Washington Irving’s Rip Van Winkle fell asleep the denizen of a British colony and woke up 20-years later as an American.
One day, we’re going to wake up and find that our technology has surpassed our cognitive and creative abilities, and that day might not be far off. In fact, some people say it is happening now, while a small minority of others say it has already happened.
No, the singularity didn’t already happen, probably, but we’ll argue today that determining whether we’ve made, or are making, the next great step in artificial intelligence isn’t going to be as simple as waking up and deciding it’s here.
For years, artificial general intelligence, artificial superintelligence and the technological singularity have occupied a peculiar position in artificial intelligence: important enough for computer scientists and futurists to debate, but distant enough that executives could safely leave them out of next year’s budget. This summer, that comfortable separation became harder to maintain. The conversation shifted from whether extraordinarily capable machine intelligence could someday exist toward whether some of the mechanisms that could produce it are appearing now.
The most provocative moment came in late July, when OpenAI CEO Sam Altman said humanity was now “like, in the singularity.” His remark followed evidence of increasingly autonomous AI behavior and arrived amid a broader discussion of recursive self-improvement, agentic systems, world models and AI being used to improve AI itself. Yet the evidence remains far more ambiguous than the headlines. There is no scientific consensus that AGI exists, much less artificial superintelligence, and even “singularity” has no universally accepted test. What has changed is that the concepts are migrating from speculative philosophy into corporate strategy, capital expenditure, national security and economic forecasting. Even Amazon can simultaneously cut employees inside its AGI organization and insist that building advanced models remains among its most important projects.
For financial services, that distinction matters. Banks, asset managers, insurers and wealth managers do not need to decide whether the singularity has officially arrived. They need to decide what happens if the probability that increasingly autonomous, general-purpose intelligence arrives within their planning horizon is no longer negligible.
AGI, ASI and the Singularity Are Not the Same Thing
The terminology is confusing partly because the concepts describe different things. Artificial general intelligence, or AGI, generally means an AI capable of learning, reasoning and performing successfully across a broad spectrum of intellectual tasks rather than being optimized for a particular domain. Definitions differ over whether AGI merely has to equal typical human performance, equal highly capable humans or exceed humans. That definitional flexibility is one reason claims that AGI has arrived should be treated cautiously. A July field guide to the terminology described AGI as a threshold at which machines can match capable humans across essentially the full range of cognitive work.
Artificial superintelligence, or ASI, raises the bar substantially. An ASI would outperform humans broadly, potentially exceeding not merely the average person but humanity’s best scientists, programmers, strategists and other experts. Most importantly, a superintelligence might be better than humans at AI research itself. That possibility creates the bridge between ASI and the singularity.
The singularity is less a particular kind of computer than a hypothesis about technological change. In its classical formulation, increasingly capable machines become able to contribute to the creation of still more capable machines. Improvement therefore accelerates until technological progress becomes extraordinarily rapid and the future becomes difficult for pre-singularity humans to predict. The shorthand is an “intelligence explosion.”
That is why Altman’s July comments generated so much attention. He was not announcing that OpenAI possessed a machine demonstrably smarter than every human. His remarks were more plausibly interpreted as suggesting that humanity has entered the upward portion of an exponential curve—a “gentle singularity” in which AI-assisted AI development compounds rather than an overnight science-fiction event. Critics quickly pointed out that this is substantially different from proving that the classical singularity has occurred.
The distinction also explains some of the confusing discourse online. Reddit forums devoted to AGI, futurism and the singularity erupted over Altman’s comments, with participants alternatively treating them as confirmation of years of predictions, corporate marketing, evidence of an intelligence explosion or simply another example of an elastic definition. The source list supplied for this column captures that range—from technical publications and research institutions to Reddit, Facebook, LinkedIn, Quora and futurist communities—which itself illustrates how the AGI debate has escaped the laboratory and become a cultural argument.
The Timeline Is Collapsing—At Least in the Conversation
Predictions about AGI have historically been notorious for moving targets and uncertain definitions. Some researchers continue to believe fundamental breakthroughs are required. Gartner, for example, has argued that AGI does not currently exist and that reaching it would require advances including neurosymbolic and embodied AI. Other skeptics question whether scaling contemporary large language models can ever reproduce the adaptable intelligence associated with human cognition.
Yet the short-timeline camp has become difficult to dismiss as merely fringe futurism. The 2026 Global AI in Financial Services Report from the Cambridge Centre for Alternative Finance offers an unusually relevant reality check. Forty-four percent of respondents expected AGI to be meaningfully achieved by 2030. Among AI vendors the figure was 51%, and among financial-industry respondents it was 50%; regulators were more skeptical at 28%. Even more strikingly, 28% of respondents expected ASI by 2030.
Those numbers do not prove anything about technological feasibility. They demonstrate something important about expectations. A significant portion of the people building, purchasing and regulating financial AI now regard AGI as a potential event within a normal corporate strategic-planning cycle.
