AGI Timelines and Their Changes Over the Last 5 Years

Source:https://chatgpt.com/share/67a291eb-7d7c-8007-a75b-109b6084bf9f

Understood! I will compile a comprehensive research report on AGI timelines and their changes over the last five years. The report will include:

  • Definitions and context around AGI
  • A historical overview of how AGI timeline predictions have evolved
  • A summary table listing influential researchers and CEOs, their AGI estimates, and how their views have shifted over time
  • Specific quotes, caveats, and explanations for why their estimates changed
  • A well-structured format ensuring clarity

I will source reputable and authoritative voices in AI research and technology leadership. I'll let you know when the research is complete!

Introduction

Artificial General Intelligence (AGI) refers to a level of artificial intelligence that matches or surpasses human cognitive capabilities across virtually all tasks, rather than being specialized or “narrow” in one domain

www.businessinsider.com. In other words, an AGI would be an AI system “that is generally smarter than humans” across the boardwww.businessinsider.com. The quest for AGI is significant because such a system could revolutionize technology, economy, and society – enabling breakthroughs in science, automating complex labor, and possibly even exhibiting forms of consciousness or agency. AGI timelines – predictions of when human-level (or greater) AI might be achieved – matter greatly. They influence how researchers prioritize projects, how companies invest in AI development, how policymakers prepare regulations, and how society at large gauges the urgency of AI’s opportunities and risks. Overly optimistic timelines can fuel hype (and perhaps underplay long-term safety), whereas overly skeptical timelines might lead to complacency in research or preparedness. Thus, tracking expert predictions and how they change over time provides insight into the evolving state of AI progress and the balance of optimism vs. caution in the field.

Historical Context (Last Five Years of AGI Predictions)

Five years ago, the consensus among many experts leaned toward AGI being a distant prospect – often decades away. Around 2018–2020, numerous prominent AI researchers suggested that human-level AI was not on the immediate horizon. For example, in 2018 a survey of AI experts put the median estimate for achieving “High-Level Machine Intelligence” around 2060

www.lesswrong.com. Pioneers like Andrew Ng frequently dismissed near-term AGI concerns; he quipped that worrying about evil killer AI is like worrying about overpopulation on Mars – implying AGI was hundreds of years away, a remote future problemquoteinvestigator.com. Similarly, robotics legend Rodney Brooks in 2018 predicted no sudden emergence of general intelligence, insisting progress would remain in “point solutions for a long, long time”rodneybrooks.comand even betting that truly human-like cognition in machines would not arrive by the 2040srodneybrooks.com.

However, the period from 2020 to 2023 saw rapid AI advances – most notably large language models like OpenAI’s GPT-3 and GPT-4 – which led many experts to update their timelines toward earlier dates. By 2023, some AI lab leaders began suggesting AGI might be just years, not decades, away. Surveys reflect this shift: between 2022 and 2023 alone, the aggregate expert prediction for a 50% chance of achieving human-level AI moved 13 years closer (from around 2060 to roughly 2047)

www.lesswrong.com. In 2023, the median forecasts in some polls jumped into the mid-21st century or sooner, whereas earlier surveys had hovered around the 2050s or 2060s.

This sharpening of timelines has been anything but uniform – opinions remain deeply divided. On one hand, optimists (including certain CEOs and futurists) argue recent breakthroughs are signs that AGI could be imminent, possibly within this decade or even the next few years. On the other hand, skeptics (including several founding figures of AI) maintain that fundamental obstacles remain and that human-level AI could still be far off, absent further major innovations. Notably, a number of thought leaders who once viewed AGI as remote have significantly revised their estimates earlier after witnessing the rapid progress of AI in the last five years. Meanwhile, others have held consistently to longer timelines, urging caution against hype.

Below, we summarize the timeline predictions of several influential AI researchers, CEOs, and thinkers – highlighting their past estimates versus their more recent statements – and examine how and why these views have changed. This sets the stage for a deeper analysis of the contrasting perspectives (the optimistic vs. the skeptical) on when the world might see true AGI.

Summary of AGI Timeline Predictions (2018–2024)

Figure

Role/Background

Earlier Prediction (circa 2018–2020)

Recent Prediction (2023–2024)

Sam Altman

OpenAI CEO

c. 2019–2020: Believed AGI possible within ~10 years (around 2030)nxtli.com(often said it may come faster than most expect).

2023–2025: Extremely bullish – “confident we know how to build AGI” and expecting first true AGI systems by 2025blog.samaltman.com. In late 2023, stated AGI will arrive “sooner than most people think”www.businessinsider.com.

Demis Hassabis

DeepMind (Google) CEO

c. 2018: Consistently cautious. Estimated “at least a decade” away (2030s timeframe) and needing multiple breakthroughscybernews.com.

2023: Remains cautious. Reiterated AGI is 5–10 years away – “maybe within a decade”cybernews.com– still requiring 2–3 major innovations (e.g. better memory, planning agents)cybernews.com.

Ray Kurzweil

Futurist, Google engineer

2029 – famously predicted in the 2000s that AGI (human-level AI) would arrive by 2029; maintained this timeline in 2017www.forbes.com.

2022–2023: Unchanged. Still targeting 2029 for human-level AI. (In recent interviews he “sticks with 2029,” though he allows it could happen slightly sooner)www.reddit.comwww.reddit.com.

Yann LeCun

Meta Chief AI Scientist, Turing Award laureate

c. 2019: Skeptical of short-term AGI; often emphasized current AI lacks common sense and is “not even at the level of a cat,” implying decades of work needed (no specific date given).

