The AI debate

Article

People who think hard about AI disagree, sometimes sharply. This series explains the main camps (accelerationists, those most worried about catastrophe, safety-focused builders and researchers, and sceptics) in terms their own members would recognise, and sets out what would have to happen for their hopes or fears to come true.

Article

Some of the most credible people in AI believe it could make expert help, scientific progress, learning and clean energy far more widely available. Here is that hopeful case in their own words, the real evidence of falling costs so far, the honest counter-arguments, and, for each promised benefit, what would have to happen for it to come true and which signs to watch.

Article

Accelerationists and techno-optimists believe faster technological progress, including AI, is the surest route to longer, richer and freer lives, and that slowing it down carries real human costs. Here is their case in their own words, the strongest objections to it, and what would have to be true for them to be right.

Article

Some serious researchers believe advanced AI could lead to human extinction, and 'p(doom)' has become shorthand for how likely people think that is. This article explains the term, shows the real spread of expert estimates, explains why a single number can mislead, and sets out honestly why a minority hold very high estimates and why most experts are more hopeful.

Article

A calm look at the main risks people raise about AI (jobs, misuse, loss of control, concentration of power, surveillance, misinformation and accidents). For each one: what would have to happen for it to become real, what the evidence shows so far, which signs would tell us it is growing or easing, and what is already being done, globally and in the Gulf.

Article

A large and influential group of researchers, company leaders and policymakers believe AI could be both enormously beneficial and seriously dangerous, and that safety science, testing and governance must keep pace with capability. Here is who they are, what they propose, the criticisms they face from both directions, and how to tell whether their approach is working.

Article

Sceptics come in several kinds: those who doubt that today's AI methods lead to human-level intelligence, those who doubt that AI poses an existential threat, and those who think both hype and doom distract from real, present-day harms. Here is their case in their own words, where it has been challenged, and what evidence would prove them right or wrong.

Article

The 'technological singularity' is the idea that AI might one day improve itself so quickly that the future becomes impossible to predict. Here is where the idea came from (I. J. Good, Vernor Vinge, Ray Kurzweil), how its predictions have fared, what its critics say, and how to think about it calmly.

Tracker entry

"We believe Artificial Intelligence can save lives – if we let it." "We believe any deceleration of AI will cost lives."

Tracker entry

The goal is to make intelligence and energy extremely cheap (manifesto).

Tracker entry

"2026 will likely see the arrival of systems that can figure out novel insights."

Tracker entry

"2027 may see the arrival of robots that can do tasks in the real world."

Tracker entry

"In the 2030s, intelligence and energy… are going to become wildly abundant."

Tracker entry

"The most likely result of building a superhumanly smart AI, under anything remotely like the current circumstances, is that literally everyone on Earth will die."

Tracker entry

Call for an indefinite, worldwide moratorium on large training runs, enforced by international agreement and monitoring of hardware.

Tracker entry

*If Anyone Builds It, Everyone Dies*: argues building superintelligence would lead to human extinction; says creation "can still be prevented".

Tracker entry

"Pause for at least 6 months the training of AI systems more powerful than GPT-4."

Tracker entry

10–20% chance AI leads to human extinction within three decades.

Tracker entry

"25% chance that things go really, really badly"; 75% really, really well.

Tracker entry

Extinction from AI by 2100: superforecasters median 0.38%, AI experts median 3%.

Tracker entry

LawZero will develop "Scientist AI", a non-agentic system to act as a guardrail.

Tracker entry

AI's impacts will unfold over decades, like electricity, as institutions adapt.

Tracker entry

Existential-risk probabilities "lack meaningful epistemic foundations".

Tracker entry

Deep learning is "hitting a wall"; scaling laws are "mere observations that might not hold forever".

Tracker entry

Worrying about evil killer AI is like worrying about "overpopulation on Mars".

Tracker entry

Large language models as "stochastic parrots": risks of size, cost, bias and documentation.

Tracker entry

"Within thirty years, we will have the technological means to create superhuman intelligence."

Tracker entry

"I'll be surprised if this event occurs before 2005 or after 2030."

Tracker entry

"By the end of the century … we will still be wondering if the singularity is near."

Tracker entry

The length of tasks AI agents can complete doubles roughly every 7 months.

Tracker entry

AI can produce science recognised at the highest level.

Tracker entry

AGI "in the next five to ten years".

Tracker entry

The end of disease: "within reach. Maybe within the next decade or so".

Tracker entry

"The cost to use a given level of AI falls about 10x every 12 months."

Tracker entry

Generative AI could raise global GDP by about 7% over ten years.

Tracker entry

AI will raise total factor productivity by no more than 0.66% over ten years.

Tracker entry

Masdar/EWEC round-the-clock project (5.2GW PV + 19GWh storage, 1GW continuous) operational in 2027; "world-first" (Masdar).

Tracker entry

Employment of 22–25-year-olds in highly AI-exposed occupations is falling relative to less-exposed peers.

Tracker entry

AI lowers the barrier to sophisticated, largely autonomous cyberattacks.

Tracker entry

Frontier models may give meaningful bio/chem uplift to novices; developers added safeguards because they could not exclude it.

Tracker entry

EU AI Act obligations for general-purpose AI models apply from 2 Aug 2025.

Tracker entry

EU AI Act high-risk rules to apply from 2 Aug 2026.

Tracker entry

EU high-risk (Annex III) rules apply from 2 Dec 2027; Annex I from 2 Aug 2028.

Tracker entry

UN to establish an Independent International Scientific Panel on AI (40 experts) and a Global Dialogue on AI Governance.

Tracker entry

No evidence that AI-enabled disinformation changed UK, French or EU 2024 election results; wider harms documented.

Tracker entry

Misinformation/disinformation ranked 2nd most severe global risk (2-year); adverse outcomes of AI 5th (10-year).

Tracker entry

Frontier AI is concentrated and becoming less transparent: industry produced 90%+ of notable models in 2025; transparency index average fell from 58 to 40.

Tracker entry

"Powerful AI" (smarter than a Nobel laureate across most fields, able to act autonomously for days) "could come as early as 2026, though there are also ways it could take much longer."

Tracker entry

After powerful AI arrives, AI could compress "the next 50-100 years of biological progress" into 5–10 years.

Tracker entry

"There's absolutely no way … that autoregressive LLMs, the type that we know today, will reach human intelligence." Scaling "is saturating".

Tracker entry

Aggregate forecast: 50% chance that unaided machines outperform humans at every task by 2047; 10% by 2027.

Tracker entry

Full automation of all occupations at 50% probability around 2116.

Tracker entry

AI progress to 2030 could plateau, continue or accelerate. Capabilities are "jagged".

Tracker entry

Superintelligence: "any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest".

Tracker entry

Human-level AI / AGI by 2029; the Singularity (a "millionfold" expansion of intelligence) by 2045.