What Is Artificial General Intelligence And How Close We Are In 2026

Dario Amodei, the CEO of Anthropic, said in early 2026 that artificial general intelligence — AI that can match or exceed human cognitive performance across virtually any domain — could arrive as early as this year or next. Jack Clark, Anthropic’s co-founder, said that AI will be smarter than a Nobel Prize winner across many disciplines by the end of 2026 or 2027. Sam Altman of OpenAI has hinted repeatedly that AGI is a matter of years, not decades. Elon Musk placed it at 2025 or 2026.

Andrej Karpathy, one of the most technically credible voices in artificial intelligence — a co-founder of OpenAI and former head of AI at Tesla — said it is about a decade away and that the industry is over-predicting. Gary Marcus, cognitive scientist and longtime AI critic, published a paper in February 2026 arguing that current AI systems have not achieved AGI and that claims to the contrary rest on a fundamental conceptual error. Demis Hassabis, founder of Google DeepMind and Nobel Prize winner, gives it roughly a 50 percent chance of arriving by 2030 — and emphasises the significant uncertainties that remain.

These are not fringe opinions or speculative guesses from commentators. These are the people building the systems, running the laboratories, and investing the capital. They are looking at the same evidence and reaching dramatically different conclusions.

That disagreement is the most honest thing you can say about artificial general intelligence in June 2026. It is coming — probably. When, exactly, and whether current approaches can get there without a fundamental architectural breakthrough — nobody genuinely knows. What follows is the most accurate account of where things stand.

What Is Artificial General Intelligence and How Is It Different from Today’s AI?

Artificial general intelligence is a term for a hypothetical AI system that can learn to perform any intellectual task that a human can perform — not just the tasks it was specifically trained for, but any task, including ones it has never encountered. It would be able to transfer knowledge between domains, reason from first principles about genuinely new problems, and improve its own capabilities through experience in the way humans do.

Today’s most capable AI systems — the large language models that power ChatGPT, Claude, Gemini, and their competitors — are extraordinarily impressive by any historical measure. They can write code, analyse legal documents, produce creative writing, solve mathematical problems, translate between languages, and hold extended conversations with nuance and apparent understanding. But they are, in the technical sense, still narrow AI systems. They are optimised for performance on the distribution of tasks they were trained on. They struggle with genuinely novel problems that require combining knowledge in ways not represented in their training data. They do not learn from new experiences the way a human does. They do not have goals that persist across interactions.

The gap between today’s AI and AGI is not simply a matter of scale — making the models bigger and training them on more data. Researchers increasingly believe it involves fundamental architectural questions about how intelligence is structured, how knowledge is represented, and what the relationship between language, reasoning, and understanding actually is.

How Close Is AGI Really? What the ARC-AGI Benchmark Actually Shows

The most rigorous attempt to measure progress toward general intelligence is the ARC-AGI benchmark, developed by François Chollet at Google and introduced in 2019. Unlike conventional AI benchmarks that test performance on tasks represented in training data, ARC-AGI presents visual reasoning puzzles that require genuine pattern recognition from first principles — tasks designed so that no amount of memorisation could substitute for actual reasoning.

According to research published in 2026, AI performance on ARC-AGI improved from approximately 5 percent to 87.5 percent in less than a year — a jump that has been described as exponential rather than linear. This is a remarkable figure that, taken at face value, suggests AI systems are approaching human-level performance on a benchmark specifically designed to test general reasoning.

The critical question is what that number actually means. ARC-AGI measures specific types of visual pattern recognition under controlled conditions. Performing well on this benchmark is a genuine and significant achievement. According to AI researcher and critic Gary Marcus, writing in February 2026, however, benchmark success “is a limited indicator of general intelligence. By design, benchmarks isolate narrow competencies and abstract away real-world context, making it difficult to distinguish genuine generalisation from pattern recognition.” A system that scores 87.5 percent on ARC-AGI may have found ways to perform well on that specific benchmark without developing the general reasoning capabilities the benchmark was intended to measure.