2023–2024: More optimistic than before. Said his timeline “is not very different” from Altman’s or Hassabis’s – “quite possibly within a decade” for human-level AIofficechai.com. Believes 5–10 years is possible “if everything goes great” (though he cautions that is an ideal scenario)officechai.com.

Geoffrey Hinton

Turing Award laureate (“Godfather of Deep Learning”)

c. 2016–2018: Believed AGI was 30 to 50 years or more awayen.wikipedia.org.

2023: Dramatically shortened. “Until quite recently I thought 20–50 years… now I think it may be 20 years or less,” Hinton said in early 2023www.foxnews.com. (Leaving Google in 2023, he voiced concern AGI could be sooner than expected.)

Yoshua Bengio

Turing Award laureate, AI researcher

c. 2020: Assumed AGI was very distant – “decades to centuries” away (focus was on narrow AI).

2023: Significantly revised after GPT-4. Acknowledged his earlier view was wrong and now estimates 5 to 20 years for human-level AI with 90% confidenceyoshuabengio.org. Even considers the possibility it could be just “a few years” away in a worst-case scenarioyoshuabengio.org.

Andrew Ng

AI pioneer, former Google Brain lead

c. 2018: Dismissed AGI as a near-term concern. Famously said worrying about superintelligent AI now is like worrying about overpopulation on Mars – i.e. hundreds of years too earlyquoteinvestigator.com.

2023: Largely unchanged skepticism. Ng still emphasizes current AI is far from true understanding, and he focuses on practical AI. (He hasn’t provided a specific year publicly, but continues to imply AGI is not imminent, aligning with his earlier “decades away” stance.)futureoflife.orgalldus.com

Elon Musk

Tech CEO (xAI, Tesla, etc.)

c. 2016–2019: Warned AGI could be sooner than expected; suggested 2025 as a plausible arrival date on multiple occasions (though viewed as optimistic).

2022–2023: Still highly optimistic. Stated an AGI “smarter than the smartest human” will likely be available by 2025 or 2026cybernews.com. (Notorious for aggressive timelines, but consistent that mid-2020s are likely for AGI.)

Gary Marcus

Cognitive scientist, AI critic

c. 2019: Very skeptical – argued deep learning alone can’t achieve AGI. Expected no human-level AI for multiple decades absent new paradigms (no exact date, but clearly not by the 2020s).

2022–2023: Remains skeptical. Published “AGI is Not Nigh” (2022) and even bet that by 2029 AI won’t master certain human-level tasksmanifold.markets. Believes current systems (LLMs) lack true understanding and that no imminent AGI is in sight barring major new breakthroughs.

Rodney Brooks

Roboticist, AI pioneer

c. 2018: Asserted no sudden AGI will occur; “not at all there” in terms of human-like cognition. Predicted a robot with common sense of a child was “Not In My Lifetime” (far beyond 2040s)rodneybrooks.comrodneybrooks.com.

2023: Unchanged in pessimism. In his 2023 scorecard, Brooks noted we’re still “<1% of the way” to human-level intelligence and ridiculed the idea that an all-encompassing general AI is nearrodneybrooks.com. Still expects mid-century or later for any AI with genuine human-like understanding (and perhaps even that is optimistic).

Ben Goertzel

AI researcher, CEO SingularityNET

c. 2017: Predicted relatively early AGI (he coined “AGI” term). In 2017 he speculated a 2020s timeline – e.g. ~2029 for human-level AI.

2023: Consistently optimistic. Reiterated expectation that AGI will likely emerge by the end of 2020s. (In one interview, Goertzel confidently projected “humanity would develop AGI by 2029”www.techradar.com.)

Table: Influential AI figures and how their AGI timeline predictions have shifted (or held) over roughly the last five years. Sources for these estimates are provided in the analysis below.

Detailed Analysis of Predictions and Shifts

In this section, we delve deeper into each figure’s viewpoint, providing direct quotes and context that illuminate why their AGI timeline predictions changed (or why they steadfastly did not). We also contrast the optimistic expectations versus the skeptical ones, exploring the reasoning and caveats behind each.

Sam Altman – From Cautious Optimism to Bold Certainty

Sam Altman (CEO of OpenAI) has become one of the most bullish public voices on AGI timelines in recent years. A few years ago, Altman already believed AGI was on the horizon, but his timeframe has dramatically tightened. Around 2019, Altman was hinting that AGI might be achieved in roughly a decade. In conversations back then, he suggested timelines “as early as 10 years” for reaching a powerful “Level 5” AI system

nxtli.com. This would have put AGI around 2030. Indeed, Altman’s general stance was that it could happen sooner than many expected – he once remarked that most people underestimate how quickly AI could progress.

By 2023-2024, after OpenAI’s successes with GPT-3 and GPT-4, Altman’s public statements became even more aggressive. In late 2023, at The New York Times DealBook Summit, Altman stated: “we’ll achieve AGI sooner than most people in the world think”

www.businessinsider.com. Perhaps more tellingly, he added that it “will matter much less” than people assumewww.businessinsider.com– a nod to his view that AGI might integrate gradually into society rather than appear as a sudden, apocalyptic “singularity.”

In early 2025, Altman went a step further by essentially declaring that the initial form of AGI is imminent. He wrote in a January 2025 blog post: “We are now confident we know how to build AGI as we have traditionally understood it.”

www.businessinsider.comblog.samaltman.com. He elaborated that 2025 could already see the first AI agents “join the workforce” in a meaningful wayblog.samaltman.com– implying that true generally capable AI systems are about to emerge. “In 2025, we may see the first AI agents…materially change the output of companies,” Altman predictedblog.samaltman.com.

