The AI race: why some people are scared and others aren't

Sam Altman says he knows how to build artificial general intelligence. Geoffrey Hinton, a Nobel laureate, says he can’t see any path that guarantees safety. Yann LeCun, a Turing Award winner, thinks those fears are nonsense. Meanwhile, Big Tech is spending hundreds of billions of dollars a year. Here are the facts, camp by camp, followed by our take.

In short: the race is mostly between American and Chinese labs, with Europe trying to keep up, and their leaders promise artificial general intelligence (AGI) within a few years. The worried camp (Hinton, Bengio) fears losing control; the skeptics (LeCun, Gary Marcus) think capabilities are overhyped; others fear a financial bubble. The latest data shows falling employment for 22-to-25-year-olds in the most AI-exposed jobs, but no mass job losses across the economy. The optimists point to results already measured in science, health, and productivity. For a student, the right response is neither panic nor euphoria: it’s building skills that AI complements.

The race: who’s running, and what they promise

• The US has the money. According to Stanford’s 2026 AI Index, US private AI investment reached $285.9 billion in 2025, more than 23 times the $12.4 billion invested in China.
• China has caught up technically. The same report says the performance gap between US and Chinese models “has effectively closed”: DeepSeek-R1 briefly matched the top US model in February 2025. More in our piece on Chinese AI.
• Europe is betting on infrastructure and a few champions. In February 2025 the EU launched InvestAI to mobilise €200 billion. Mistral AI raised €1.7 billion in September 2025, and Yann LeCun’s Paris start-up, AMI Labs, raised $1.03 billion in March 2026 (we cover that bet here).

As for promises, the heads of the three most prominent US labs all point to short timelines:

• Sam Altman (OpenAI), early January 2025: “We are now confident we know how to build AGI as we have traditionally understood it” (Reflections). In June 2025 he opened a post with “We are past the event horizon; the takeoff has started” (The Gentle Singularity).
• Dario Amodei (Anthropic) described powerful AI in October 2024 as a “country of geniuses in a datacenter” that “could come as early as 2026”, while admitting it could take much longer (Machines of Loving Grace). In January 2026 he wrote it may be “as little as 1–2 years away” (The Adolescence of Technology).
• Demis Hassabis (Google DeepMind) talked about a five-to-ten-year horizon in March 2025, then in February 2026 said AGI is “on the horizon, maybe in the next five to eight years” (Business Today).

The worried camp: “we don’t know how to control what we’re building”

Geoffrey Hinton is its best-known voice. In May 2023 he left Google to speak freely about the risks, after saying he had “suddenly switched” his views on whether these systems would become more intelligent than us (MIT Technology Review). When he won the 2024 Nobel Prize in Physics with John Hopfield, he mentioned “the threat of these things getting out of control” the same day (University of Toronto). On 60 Minutes, he put it bluntly: “I can’t see a path that guarantees safety” (CBS News).

Yoshua Bengio, another Turing Award winner, chairs the International AI Safety Report. Its second edition (February 3, 2026, more than 100 experts, backed by over 30 countries and organisations) notes that AI systems can already find software vulnerabilities and that some models increasingly tell a test apart from real-world use. Its core point: capabilities are moving fast, while evidence about risks arrives slowly. Bengio also founded LawZero, a nonprofit building “safe-by-design” AI; on September 16, 2026, Canada and Germany committed up to CAD 300 million to it.

Two collective texts shaped this camp:

• The Future of Life Institute open letter of March 22, 2023 called for a pause of at least six months on training systems more powerful than GPT-4. Over 31,000 signatures, no pause.
• The Center for AI Safety’s Statement on AI Risk (May 2023) is a single sentence: mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war. Striking detail: Altman, Amodei, and Hassabis signed it.

The most radical version is a book: If Anyone Builds It, Everyone Dies, by Eliezer Yudkowsky and Nate Soares, published on September 16, 2025 by Little, Brown. The thesis is in the title: a superhuman AI built with current techniques would kill everyone, so the race has to stop.

The skeptical camp: “it’s mostly hype”

Yann LeCun is the most famous skeptic. In October 2024, asked by the Wall Street Journal about AI as an existential threat, he called it “complete B.S.” (TechCrunch). His argument: today’s models lack what a house cat has, namely persistent memory, reasoning, planning, and an understanding of the physical world. LeCun isn’t anti-AI, though: he’s betting on a different approach.

Gary Marcus, a cognitive scientist, published “Deep Learning Is Hitting a Wall” in Nautilus back in March 2022. He argues that deep learning alone lacks reliability and reasoning, and advocates hybrid systems that combine neural networks with symbolic logic.

For Princeton researchers Arvind Narayanan and Sayash Kapoor, AI is a “normal technology”: like electricity, its impact will spread over decades, slowed by safety requirements, how organisations adapt, and slow adoption.

This camp’s weak spot: capabilities have often outrun its predictions. In July 2025, a version of Gemini scored 35 out of 42 at the International Mathematical Olympiad, a gold-medal score certified by IMO coordinators.

The bubble fear

Here the question isn’t “is AI dangerous?” but “will the money come back?”

The spending. According to JPMorgan, as reported by Fortune in June 2026, capital spending by the big cloud providers should reach $650 billion in 2026 and top $1.1 trillion in 2027, with global AI-related spending hitting $5.5 trillion by 2030. The bank thinks the economics hold for now, with one main risk: demand that doesn’t grow fast enough.