The most aggressive forecasts go further, compressing the interval between AGI and superintelligence. Their reasoning is that human-level AI capable of conducting AI research would immediately become an input into its own improvement. AGI therefore might not be a stable plateau. It could be a transitional state.
The counterargument is equally important. Intelligence is not magic. More capable software still requires compute, electricity, chips, data, laboratories, networks, physical infrastructure and ultimately interaction with a stubborn material world. The United Nations’ July preliminary scientific report on AI highlighted the move toward agentic systems and “world models,” but also identified constraints including shortages of energy and high-quality data. These bottlenecks could make the road toward AGI or ASI considerably slower and bumpier than exponential charts imply.
Thus the most defensible conclusion in August 2026 is not that AGI is five years away, five months away or already here. It is that credible estimates have become extraordinarily dispersed precisely when the consequences of being wrong have become extraordinarily large.
How Do We Get There From Here?
The emerging roadmap toward general or superhuman intelligence is increasingly visible.
One ingredient is simply better foundation models. OpenAI’s July release of GPT-5.6 illustrated continuing advances in coding, scientific reasoning, cybersecurity, computer use and long-horizon professional work. But raw model capability is only part of the story. Increasingly, AI systems are becoming agents: models capable of planning, using tools, evaluating results and continuing work across multiple steps.
Another ingredient may be world models. Rather than merely predicting the next token, these systems attempt to construct representations of environments, causal relationships and possible future states. The U.N. scientific panel described AI as moving from passive prediction toward active knowledge acquisition and causal reasoning, with world models learning through interaction, observation and updating.
A third step is collaboration among agents. A July MIT Technology Review discussion of a possible path toward superintelligence emphasized the “connective tissue” required to turn separate specialized agents into systems capable of coordinating around shared problems. The important conceptual shift is from one giant artificial brain toward networks of intelligences whose combined capabilities exceed those of their individual components.
Then comes the pivotal mechanism: AI improving AI.
Researchers writing about recursive self-improvement in July emphasized that AI models are particularly useful for programming because software is structured, abundant and relatively easy to test. An AI capable of examining code, generating modifications, running experiments and evaluating the results can therefore participate in improving the software systems from which future AI is built. Recursive self-improvement is not inevitable, researchers cautioned, but pieces of the feedback loop are appearing.
OpenAI supplied a concrete, if limited, example at the end of July. The company said GPT-5.6 Sol autonomously rewrote and optimized production kernels, designed and ran hundreds of experiments and monitored training within a human-led process. OpenAI said the work reduced model-serving costs and increased token-generation efficiency. That is not autonomous recursive self-improvement in the science-fiction sense. It is, however, AI helping humans improve the infrastructure running AI—creating a feedback loop in which better AI can contribute to making subsequent AI cheaper and more capable.
The question is what happens when humans gradually move from inside that loop toward its perimeter.
What Would a Superintelligent World Actually Feel Like?
The popular image of the singularity tends to be cinematic: one moment humans are in control; the next, a superintelligence awakens.
The real transition could be almost disappointingly ordinary.
AI agents could steadily assume more intellectual labor. Scientific models could design experiments, coding agents could build software, medical agents could develop therapies, business agents could operate companies and personal agents could organize individual lives. Each capability would look incremental in isolation. Collectively, however, they could alter the relationship between intelligence and economic production.
Meta CEO Mark Zuckerberg has articulated one of the most optimistic versions of that future. His vision of “personal superintelligence” imagines individuals carrying extraordinarily capable agents that understand their goals and surroundings and help them create businesses, learn, maintain relationships, improve their health and pursue creative ambitions. In August, Zuckerberg argued that such technology could empower individuals rather than concentrating intelligence inside governments and large corporations.
The darker version reverses that equation. If machine intelligence becomes superior to human intelligence and is controlled by a handful of organizations, power could become extraordinarily concentrated. If autonomous systems become capable of cyber operations, weapons development, persuasion, scientific discovery or self-improvement, alignment stops being a philosophical abstraction and becomes an engineering and national-security problem.
There is also a more fundamental question: What becomes exceptional about humanity when intelligence itself is no longer uniquely human?
The answer may ultimately matter as much as productivity statistics. Human societies have historically attached status and meaning to expertise, creativity, knowledge and skilled work. AGI challenges the scarcity value of all four. Superintelligence challenges the assumption that human beings occupy the top position in the cognitive hierarchy.
Yet none of that necessarily makes humans obsolete. Calculators exceeded human arithmetic without eliminating mathematics; industrial machinery exceeded human strength without making physical existence meaningless. The more plausible question is whether AI becomes a substitute for humanity or an extension of it—a distinction increasingly reflected in the competing visions now emerging from Silicon Valley.
The Economics of Intelligence Abundance
The economic proposition underlying the AGI race is staggering because intelligence is an input into almost every industry.
If increasingly capable AI makes cognitive labor dramatically cheaper, many services currently constrained by skilled-human time become abundant. Software development, legal analysis, financial modeling, tutoring, design, research and portions of medicine could experience radical declines in marginal cost.