This represents a significant acceleration from even Altman’s own position a few years prior. The reasoning behind Altman’s shortened timeline is rooted in the empirical progress OpenAI witnessed. Systems like ChatGPT demonstrated surprisingly general capabilities (reasoning, coding, language understanding), leading Altman to believe that scaling up models plus incremental innovations are enough to reach AGI soon. By his account, the core ingredients for AGI are essentially here – it’s now a matter of engineering: “iteratively putting great tools in the hands of people” to reach general intelligence

blog.samaltman.com. That confidence is bolstered by Altman’s inside view at OpenAI, where they presumably see a feasible path to building progressively more powerful models.

Altman does include caveats – he acknowledges it’s “still so early” in understanding and that there’s “so much we don’t know”

blog.samaltman.com. He also frequently discusses safety and alignment challenges that must be solved in parallel. Nonetheless, his timeline for when AGI will arrive has moved into the immediate term. This optimism is tempered by his notion that AGI won’t be a single earth-shattering moment. As he put it, everyone may be surprised how soon it comes, but also surprised by how it’s not a sudden overnight utopia or doom scenario. It’ll augment human work and productivity more gradually than science fiction might suggestwww.businessinsider.com.

Key quotes (Sam Altman):

  • “My guess is we will hit AGI sooner than most people think and it will matter much less [than people imagine].”www.businessinsider.com– (DealBook Summit, Dec 2023)
  • “We are now confident we know how to build AGI… We believe that, in 2025, we may see the first AI agents ‘join the workforce’ and materially change the output of companies.”blog.samaltman.com– (Altman’s blog, Jan 2025)

Altman’s evolving stance encapsulates the growing optimistic camp: those who, witnessing rapid AI progress, have pulled their AGI expectations from “someday in 2030+” to “likely within this very decade (or sooner)”.

Demis Hassabis – Steady as She Goes (Still Some Breakthroughs Needed)

Demis Hassabis, CEO of Google DeepMind, provides a contrast – his timeline has remained relatively consistent over the last five years, leaning cautious. Around 2018–2020, Hassabis often said that AGI was still far off. He emphasized the need for fundamental research and multiple “missing pieces” before general intelligence is achieved. In interviews, he suggested AGI was on the order of a decade or more away. For instance, he mentioned needing “2 or 3 big innovations” beyond what existed, implying at least 10+ years of work

cybernews.com.

Fast forward to 2023, and Hassabis is still saying roughly the same thing – albeit acknowledging that we may be a few years closer now. In an October 2023 interview with The Times (UK), Hassabis said “AGI is still at least a decade away”

cybernews.com. He described AGI as an “epochal, defining” achievement for humanity, but one that demands further breakthroughs, such as truly capable agent-based AI systems with better memory and reasoningcybernews.com. Later in 2023, he reiterated to The Wall Street Journal that human-level AI “could be just a few years, maybe within a decade away.”cybernews.com. In other words, his estimate centers around the 2030s timeframe – perhaps somewhere in the 5–10 year range as of 2023. This isn’t a dramatic shift; it’s essentially the timeline he’s hinted at for years, just rolling forward with the calendar.

Hassabis’s reasoning has consistently been grounded in technical realism. As one of the minds behind breakthroughs like DeepMind’s AlphaGo and AlphaFold, he is optimistic about AI’s potential, but he also directly confronts the current limitations. He identifies that current AI systems still lack key attributes of human cognition, such as robust generalization, flexible memory, and the ability to plan over long horizons. In 2023 he noted that we need new ideas – for example, “AI agents” that can autonomously explore and learn – before AGI is achieved

cybernews.com. This view naturally leads to a later timeline than those who believe scaling today’s techniques is enough.

It’s worth noting that while Hassabis has maintained a cautious timeline, he has become slightly more specific as progress unfolds. Saying “at least a decade” in 2023 suggests he’s eyeing early-to-mid 2030s. He also sometimes adds optimistic caveats like “a few years, maybe within a decade.” But importantly, Hassabis pushes back against the hype from both extremes: he criticizes “a lot of…crazy hype” – both the alarmist narratives and the over-optimistic ones

cybernews.com. He sees AGI as inevitable but not imminent, and urges steady, careful progress.

In summary, Demis Hassabis’s timeline has not significantly changed in the past five years – he was saying ~10+ years a few years ago, and he’s still saying ~10 years now (meaning the goal is always about a decade out on the current trajectory). This measured stance underscores the skeptical/guarded perspective: even as AI leaps forward, some experts believe we still have a long road and multiple unknowns to resolve before crossing the threshold to true AGI.

Key quotes (Demis Hassabis):

  • “AGI is still at least a decade away… there are still two or three big innovations needed from here until we get to AGI.”cybernews.com– (Interview in The Times, Oct 2023)
  • “[AGI] could be just a few years, maybe within a decade away.”cybernews.com– (Interview in WSJ, 2023, indicating his best-case range, with caution)

Hassabis’s consistency provides a baseline in the debate: even as milestones like GPT-4 emerged, he did not dramatically shorten his estimate. This contrasts with the next category of experts – those whose timelines did shift notably due to recent progress.