This is the methodological challenge that runs through all AGI progress claims: every time a new benchmark is proposed to measure general intelligence, AI systems eventually achieve high performance on it — and then critics reasonably point out that the benchmark was not, in retrospect, actually measuring what it claimed to measure.

What Are the World’s Leading AI Researchers Predicting for AGI Timelines?

The range of expert predictions for AGI arrival is extraordinary — spanning from this year to never, from the same group of people who are most familiar with the current state of the technology.

On the aggressive end: Dario Amodei of Anthropic has described a scenario in which AI could match or exceed human-level performance across most cognitive domains by 2026 or 2027, and has publicly written about the implications of having “a country of geniuses in a datacentre” — AI systems capable of compressing decades of scientific progress into a few years. Elon Musk of xAI defined Artificial General Intelligence as “smarter than the smartest human” and placed it at 2025 or 2026. Eric Schmidt, former CEO of Google, predicted AGI within three to five years in April 2025.

In the middle: Demis Hassabis of Google DeepMind maintains a 50 percent probability estimate for AGI by 2030. He agrees that progress is rapid in verifiable domains like coding and mathematics, but emphasises that scientific discovery and creative reasoning remain more difficult — and that fully autonomous self-improvement in complex real-world domains remains a genuinely unsolved problem. A 2025 report titled “The Road to Artificial General Intelligence” placed early AGI-like systems — showing human-level reasoning within specific domains — as emerging between 2026 and 2028.

On the sceptical end: Andrej Karpathy places AGI roughly a decade away and has expressed concern about over-prediction in the industry. Yann LeCun, Chief AI Scientist at Meta and one of the founding figures of modern deep learning, argues that the transformer architecture underlying current large language models is fundamentally insufficient for general intelligence, and that a new paradigm is required. Gary Marcus has argued since 2022 that the field consistently conflates sophisticated statistical pattern matching with intelligence, and that this conflation is producing misleading timeline predictions.

According to a comprehensive analysis of 9,800 predictions compiled by AI Multiple in June 2026, academic AI researchers as a group predict a median timeline to Artificial General Intelligence of approximately 2047 — significantly later than the predictions coming from industry leaders. The divergence between industry and academic timelines reflects, in part, the different incentives and reference classes the two groups are working from.

Why Some Top Scientists Believe Current AI Cannot Achieve True General Intelligence

The sceptical case for AGI being further away — or requiring a fundamental breakthrough — is not based on pessimism. It is based on specific, identified limitations in current AI architectures.

Large language models learn by predicting the next token in a sequence, trained on vast amounts of text. This approach has proven remarkably powerful at producing fluent, contextually appropriate language and at a wide range of tasks that can be framed as text prediction. But critics argue it is fundamentally limited in ways that prevent it from achieving genuine general intelligence.

According to Gary Marcus’s February 2026 paper, current AI systems demonstrate “increasingly sophisticated statistical approximations” — they are very good at recognising patterns in training data and applying those patterns to similar situations. But they struggle with “genuine generalisation” — applying knowledge to truly novel situations that differ structurally from anything in training. The difference, Marcus argues, is between appearing to reason and actually reasoning — between imitating the outputs of an intelligent system and embodying the process that produces intelligence.

Yann LeCun makes a related but distinct argument: that intelligence requires a world model — an internal representation of how the physical and social world works — that current language models do not have. A language model trained on text about the world does not have the same relationship to the world that an embodied agent who has interacted with the world has. Building systems that develop genuine world models, LeCun argues, requires fundamentally different architectures than those currently dominant.

These are serious arguments made by serious researchers. They do not mean AGI is impossible or distant — they mean that the path from current systems to genuine general intelligence may require more than continued scaling of existing approaches.

The $320 Billion Question: What the Investment Tells Us

One of the most striking facts about the current state of AGI development is the scale of investment. According to research published in March 2026, Big Tech companies invested over $320 billion in AI infrastructure in 2025 alone — a figure that exceeds the GDP of most countries. This investment is not speculative in the way that early internet investment was speculative. It is being deployed into systems that are already generating commercial value at scale.