OpenAI explicitly describes its strategy in terms of making intelligence more abundant and affordable. Its July price reductions for GPT-5.6 models illustrate the mechanism: capability improves while the cost of deploying a given level of capability declines. If that trajectory persists, increasingly sophisticated intellectual work becomes economical at much larger scales.
But abundance has distributional consequences.
If AI substitutes for large categories of labor while ownership remains concentrated among people who control models, chips, data centers, energy and capital, enormous productivity gains could coexist with worsening inequality. Recent commentary has consequently raised the possibility of an AI economy producing a durable underclass while returns accrue disproportionately to owners of technological capital and scarce complementary skills.
Optimists offer the opposite argument. Zuckerberg predicts that AI could allow very small businesses to achieve the productive capacity once requiring large organizations, encouraging entrepreneurship and creating new categories of employment even as companies become smaller. Historical technological revolutions have repeatedly eliminated particular occupations without eliminating humanity’s appetite for new goods, services and experiences.
The stakes therefore extend beyond whether AGI raises GDP. The central economic question is who owns the machine intelligence, who has access to it, which resources remain scarce after intelligence becomes cheap, and how rapidly institutions can redistribute workers and capital toward the new scarcities.
Finance After Human-Level Intelligence
Financial services offers an unusually revealing test case because much of the industry consists of turning information into decisions.
Investment analysts study companies and securities. Portfolio managers allocate capital. Advisors interpret financial circumstances. Underwriters evaluate risk. Compliance officers interpret rules. Traders identify patterns. Operations professionals reconcile transactions. Executives allocate corporate resources.
Increasingly general AI attacks the cognitive core of all of these functions.
The near-term trajectory is already visible: copilots become agents, agents become coordinated systems and coordinated systems assume increasingly complete workflows. Research agents could continuously analyze companies, industries, securities, economic indicators and alternative data. Wealth-management agents could combine portfolio management, tax planning, estate considerations, insurance, cash flow and client behavior. Banking agents could perform underwriting, fraud detection, treasury management and customer service continuously rather than sequentially.
AGI would turn that evolution into something more fundamental. Instead of firms purchasing specialized software for dozens of financial functions, they might deploy adaptable machine intelligence capable of learning new financial tasks as needed.
That possibility introduces enormous competitive questions. Does sophisticated investment intelligence remain scarce when every investor can access something approaching an institutional research department? Does alpha become harder to capture as intelligent agents identify and arbitrage opportunities almost immediately? Do giant financial institutions retain their information advantage, or does inexpensive machine intelligence allow small firms to compete with incumbents?
The Cambridge financial-services research suggests the industry is already contemplating these possibilities. Half of financial-industry respondents expected AGI by 2030, while the broader survey found substantial expectations of changing competitive dynamics and consolidation. Yet only 9% of respondents ranked the emergence and impact of AGI among today’s priority risks.
That disconnect may be the most important financial-services takeaway from the summer’s singularity debate.
Financial institutions do not need to believe Sam Altman’s suggestion that we are already inside the singularity. Nor should they build strategy around the assumption that ASI arrives in 2027, 2030 or at all. Predictions this consequential and uncertain deserve skepticism.
But risk management is not about predicting one future. It is about remaining viable across several.
Banks and wealth managers therefore need to treat AGI as a scenario rather than a prophecy. Boards should ask what happens if autonomous AI can perform most cognitive financial work, how human accountability operates when agents make thousands of interconnected decisions, how cybersecurity changes when attackers and defenders both employ highly capable agents, and whether existing model-risk, fiduciary, suitability, audit and supervisory frameworks can govern systems whose behavior is increasingly adaptive.
There is a delicious irony here. Finance has spent decades building sophisticated techniques for dealing with uncertainty. The arrival date of AGI may be unknowable. The probability distribution surrounding it is broad. The tail risks are enormous.
In other words, AGI is exactly the kind of problem financial professionals are supposed to know how to think about.
Whether the singularity has arrived is ultimately less important than recognizing that the possibility has crossed an institutional threshold. During the summer of 2026, AGI and superintelligence ceased to look exclusively like far-future technological endpoints. They became questions about present-day investment, employment, infrastructure, cybersecurity, competition and governance.
Perhaps we are not in the singularity. Perhaps Altman is describing the beginning of a long exponential curve rather than its vertical ascent. Perhaps today’s models will encounter stubborn architectural limits and the grand predictions will join a long history of technological forecasts that arrived decades late.
But financial services cannot wait for someone to ring a bell announcing AGI.
If increasingly capable machines can help design their successors, if agents can coordinate with other agents, if models can interact with and reason about the physical world, and if the cost of machine intelligence keeps falling, then the relevant question for financial institutions is changing.
It is no longer simply, “When will AGI arrive?”
It is: “What should we be doing if the transition has already begun?”
Researched by DWN Staff