Geoffrey Hinton & Yoshua Bengio – From Long-Term Skepticism to Shorter-Term Concern

It is particularly striking when founding fathers of the field change their tune. Geoffrey Hinton and Yoshua Bengio, two deep learning pioneers (and Turing Award winners), have in the past couple of years revised their AGI timelines earlier, citing the surprisingly rapid advancements in AI capabilities.

Geoffrey Hinton

Up until recently, Hinton was on record expressing that human-level AI was likely many decades away. For example, in the 2010s he would say things like “It’s going to be 20 to 50 years or even longer” before we have general AI comparable to a human

en.wikipedia.org. This aligned with a cautious, academic consensus at the time.

However, in March 2023, Hinton made headlines with a much more urgent tone. In an interview with CBS (shortly before he left Google), Hinton acknowledged the timeline had compressed: “Until quite recently, I thought it was 20 to 50 years… And now I think it may be 20 years or less.”

www.foxnews.com. This quote encapsulates a stark shift. By “quite recently,” he likely meant that the progress of models like GPT-3 and GPT-4 forced him to update his priors. If AGI might be under 20 years away, that implies a rough target of the 2040s or even 2030s in Hinton’s mind – a significant jump forward from saying “2070 or later” as a possibility before.

Hinton’s reasoning is both awe and anxiety at what current AI can do and what might be on the horizon. He reportedly grew concerned that massive neural networks were starting to show glimmers of general abilities, and he famously said in 2023, “It’s not inconceivable” that AI could even pose an existential threat (the “wiping out humanity” scenario) if it continues to surprise us

www.foxnews.com. Such statements were a departure from his earlier stance that AI was too dumb to worry about yet. In short, Hinton saw the writing on the wall with current AI progress and felt compelled to warn that AGI might come much sooner than he’d thought – perhaps prompting his departure from Google to speak more freely on the topic.

Yoshua Bengio

Yoshua Bengio underwent a similar transformation around the same time. Bengio admitted explicitly that he “needed to…radically” change his timeline estimates for AGI

yoshuabengio.org. For most of his career (spanning the 80s, 90s, 2000s), Bengio wasn’t focused on when AGI would happen; like many, he assumed it was far off. As late as the mid-2010s, he, like Hinton and LeCun, likely imagined AGI was many decades away. In early 2020s interviews, Bengio sometimes mentioned centuries as a possibility for when AI might reach human intelligence, essentially putting it so far ahead as to not be an immediate concern.

But by mid-2023, Bengio’s view changed dramatically. In a candid blog post (August 2023), he wrote that seeing the new wave of AI capabilities (like GPT-4’s reasoning improvements) made him update his forecast. He went from “decades to centuries” to now believing “5 to 20 years” is a plausible timeline for human-level AI, with 90% confidence in that range

yoshuabengio.org. In other words, he’s saying there’s only a 10% chance we don’t get AGI by roughly the 2040s. That is a huge shift toward sooner timelines. Bengio even mused: what if it’s just a few years away? – a prospect that clearly unnerved him: “And what if it was, indeed, just a few years?” he added, highlighting that we must consider even that scenarioyoshuabengio.org.

The context around Bengio’s change is his growing involvement in AI safety and policy discussions. He signed the 2023 open letter calling for a pause on giant AI experiments, and he’s spoken about the need for regulation. The speed of AI advancements “started to dawn on [him]” that AGI could come much faster

yoshuabengio.org, and this created a personal ethical crisis (as someone who helped lay the groundwork for these AI techniques). Bengio cited the emergence of “system 2”-like reasoning in GPT-4 (more analytical, general problem solving) as evidence that even without fundamentally new architectures, AI was approaching more general cognitive abilitiesyoshuabengio.org. This, combined with the dual-use risks of AI, led him to substantially shorten his expected timeline and become more vocal about preparing for AGI.

Key quotes (Hinton & Bengio):

  • Hinton: “Until quite recently, I thought it was going to be like 20 to 50 years… And now I think it may be 20 years or less.”www.foxnews.com
  • Bengio: “It started to dawn on me that my previous estimates… needed to be radically changed. Instead of decades to centuries, I now see it as 5 to 20 years with 90% confidence.”yoshuabengio.org

Both Hinton and Bengio essentially moved from the skeptics’ camp toward the optimistic camp (or at least toward the concerned camp that sees AGI on the near horizon). Their credibility as researchers who never hyped AI in the past makes their revised predictions particularly notable. It signals that even some originally conservative experts were surprised by the pace of recent AI progress. This trend – experts updating toward shorter timelines – is a theme echoed by many in the AI community around 2022-2023.

Optimists: Futurists and CEOs Betting on the 2020s

While Hinton and Bengio were late converts to shorter timelines, some figures have always worn the optimist label – and remain so, doubling down that AGI is coming very soon (if it’s not “here” in rudimentary form already). This camp includes tech entrepreneurs like Elon Musk, futurists like Ray Kurzweil and Ben Goertzel, and certain AI startup leaders. They provide the optimistic perspective that AGI is likely within this decade (the 2020s) or by around 2030 at the latest.

  • Elon Musk – Musk has long warned about AI risks and has often suggested relatively near-term dates. In 2017, for example, he mentioned 2025 as a potential point by which AI could surpass human intelligence (he worried about Google’s DeepMind reaching AGI first). In a 2022 interview with a Norwegian fund, Musk predicted an AGI that is “smarter than the smartest human” by 2025 or 2026

    cybernews.com. His rationale: the exponential improvements in AI hardware and software – “AI… is the fastest advancing technology I’ve seen… [compute] increasing by 10x every year”cybernews.com. Musk, however, is known to give aggressive timelines for all his projects (from self-driving cars to space travel), many of which turn out too optimistic. Still, he has been consistent in saying the world should brace itself for AGI in the mid-2020s. This optimism is coupled with deep anxiety: Musk also frequently states that unaligned superintelligent AI could be catastrophic (“Terminator future”), hence his push for responsible development (he even founded a new AI company, xAI, to work toward “safer” AGI).