The circular deals. In autumn 2025, Nvidia announced plans to invest up to $100 billion in OpenAI, which buys Nvidia chips; AMD gave OpenAI the right to acquire roughly 10% of its shares in exchange for large GPU purchases; Oracle signed a $300 billion, five-year cloud deal with OpenAI and buys Nvidia chips itself (The Register). Money circulates among the same players, which can inflate the appearance of growth. One analyst quoted by NBC News flagged “related-party transactions.” Sam Altman himself says there will be “booms and busts.”

The returns. In August 2025, an MIT report (NANDA initiative) found that about 5% of corporate generative AI pilots achieve rapid revenue acceleration, with the vast majority showing no measurable impact (Fortune). Method: 150 interviews, a survey of 350 employees, 300 public deployments. A signal, not a law. Where the money goes: why AI is so expensive.

What about new graduates?

The best-documented answer comes from Stanford. In the August 2026 revision of “Canaries in the Coal Mine?”, Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen analyse ADP payroll data through June 2026. Their findings:

• employment of 22-to-25-year-olds in the most AI-exposed occupations is 19% below where it would be had it kept pace with less-exposed peers, and the gap has widened since the first version in August 2025;
• the effect works mainly through fewer young hires, not layoffs;
• declines are concentrated where AI substitutes for tasks; where it complements workers, employment is flat or rising;
• they find no evidence of widespread, economy-wide job displacement.

The authors call these descriptive signals, not causal estimates. The International AI Safety Report likewise sees no overall job loss so far. Dario Amodei, for his part, predicts AI could displace half of all entry-level white-collar jobs within 1 to 5 years: a forecast, not a measurement.

The optimistic camp: “look at what it already does”

The optimists point to measured results:

• Science. Demis Hassabis and John Jumper won the 2024 Nobel Prize in Chemistry for AlphaFold, which predicts the 3D structure of proteins. According to Google DeepMind, its database has served more than 2 million researchers in 190 countries.
• Productivity. Across 5,179 customer support agents, a generative AI assistant raised productivity by 14% on average and 34% for novices, with minimal impact on the most experienced (NBER).
• Health. In the randomised MASAI trial published in The Lancet Digital Health, AI-supported breast cancer screening detected 6.4 cancers per 1,000 women screened versus 5.0 with standard double reading, with no rise in false positives and 44% less screen-reading work for radiologists (Europe PMC).

The camps at a glance

Camp	Key figures	Core idea	What supports it	What weakens it
Accelerators	Altman, Amodei, Hassabis	AGI is a few years away	Fast progress, IMO gold	Direct interest in raising money
Worried	Hinton, Bengio, Yudkowsky	We can’t control what we build	International report, AI finding vulnerabilities	Scenarios hard to test
Skeptics	LeCun, Marcus, Narayanan	Capabilities and timelines are overhyped	Enterprise failures, persistent errors	Predictions often overtaken
Bubble watchers	Economists, tech observers	Spending outruns revenue	Circular deals, 5% of pilots paying off	JPMorgan sees the economics holding for now
Optimists	Applied researchers	Benefits are already measurable	AlphaFold, MASAI trial, productivity	Uneven gains across jobs

Our take

Our take: these camps contradict each other less than it seems. The 2026 AI Index sums it up in one line: AI models can win a gold medal at the International Mathematical Olympiad but cannot reliably tell time. Researchers call this the “jagged frontier” of AI. Optimists look at one side, skeptics at the other, and both are right.

Second: always check who’s talking. Those announcing AGI for tomorrow are raising billions; LeCun, who says the opposite, is building an alternative. Nobody is neutral.

Third: a financial bubble and a real transformation can coexist, as the internet and the 2000 crash showed. The scenario we find most likely is a powerful but uneven technology, a possible financial correction, and job effects concentrated on the entry-level tasks AI substitutes for.

What it means for you as a student

1. Aim for tasks AI complements. The Stanford data is clear: employment falls where AI substitutes, not where it helps. Judgement, human relationships, accountability, fieldwork: that’s where your value builds up.
2. Keep your fundamentals. If AI does the exercise for you, you won’t be able to check what it produces. Use it as a coach that explains and asks questions (studying with AI without cheating).
3. Learn to read a claim. Who published this number? A benchmark announced by a model’s own maker is a self-reported measurement. Go back to the primary source.
4. Build concrete proof. If junior hiring tightens, an internship, a published project, or a portfolio count for more than a line on a CV.
5. Choose with data, not fear. Look at real graduate outcomes for your programme before deciding.

FAQ

Will AI wipe out jobs for new graduates?

According to Stanford (August 2026), employment of 22-to-25-year-olds in the most exposed occupations is 19% below the trend of their less-exposed peers, mostly because fewer of them are hired. But there’s no sign of mass job losses across the economy, and employment is growing where AI complements the work.

What is AGI, and when will it arrive?

AGI means an AI able to perform the full range of human cognitive tasks. The heads of OpenAI, Anthropic, and Google DeepMind talk about one to eight years; Yann LeCun thinks current methods won’t get there. Nobody can give a reliable date.

Is there an AI bubble?

Spending is enormous ($650 billion expected in 2026 from the big cloud providers, according to JPMorgan) and part of the money circulates among the same companies. A correction is possible, without the technology ceasing to be useful.

Should I avoid fields exposed to AI?

Not necessarily: within the same field, roles where AI acts as support are growing. Check graduate outcomes for your programme and focus on skills that AI complements.

Further reading

• Why Yann LeCun wants AMI: AI beyond LLMs: the bet of the most famous skeptic, explained simply.
• AI, internships, apprenticeships, and your first job: how to approach entering the job market with AI around.
• Choosing a degree with graduate-outcome data: a method for deciding on real numbers.
• The insertion-pro connector: query public graduate-outcome data straight from your AI assistant.