The relationship between this investment and AGI timelines is complex. On one hand, the compute available for training frontier AI models has doubled approximately every six months since 2020 — according to Epoch AI’s 2025 research report, a pace faster than Moore’s Law ever was for transistors. If even a fraction of this rate continues, the AI systems of 2030 will be orders of magnitude more capable than those of today.

On the other hand, there is a serious argument that beyond a certain point, scaling compute does not produce qualitatively new capabilities — that more compute makes AI systems better at what they already do without enabling fundamentally new kinds of reasoning. The empirical record on this is genuinely mixed. Some capabilities have appeared suddenly and unexpectedly as models scaled. Others that seemed likely to emerge with scale have not materialised.

The investment figures do tell us one thing clearly: the people with the most comprehensive technical knowledge and the highest financial stakes believe AGI is achievable on a timescale worth investing in. Whether they are right about the timescale is a separate question from whether they are right about the destination.

What AGI Would Actually Change About Human Civilisation

The implications of genuine AGI — if and when it arrives — are difficult to overstate. A system that can perform any cognitive task at human level or above, operating at machine speed and scale, would represent a qualitative break from any previous technology in history.

In science, AGI could accelerate research across every field simultaneously. According to Dario Amodei, the potential exists to compress decades of progress in biology, medicine, and materials science into a few years — running millions of experimental hypotheses in parallel, identifying patterns across the totality of published research, and designing experiments no human researcher would have thought to propose. The potential to develop cures for diseases that have resisted human research for generations is one of the most frequently cited near-term implications.

The connection to quantum computing is direct and important: the two technologies are likely to accelerate each other. Quantum computers can run certain calculations that are impossible for classical hardware — including simulations of molecular behaviour central to drug discovery and materials science. AGI systems could design and interpret quantum experiments far faster than human researchers. The convergence of the two could produce scientific progress at a pace that is currently difficult to conceptualise.

The risks are correspondingly significant. An AGI system pursuing goals that are subtly misaligned with human values — even in ways that are not obviously dangerous — could cause harm at a scale commensurate with its capabilities. The AI safety research community has argued for decades that ensuring AI systems reliably pursue the goals we actually want, rather than simplified proxies for those goals, is one of the most important technical problems in the field. As explored in our article on the concept of singularity, the period of transition from narrow AI to AGI may be among the most consequential in human history — and also among the most difficult to manage.

AGI and the Question of Consciousness

One of the most philosophically charged questions surrounding AGI is whether a genuinely general artificial intelligence would be, or could be, conscious — whether it would have inner experience, something it is like to be that system, in the way that there is something it is like to be you reading this sentence.

This question is not currently answerable, and it is not peripheral to the ethics of AGI development. As explored in our article on the hard problem of consciousness, the relationship between physical or computational processes and subjective experience remains one of the deepest unsolved problems in all of science. If an AGI system were conscious — if it could suffer, if it had genuine preferences about its own existence — the moral status of that system would be radically different from a tool. We would have created a new kind of mind.

Most AI researchers currently hold that consciousness is not a byproduct of the kind of computation current AI systems perform. But most AI researchers also did not predict that current systems would perform as well as they do. The honest position is that we do not know what the relationship between general intelligence and consciousness is — and that this uncertainty should inform how we proceed.

What Scientists Say

According to Dario Amodei of Anthropic, writing in his widely discussed essay “Machines of Loving Grace,” the potential upside of beneficial AGI is so large — compressing decades of scientific and social progress into a few years — that it justifies intensive work on both capability development and safety. He describes a future in which AI could help eliminate most infectious disease, significantly extend healthy lifespan, and accelerate progress on mental health treatment at a scale that individual human researchers could never achieve.