  • Ray Kurzweil – A famed futurist, Kurzweil is notable for how stable his prediction has been. Since at least the early 2000s, Kurzweil has been saying 2029 is the year he expects a true AI with human-level intelligence (and 2045 for the “Singularity” where AI vastly surpasses humans)

    www.reddit.comwww.reddit.com. Criticized by some as overly optimistic, Kurzweil’s timing initially seemed like outlier futurism. But interestingly, as 2029 draws nearer, some in Silicon Valley note that Kurzweil might end up “not too far off actually”www.reddit.comgiven the trajectory. In recent interviews (2022), Kurzweil has stood by 2029, even suggesting that timeline might be “conservative” nowlifearchitect.ai. He said he’ll “stick with 2029 prediction but it can happen before”www.reddit.com. Kurzweil’s unwavering stance provides a benchmark for optimism; to him, the exponential growth of computing and AI capabilities all pointed to late-2020s as the crossover to AGI, and so far nothing has dissuaded him – in fact current progress only reaffirms his faith in that schedule.

  • Ben Goertzel – Goertzel, who leads SingularityNET and chaired the AGI conference series, is another long-time believer that AGI is achievable soon. In 2017, for example, at SXSW he said mid-to-late 2020s could see AGI. He has often mentioned 2029 (perhaps coincidentally matching Kurzweil) as a reasonable target. In a 2022 interview with TechRadar, an AI researcher recalled: “Dr. Ben Goertzel… predicted that humanity would develop AGI by 2029”

    www.techradar.com. Goertzel’s reasoning often comes from a perspective outside the mainstream deep learning – he integrates approaches like symbolic AI and cognitive architectures, and he believes with focused effort an “artificial general intelligence” could emerge within a decade or so of work. His confidence has, if anything, grown after seeing things like GPT-4, which he considers steps in the right direction (though he also argues current AI isn’t true AGI yet). Essentially, Goertzel has been on record for many years anticipating an early breakthrough, and the timeline in his statements hasn’t significantly changed – except that as years pass, he too now frames it as “within a few years.”

These optimists bring direct quotes that exemplify their viewpoint:

  • Kurzweil: “AGI will be achieved by 2029” – a line he’s repeated, and which he affirmed in a 2022 assessmentwww.reddit.com.
  • Musk: “[AGI] will be available in 2025 or by 2026”cybernews.com, and “AI is the fastest advancing tech… many breakthroughs on [an exponential] curve”cybernews.comhighlighting why he sees such a short timeline.

It’s interesting to observe that some of these optimistic predictions, once seen as outliers, have been inching toward mainstream consideration as AI systems get more capable. What was “crazy soon” a decade ago (e.g. 2025 or 2030 for human-level AI) is now at least being taken seriously by many, given current trends.

Of course, the optimistic camp also has its detractors and caveats. Often, critics note that predicting AGI within a decade has a long history of failure (people predicted AI as early as the 1960s and 1970s and were wildly off). The optimists might be repeating that mistake, blinded by recent hype. In response, today’s optimists argue “this time is different” because AI systems are demonstrably more general than ever before, and scaling is a viable path.

The tension between these optimists and the skeptics (like Gary Marcus or Rodney Brooks) fuels much of the debate in AI. Let’s examine the skeptical side more closely.

Skeptics: Why Some Still Think AGI Is Far Away

In contrast to the entrepreneurs and recently-converted worriers, a number of respected voices maintain that AGI remains a far-off goal – beyond the 2030s, perhaps many decades out. Their predictions have either not budged in the face of recent progress, or have shifted only modestly. They tend to emphasize the gaps and flaws in current AI, arguing that no matter how impressive chatbots and games-playing AIs are, they still lack fundamental aspects of general intelligence (such as true understanding of the physical world, common sense reasoning, causal reasoning, and so on).

Key representatives of this skeptical camp include Gary Marcus, Rodney Brooks, and (formerly) Andrew Ng, among others like Ernie Davis or older-school AI researchers.

  • Gary Marcus – Marcus has been an outspoken critic of deep learning hype. In 2022, he penned a piece titled “Superhuman AGI is not nigh”, explicitly refuting the idea that general AI is around the corner

    manifold.markets. He even entered into a wager with an OpenAI policy researcher (Miles Brundage) setting specific milestones that current AI must achieve by 2029 – Marcus is betting that the AI will fail to meet those benchmarks, thereby demonstrating we won’t have human-like AI by thenmanifold.marketsmanifold.markets. His stance has essentially been that today’s AI systems are brittle and domain-limited; they can’t truly understand or reason about the world in the open-ended way humans can. Marcus often points to examples of chatbots making absurd errors as evidence that we’re missing something big. Thus, he might concede eventual AGI is possible, but insist it could be many decades unless we discover new techniques. He hasn’t given a precise year publicly (perhaps to avoid the same trap he accuses optimists of), but phrases like “not nigh” and his 2029 bet (that AGI-level tasks won’t be solved by then) indicate he expects no AGI in the 2020s and likely not in the 2030s either without a revolution in approach. Marcus’s reasoning: purely data-driven neural nets lack an understanding of the world’s abstractions; until AI has a way to ground knowledge, integrate symbolic reasoning, or something fundamentally new, it won’t achieve general intelligencenews.ycombinator.com. So unlike Hinton/Bengio, Marcus did not update toward optimism after GPT-4 – in fact, he doubled down, saying essentially “it’s still just autocomplete; scaling won’t get us to real AGI”news.ycombinator.com.

