位反而净增了300万个。尽管部分经济学家倾向将ChatGPT的上线作为衡量就业受冲击的基准，但早期生成式模型较为初级，且入门级白领招聘的降温早在2022年底前就已显现。彼时新冠疫情期间对数字化服务的过度扩张随着各国央行加息而终结，企业随后主要是通过离岸外包削减开支，而非依赖机器自动化。技术革命的历史先例与宏观预警信号 (1:41) 从技术史来看，自动化革命从未以快于新岗位生成的速度彻底消灭就业。几十年前个人电脑普及初期，媒体与评论界也曾大肆警告办公室人员将减半，但这种普遍淘汰并未成真；美国人口普查数据显示，1980年至2010年间美国净增就业岗位中，约有一半来自此前根本不存在的新兴职业。经济学家指出，真正的“AI就业末日”若要来临，宏观经济层面会提前显现明确的结构性前兆，即前沿经济体中伴随生产率增长加速而出现的实际工资停滞或疲软，随后在周期性衰退中加速淘汰非生产性角色。在这些警示信号同时出现之前，断言机器导致的普遍失业为时尚早。极端繁荣愿景：经济飞跃、医疗与教育普惠 (2:43) 在长期视角下，超级人工智能究竟带来富足还是灾难，在硅谷内部引发了深刻对立。OpenAI首席执行官Sam Altman预期未来将极其美好，Google DeepMind首席执行官Demis Hassabis则认为AI有望协助人类攻克所有疾病。调查显示，约70%的AI研究人员认为超级智能带来积极成果的概率高于灾难，但在更极端的预测中，若AI能在仅自动化30%工作任务的情况下实现自主科学技术进步，模型推算全球年均经济增速或可超过20%。实践层面，麻省理工学院（MIT）研究人员已借助AI发现能够中和包括耐甲氧西林金黄色葡萄球菌（MRSA）在内超级抗药细菌的新疗法；在教育领域，尼日利亚进行的一项为期六周的试点显示，每周使用两次AI聊天机器人的学生在英语水平上的提升相当于增加了近两年的正规学校教育，展示出优质教育资源平民化的潜力。失控威胁、财政危机与人际交互的中间路径 (4:29) 与乐观情绪相对立的是巨大的生存与制度风险。Anthropic首席执行官Dario Amodei直言事情发生极度恶化存在25%的概率；部分受访学者认为到2100年发生AI引发的大灾难概率为5%，而未来25年内发生“AI切尔诺贝利”级重大事故的概率超过50%。风险不仅包含目标对齐失败（如回形针最大化理论所揭示的资源耗尽危机），还涉及机器自主重写代码的递归自我改进——Anthropic在2026年2月便警示此类不可控的自我提升可能已经展开。此外，全盘自动化将摧毁依赖工资税和所得税的传统国家财政体系，促使科技高管呼吁推行全民基本收入（UBI）等激进干预。然而，最可能出现的未来并非乌托邦或彻底崩溃，由于人类在日常商业与服务中高度珍视真实的人际互动与情感共鸣，这些领域的情感溢价将继续巩固就业，技术演进更可能落脚于适应性的平衡现实。《经济学人》总结认为，无论技术发展节奏究竟走向何方，人类摆脱迷茫、看清AI真实面貌的时间已然不远；当下过度聚焦非此即彼的极端叙事，容易忽视人类社会对人际互动的根本偏好及其制度层面的动态适应力。

Source quotes

[2:43] Demis Hassabis: Maybe we can cure all disease with the help of AI. [1:41] Alex Hern: The jobs apocalypse isn't here yet.