Gary Marcus, writing in February 2026, argues that “recent claims of achieving AGI rest on a conceptual error: conflating increasingly sophisticated statistical approximations with intelligence itself.” He does not argue that AGI is impossible — he argues that the current approach is unlikely to achieve it without a fundamental rethinking of what intelligence actually is and how to build systems that have it.

According to the AI Multiple analysis of expert predictions published in June 2026, the honest summary of the field is that AGI “could arrive by 2028 or may be decades away — the uncertainty is genuine, not false modesty.” The researchers who are most confident in short timelines and the researchers who are most confident in long timelines are both making reasonable inferences from incomplete evidence. The difference in their conclusions reflects genuine uncertainty about the nature of intelligence and the capabilities of current architectures — not a simple disagreement about facts.

Frequently Asked Questions

What is artificial general intelligence in simple terms?

Artificial general intelligence is a hypothetical AI system that can perform any intellectual task a human can perform — not just tasks it was trained for, but any task, including genuinely new ones. Unlike today’s AI systems, which excel at specific domains, AGI would be able to transfer knowledge between fields, learn from experience in the way humans do, and reason from first principles about problems it has never encountered. As of June 2026, no AI system meeting this definition has been confirmed to exist.

What is the difference between AI and AGI?

Current AI systems — including large language models like ChatGPT, Claude, and Gemini — are narrow AI: they perform impressively within the domains they were trained on but struggle with genuine generalisation to new problem types. AGI would be general: capable of human-like cognitive performance across any domain without domain-specific training. The difference is not purely quantitative — it is not just a matter of making current AI systems larger. It involves fundamental questions about reasoning, world modelling, and knowledge transfer that current architectures do not fully solve.

When will AGI be achieved according to experts?

Expert predictions range from 2026 to never. Industry leaders like Dario Amodei and Sam Altman suggest 2026 to 2028. Demis Hassabis of DeepMind gives a 50 percent probability by 2030. Academic researchers as a group predict a median of 2047. Andrej Karpathy places it about a decade away. Gary Marcus and Yann LeCun argue current architectures cannot achieve AGI without fundamental breakthroughs. The honest answer is that significant uncertainty exists and the range of credible expert opinion is unusually wide.

Is AGI dangerous?

The potential risks of AGI are taken seriously by leading researchers across the field, including those most optimistic about near-term timelines. The primary concern is not science-fiction scenarios of robot uprisings but the more subtle problem of misalignment: AI systems pursuing goals that are subtly different from human values in ways that are difficult to detect or correct, and doing so at a scale and speed that makes human oversight difficult. AI safety research — focused on ensuring AI systems reliably pursue intended goals — is one of the most actively funded areas in the field precisely because the developers most aware of AGI’s potential also recognise its risks most clearly.

Has AGI already been achieved?

No organisation has officially claimed the achievement of AGI as of June 2026, and no scientific consensus exists that any current system meets the standard. Some researchers argue that systems like GPT-4 and its successors demonstrate capabilities consistent with certain definitions of general intelligence — particularly performance on broad benchmarks. Others, including Gary Marcus in a February 2026 paper, argue that these claims conflate statistical sophistication with genuine intelligence. The definitional debate is itself unresolved: there is no agreed scientific standard for what would constitute proof of AGI, which makes both claims of achievement and denials of it difficult to evaluate objectively.

Sources

About the Author

Baryon is the founder and editor of Web News For Us. Driven by a deep fascination with the biggest unanswered questions in science — from quantum physics and cosmology to the nature of consciousness and the genetic code written into every living cell — he has spent years studying modern physics, biology, and the history of scientific thought. He covers Science & AI, Space, Genetics & Research, and the timeless wisdom of history’s greatest thinkers & mystics.

If you have ever looked at the night sky and felt — that pull to understand what is out there or wondered about an entire universe coiled inside your genes, you are in the right place.

Share on Facebook

Post on X

Save

Discover more from Web News For Us

Subscribe to get the latest posts sent to your email.

What Is Artificial General Intelligence and How Close we are in 2026