  • Rodney Brooks – Known for both achievements in AI/robotics and for debunking AI hype cycles, Brooks has made concrete predictions that no, we would not have human-like AI any time soon. In 2018, he published a list of predictions with target dates. Notably, he gave a prediction that a truly human-like intelligent robot (with the understanding of a young child) was “NIML – Not In My Lifetime.”

    rodneybrooks.com(Brooks was in his early 70s then, so he effectively meant “not by mid-21st century”). He wrote “many think we are already [near AGI]; I say we are not at all there”rodneybrooks.com. Five years later, in his 2023 update, Brooks remained unmoved: he remarked that despite the rise of ChatGPT, we are still nowhere near full human-level AI. He observed that while people are amazed by chatbots, these systems still “confabulate” and lack true understanding – essentially performing impressive mimicry. Brooks asserts “there is not going to be a [single] general intelligence that can suddenly do all sorts of things… It is going to be point solutions for a long, long time to come.”rodneybrooks.com. He even quantified provocatively that we might be “less than 1% of the way” to human-level AIrodneybrooks.com. So Brooks would place AGI well beyond 2040, likely post-2050. If pressed, he might say late 21st century or that it’s so uncertain we shouldn’t name a date. His unchanged skepticism through 2023 highlights that some experts believe the hard problems (like embodied intelligence, self-awareness, common sense) haven’t been solved by the current deep learning paradigm, and until they are, AGI is out of reach.

  • Andrew Ng – While not as public in timeline betting, Ng’s oft-quoted analogy about Mars overpopulation

    quoteinvestigator.comsymbolized a broader sentiment in the late 2010s: focus on near-term AI, because general AI is too far to meaningfully discuss. As of 2023, Ng still advises concentrating on “data-centric AI” for real-world applications rather than worrying about AGI. He acknowledges the improvements in models but doesn’t see them as truly understanding or reasoning; they’re powerful tools, not nearly human-level entities. In effect, Ng likely still believes AGI is decades away. (He has said things like, “I don’t know what will happen 5 years from now… hundreds of years from now maybe a computer could turn evil”quoteinvestigator.com, implying he’s not expecting it in the foreseeable future.)

These skeptics ground their predictions in the pitfalls of current AI. Key arguments they raise include:

  • Lack of true understanding or grounding: AI can use language but doesn’t mean things the way humans do (Marcus often shows GPT-3 failing basic common sense to illustrate this). Until AI can reliably understand contexts like a human, it’s not AGI.
  • Specialized vs general: Achieving superhuman performance in narrow tasks (chess, Go, protein folding) doesn’t directly translate to general cognition. Brooks emphasizes the brittleness of AI – it excels in fixed domains but can’t transfer knowledge well or deal with unforeseen situations like a human toddler can.
  • The last mile problem: Even if AI gets to 90% of human ability, the last 10% – which includes self-awareness, independent learning of new concepts, and creativity – might prove vastly harder and take far longer than optimistic forecasts assume.
  • Historical precedent of AI booms: Both Marcus and Brooks cite the cycles of hype (e.g., the 1960s perceptron hype, the 1980s expert systems hype) that over-promised and under-delivered on general AI. They urge caution that we might be in another hype upswing that overlooks fundamental limits.

Key quotes (Marcus & Brooks):

  • Marcus: “There could be a race of killer robots in the far future, but…I don’t worry about [AGI]… same reason I don’t worry about overpopulation on Mars.” (2015)quoteinvestigator.com; and in 2022, “No single [AI] will solve more than 4 of [a set of] tasks by 2027” (his public wager – effectively betting against near-term AGI)www.metaculus.com.
  • Brooks: “Many think we are [near AGI]; I say we are not at all there.”rodneybrooks.com; “Building human-level intelligence… is really, really hard… In reality we are less than 1% of the way there.”rodneybrooks.com.

In essence, the skeptical camp’s timeline might put AGI in 2040, 2050, or well beyond – or even cast doubt on whether current approaches will ever get us to full AGI without new paradigms. They serve as a counterweight to the hype, ensuring that the discussion of timelines includes the possibility that today’s excitement could be premature.

Bridging the Perspectives: Why Such Divergence?

It’s noteworthy that in 2023–2024, the range of credible AGI timelines from experts spans from just a few years to many decades. This divergence stems from different interpretations of recent progress and different philosophies on what intelligence entails:

  • Those shortening timelines (Altman, Hinton, Bengio, etc.) often cite empirical evidence: large models learning skills unforeseen by their programmers (emergent behavior) suggests perhaps a scalable path to general intelligence. Each new model (GPT-2 → GPT-3 → GPT-4) has shown qualitative leaps, so extrapolating that trend underpins their optimism. They also worry about being caught unprepared if AGI indeed arrives soon, hence taking nearer timelines seriously. Another factor: the investment and competition in AI is unprecedented (big tech and many startups pouring billions into AI); this acceleration of resources could bring AGI faster than an academic linear trajectory would imply.