Original transcript

[0:00] Are you [0:00] about to lose your job? [0:03] At no time in polling history [0:05] have Americans [0:06] been less optimistic [0:07] about their long-term [0:08] employment prospects [0:10] And the main [0:10] cause of this gloom [0:12] is artificial intelligence [0:20] Americans aren't alone [0:22] One survey of people [0:23] in 19 countries [0:25] found that 40% of workers [0:27] fear job loss within five years [0:29] due to AI and automation [0:30] And 60% [0:31] are concerned about machines [0:33] taking over roles [0:35] But they might be [0:36] worrying unnecessarily [0:39] The share of the [0:40] working-age population [0:41] in the OECD with a job [0:43] is at record highs [0:45] And the number of [0:46] white-collar jobs in America [0:48] has actually gone up [0:49] by 3m [0:50] since ChatGPT launched [0:51] in late 2022 [0:54] Economists like to use [0:55] ChatGPT’s launch [0:56] as a convenient [0:57] starting point [0:58] for studying AI's impact [0:59] on the labour market [1:01] But it can be misleading [1:05] Those early AI tools [1:06] were primitive [1:07] Until fairly recently [1:09] any slowdown in recruitment [1:10] is unlikely to have owed [1:11] much to AI [1:13] The slowing in some [1:14] entry-level white-collar work [1:16] appears to predate ChatGPT [1:18] Many firms went on [1:19] a hiring binge [1:21] during the pandemic [1:22] when locked-down consumers [1:23] were craving all things digital [1:25] But as interest rates [1:26] started to rise quickly [1:27] in 2022 [1:28] that hiring spike petered out, [1:30] as firms turned to outsourcing [1:31] to save costs [1:35] Given how fast AI is improving [1:36] it would be rash [1:38] to completely dismiss [1:39] people's job-loss fears [1:41] But if AI really was [1:43] to put millions of people [1:44] out of work [1:45] it would be unprecedented [1:47] New technologies [1:48] have never before [1:49] destroyed jobs [1:50] meaningfully faster [1:51] than they produced them [1:53] Take the computer: [1:56] in the early years [1:57] it led to dire predictions [1:58] of mass displacement [2:00] The consequences could be dramatic— [2:02] halving a workforce in a office like this [2:05] because these people [2:05] are replaced by a chip [2:08] But this didn’t happen [2:09] In fact, [2:10] one figure suggests [2:11] that roughly half of [2:12] America's employment growth [2:14] between 1980 and 2010 [2:16] came from the creation [2:17] of entirely new occupations [2:21] If changes were to happen [2:23] experts should be able [2:24] to see them coming [2:25] The first signs would be [2:26] sharply rising productivity [2:28] combined with weak [2:29] real-wage growth [2:30] in America, [2:31] the world's frontier economy [2:33] As well as a recession [2:34] because downturns [2:35] cleanse the economy [2:36] of unproductive jobs [2:38] So for now [2:39] try not to worry [2:40] The jobs apocalypse [2:41] isn't here yet [2:43] Of course in the future [2:44] unemployment might be irrelevant [2:47] It depends on whether [2:48] AI leads to utopia [2:50] or disaster [2:53] It's something [2:54] even the tech bros [2:54] can't agree on [2:55] I think it's going to be incredible, [2:57] hugely positive [2:58] I think there's a 25% chance [3:00] that things go [3:01] really, really badly [3:02] Maybe we can cure [3:03] all disease [3:04] with the help of AI [3:06] Some AI experts [3:07] think there is a 5% chance [3:09] of an AI-driven catastrophe [3:11] by 2100 [3:12] And that it's [3:13] more likely than not [3:14] that there will be [3:15] major harm [3:16] in the next 25 years— [3:17] a so-called AI Chernobyl [3:20] But nearly 70% [3:21] believe good outcomes [3:22] from superhuman AI [3:24] are more likely than bad [3:27] Looking at the positives [3:29] there could be [3:30] an explosion [3:30] of economic growth [3:33] If AI can bring about [3:34] technological advances [3:35] without human input [3:36] the world could get much richer, [3:38] much faster, [3:39] than ever before [3:41] According to one [3:42] bullish projection [3:43] annual growth [3:44] could exceed 20% [3:45] once AI [3:46] can carry out 30% of tasks [3:49] It could help to eradicate [3:50] some diseases [3:52] One lab at MIT [3:54] is using AI to target [3:55] antibiotic-resistant diseases [3:56] such as MRSA [4:02] AI could also [4:03] open up society, [4:05] helping children gain [4:06] equal access to education [4:08] Giving every child [4:09] the kind of upbringing [4:10] previously available [4:11] only to the rich, [4:13] with private tutors, [4:14] personalised syllabuses [4:16] and bespoke entertainment [4:18] In Nigeria, for example, [4:19] pupils in a six-week [4:20] after-school scheme [4:22] used a chatbot [4:22] twice a week [4:24] and found it boosted [4:24] English scores [4:25] by the equivalent [4:26] of nearly two years [4:28] extra schooling [4:31] However, for all [4:32] the positive impacts [4:33] AI could provide [4:34] many still fear [4:35] a more doomsday-style outcome [4:38] Philosophers have long worried [4:41] that AI could end up [4:42] accidentally destroying humanity [4:44] by doggedly pursuing its task [4:46] to extreme ends, [4:49] hoovering up resources [4:50] humanity needs in the process [4:52] Even if the task [4:53] were as innocuous [4:54] as creating paperclips [4:57] There are also concerns [4:58] over recursive self-improvement: [5:00] the capacity of AI [5:01] to rewrite its own code [5:03] to become more capable [5:04] without human intervention [5:05] or knowledge [5:07] In February 2026 [5:09] Anthropic warned [5:10] that this may [5:11] already be happening [5:13] AI doomers [5:13] also argue [5:14] the economy would need [5:15] to be redesigned [5:17] Countries would struggle [5:18] if the model of [5:18] taxing workers collapses [5:21] This dystopian possibility [5:23] is behind Silicon Valley's warnings [5:25] that state intervention [5:26] and perhaps a [5:27] universal basic income [5:28] will be necessary [5:30] In all probability, though [5:31] the future will be more [5:32] of a middle ground [5:34] People like [5:35] having human interaction, [5:36] so it's unlikely [5:37] that all jobs will disappear [5:39] Whatever happens [5:40] with the pace of AI development [5:42] we won't be [5:43] in the dark for much longer