  • Those holding longer timelines (Marcus, Brooks, Hassabis to an extent) emphasize the unknowns: we still don’t know how to achieve attributes like true reasoning, common sense, or consciousness in AI. They argue current AI is still fundamentally lacking in areas that humans find simple, and that we may hit diminishing returns from just scaling up language models. In their view, without new conceptual breakthroughs (akin to going from alchemy to chemistry, so to speak), we won’t get to AGI. So their estimates remain farther out, awaiting those new ideas.

An interesting middle-ground perspective comes from people like Yann LeCun (Meta’s chief AI scientist). He has criticized both excessive short-term hype and excessive doom. LeCun historically said we need new architectures (for example, AI that can learn models of the world like animals do) and that just scaling up GPT won’t directly yield AGI. This made his timeline sound long. But in late 2024, even LeCun said he doesn’t think his view is very different from Altman’s or Hassabis’s – possibly within a decade for AGI if all goes well

officechai.com. “It’s not going to happen next year or in two years… [but] quite possibly within a decade,” LeCun clarifiedofficechai.com. He added the caveat that this would require new architectures and learning paradigms to be developed and no major roadblocks – implying that in a favorable scenario we could have AGI by ~2030, but if things don’t go perfectly, it could take longerofficechai.com. This nuanced stance bridges optimism and caution: yes, AGI is attainable maybe in 5–10 years, but only with significant innovation (which he himself is working on).

Thus, the difference in timelines often comes down to whether one believes current progress is on a direct path to AGI or not. Optimists say yes, just go bigger and refine alignment; pessimists say no, we’re missing pieces and might plateau. Both sides bring valid arguments, which is why we see even experts shifting sides as new data emerges.

Key Takeaways and Trends

  • General Trend: Timelines Shortening (for many) – Over the last five years, there’s been a noticeable convergence toward earlier AGI predictions among a significant subset of experts. Breakthroughs such as GPT-3 (2020) and GPT-4 (2023) led many who once said “2050+” to consider dates like 2030 or 2040 or even sooner. Surveys reflect this, with the median expert forecast moving up by over a decade from 2022 to 2023

    www.lesswrong.com. The pace of progress has injected a sense of urgency in discussions that was largely absent in the late 2010s.

  • Optimistic vs. Skeptical Split – There is still a healthy split in opinion. Optimists (often lab CEOs or futurists) now talk about single-digit years until AGI. They point to the rapid scaling of AI models and emergent abilities as evidence that we might wake up with human-level AI by, say, 2028. On the other hand, skeptics (often veteran academics or those focused on AI’s fundamental limits) caution that we might be decades away and that current AI is fundamentally narrow or brittle. They often use analogies like “no one in the 1800s could predict when a fusion reactor would be built” – i.e., the timeline is fundamentally uncertain and likely long. This divergence means there is no single consensus in the AI community; instead, there’s a spectrum from “AGI is imminent” to “AGI is not in sight yet.”

  • Recent Conversions Add Weight to Warnings – The fact that figures like Hinton and Bengio changed their minds toward shorter timelines is a major development. It suggests that recent AI progress crossed a threshold of impressiveness that even deeply knowledgeable people found surprising. Their updated views lend credibility to the notion that we should take the near-term AGI scenario seriously. In other words, even if one remains skeptical, it’s notable that some previous skeptics are now sounding alarms that AGI (and associated risks) might be closer than they thought

    www.foxnews.comyoshuabengio.org.

  • Definitions and Semantics – Part of the debate hinges on what counts as AGI. Some optimistic statements might define AGI as “AI that can do most economically relevant tasks as well as a human” – a relatively bounded definition. Others think of AGI as “AI with consciousness and full human-level autonomous cognitive ability.” These are different bars. For instance, OpenAI and Microsoft reportedly define achieving AGI in operational terms (an AI that can generate $100B in value is one metric)

    cybernews.com. If one uses a lower bar, one might claim we’re close or even partially there. If one uses a higher bar (including qualities like self-awareness), then timelines might be longer. This semantic nuance means two experts might appear to disagree on timeline but actually envision different endpoints. It’s important to note the context and caveats each person gives – e.g., Altman’s “AGI” might mean something slightly different from Hassabis’s stricter definition.

  • Consensus on “This Century” – One area of broad agreement: most experts on either side now believe it is likely that AGI will be achieved at some point in this century (barring global catastrophes). In 2015, some might have said “maybe never” or centuries away. By 2025, relatively few credible AI researchers say “AGI will never happen.” The debate has shifted to when and how. Even the skeptics like Marcus don’t claim AGI is impossible – Marcus writes about hybrid approaches that could eventually yield general AI, just not as fast as hype suggests. So the discourse has matured: we’ve moved from “if” to mostly “when,” and “when” ranges from a handful of years to many decades, but rarely “never” anymore. This in itself is telling – it implies a collective intuition that AGI is a matter of time and progress.

  • Implications of Timeline Differences – These predictions aren’t just academic; they have real implications. Optimistic timelines raise pressing questions about AI safety and governance right now. If AGI could be, say, 5 years away, the world has very little time to prepare policies, guardrails, and align such a system with human values. This is why Altman, Musk, Hinton, Bengio (despite varying timelines) all agree on urging more AI safety research and some form of regulation now. Conversely, if AGI is 30+ years away, one might prioritize current issues (like AI bias, or using AI for climate modeling, etc.) and not divert too much focus to sci-fi scenarios. Thus, how one perceives the timeline can influence one’s priorities in research and policy. It’s notable that even some who say “it’s far” (like Hassabis) still advocate working on safety early – because whenever it comes, it will be “epochal.” So, a bit of a precautionary principle is at play.

  • Convergence on mid-2030s? A possible point of emerging semi-consensus could be the early-to-mid 2030s as a median expectation for human-level AI, with huge uncertainty bounds. For example, when pressed, Altman has said he’d be surprised if it took until 2040 (implying he expects sooner); Hassabis would be surprised if it happened before 2030 (implying later). That places a lot of probability mass roughly 2030-2040 among many experts. Surveys back this: the 2023 expert survey gave 50% chance by 2047

    www.lesswrong.com, which roughly averages the optimists and pessimists. In conversation, one often hears “within 10-20 years” as a safest guess, which lands around 2035. So while the extremes get attention (5 years vs 50 years), if we average out, many think somewhere in the 2030s is plausible. That said, there remains a non-trivial chance assigned to much sooner or much later by different parties – reflecting genuine uncertainty in a fast-moving field.

In conclusion, the landscape of AGI timeline predictions from 2018 to 2024 has shifted toward earlier estimates overall, yet it remains highly varied. Many tech leaders are openly optimistic that AGI will arrive by the late-2020s, citing recent AI leaps. In parallel, a number of esteemed researchers urge patience and caution, insisting that AGI might still be a multi-decade challenge and warning against believing the hype too quickly. This dynamic, evolving discussion underscores both the astonishing progress in AI and the persistent complexity of intelligence itself.

Whether AGI comes in 5 years or 50, nearly everyone agrees it will be a watershed for humanity – which is why nailing the timeline (to the extent possible) and preparing for its consequences are such important endeavors. As AGI predictions become less like sci-fi and more like concrete near-future forecasts, the world is paying much closer attention to what AI’s top minds have to say – and as we’ve seen, even those minds sometimes change their projections as the technology races ahead.

Sources and References

  • OpenAI’s definition of “AGI” (generally smarter than humans)www.businessinsider.comand focus on developing it (Sam Altman interview, Business Insider).
  • TechRadar – “Sam Altman predicts artificial superintelligence (AGI) will happen this year” (Jan 13, 2025) – Discusses Altman’s 2025 prediction and compares others like Goertzelwww.techradar.comwww.techradar.com.
  • Sam Altman’s personal blog – “Reflections” (Jan 2025) – Altman’s own words on knowing how to build AGI and expecting workforce AI agents in 2025blog.samaltman.com.
  • Business Insider – Altman’s DealBook Summit quote about AGI sooner than expectedwww.businessinsider.com.
  • Cybernews – “Tech leaders on AGI: when will it change the world?” (Jan 2024) – Compilation of quotes: Elon Musk’s 2025/26 predictioncybernews.com; Demis Hassabis’s “at least a decade” stancecybernews.comand “few years, maybe within a decade” updatecybernews.com; Yann LeCun’s “quite possible within a decade” quotecybernews.com.
  • The New York Times / WSJ interviews (referenced via Cybernews) – Hassabis on needing breakthroughs for AGIcybernews.com.
  • Fox News (via CBS News) – Geoffrey Hinton’s March 2023 interview quote: 20–50 years down to 20 or lesswww.foxnews.com.
  • Yoshua Bengio’s blog – “Personal and Psychological Dimensions of AI Researchers Confronting AI Catastrophic Risks” (Aug 2023) – Bengio’s update from “decades/centuries” to “5 to 20 years”yoshuabengio.org.
  • Reddit discussion citing Bengio (with link to his blog) confirming the quoteyoshuabengio.org.
  • Quote Investigator – Tracing Andrew Ng’s “overpopulation on Mars” analogy (originally from 2015)quoteinvestigator.comand Wired interview quote “hundreds of years from now…maybe a computer could turn evil”quoteinvestigator.com.
  • Rodney Brooks’s blog – “My Dated Predictions” (Jan 2018) and 2025 Update – Contains his predictions: “A robot as intelligent as a dog: NET 2048… not at all there”rodneybrooks.com, “a six-year-old’s understanding: NIML”rodneybrooks.com, and commentary about no sudden AGI, 1% of the way there, etc.rodneybrooks.comrodneybrooks.com.
  • Reddit summary of Gary Marcus’s predictions – references to his Substack 2022 predictions for 2029 and bet with Miles Brundagewww.metaculus.com.
  • Manifold Markets post linking to Gary Marcus’s Substack (“Superhuman AGI is Not Nigh”) and the Marcus-Brundage wager specsmanifold.marketsmanifold.markets.
  • Interview with Ray Kurzweil (via Reddit discussion) – confirms Kurzweil sticking to 2029 for AGIwww.reddit.comand considering it possibly conservative.
  • OfficeChai – “AGI Possible In 5-10 Years: Yann LeCun” (Nov 28, 2024) – LeCun quotes: “No, I don’t think it’s that far away… not very different from Altman or Hassabis… quite possibly within a decade.”officechai.comand “5 to 10 years if everything goes great… almost certainly not [so smooth]”officechai.com.
  • LessWrong post – “Timelines to Transformative AI: an investigation” (2023) – gives data on expert surveys shifting: 2022 survey median 2060 vs 2023 survey median 2047 (50% probability)www.lesswrong.comwww.lesswrong.comand 10% chance by 2029www.lesswrong.com.
  • MIT Technology Review / Time / etc., as secondary sources for general context (some referenced through quotes in other articles).

(All sources above have been referenced inline in the report with the format 【citation†】 for verification.)