Artificial general intelligence (AGI) is a type of artificial intelligence (AI) that matches or surpasses human cognitive capabilities across a wide range of cognitive tasks. This contrasts with narrow AI, which is limited to specific tasks.^[1] Artificial superintelligence (ASI), on the other hand, refers to AGI that greatly exceeds human cognitive capabilities. AGI is considered one of the definitions of strong AI.

Creating AGI is a primary goal of AI research and of companies such as OpenAI^[2] and Meta.^[3] A 2020 survey identified 72 active AGI research and development projects across 37 countries.^[4]

The timeline for achieving AGI remains a subject of ongoing debate among researchers and experts. As of 2023, some argue that it may be possible in years or decades; others maintain it might take a century or longer; a minority believe it may never be achieved; and another minority claims that it is already here.^[5][6][7] Notable AI researcher Geoffrey Hinton has expressed concerns about the rapid progress towards AGI, suggesting it could be achieved sooner than many expect.^[8]

An OpenAI employee, Vahid Kazemi, recently claimed that the company has achieved AGI with its latest model, O1, stating, "In my opinion, we have already achieved AGI and it’s even more clear with O1." Kazemi clarified that while the AI is not yet "better than any human at any task," it is "better than most humans at most tasks." He also addressed criticisms that large language models (LLMs) merely follow predefined patterns, comparing their learning process to the scientific method of observing, hypothesizing, and verifying. These statements have sparked debate, as they rely on a broad and unconventional definition of AGI—traditionally understood as AI that matches human intelligence across all domains. Critics argue that, while OpenAI's models demonstrate remarkable versatility, they may not fully meet this standard. Notably, Kazemi's comments came shortly after OpenAI removed "AGI" from the terms of its partnership with Microsoft, prompting speculation about the company’s strategic intentions.^[9]

There is debate on the exact definition of AGI and regarding whether modern large language models (LLMs) such as GPT-4 are early forms of AGI.^[10] AGI is a common topic in science fiction and futures studies.^[11][12]

Contention exists over whether AGI represents an existential risk.^[13][14][15] Many experts on AI have stated that mitigating the risk of human extinction posed by AGI should be a global priority.^[16][17] Others find the development of AGI to be too remote to present such a risk.^[18][19]

AGI is also known as strong AI,^[20][21] full AI,^[22] human-level AI,^[5] human-level intelligent AI, or general intelligent action.^[23]

However, some academic sources—often arising from an idealist philosophical stance—reserve the term "strong AI" only for computer programs that experience sentience or consciousness.^[a] From a materialist philosophy and science view, in which the brain and consciousness are regarded as one and the same physical reality, these idealist arguments are considered philosophically distinct or even misguided, because they treat consciousness as something separate from physical processes. In contrast, weak AI (or narrow AI) is able to solve one specific problem but lacks general cognitive abilities.^[24][21] Some academic sources use "weak AI" to refer more broadly to any programs that neither experience consciousness nor have a mind in the same sense as humans.^[a]

Related concepts include artificial superintelligence and transformative AI. An artificial superintelligence (ASI) is a hypothetical type of AGI that is much more generally intelligent than humans,^[25] while the notion of transformative AI relates to AI having a large impact on society, for example, similar to the agricultural or industrial revolution.^[26]

A framework for classifying AGI in levels was proposed in 2023 by Google DeepMind researchers. They define five levels of AGI: emerging, competent, expert, virtuoso, and superhuman. For example, a competent AGI is defined as an AI that outperforms 50% of skilled adults in a wide range of non-physical tasks, and a superhuman AGI (i.e. an artificial superintelligence) is similarly defined but with a threshold of 100%. They consider large language models like ChatGPT or LLaMA 2 to be instances of emerging AGI.^[27]

Various popular definitions of intelligence have been proposed. One of the leading proposals is the Turing test. However, there are other well-known definitions, and some researchers disagree with the more popular approaches. ^[b]

However, researchers generally hold that intelligence is required to do all of the following:^[29]

• reason, use strategy, solve puzzles, and make judgments under uncertainty
• represent knowledge, including common sense knowledge
• plan
• learn
• communicate in natural language
• if necessary, integrate these skills in completion of any given goal

Many interdisciplinary approaches (e.g. cognitive science, computational intelligence, and decision making) consider additional traits such as imagination (the ability to form novel mental images and concepts)^[30] and autonomy.^[31]

Computer-based systems that exhibit many of these capabilities exist (e.g. see computational creativity, automated reasoning, decision support system, robot, evolutionary computation, intelligent agent). There is debate about whether modern AI systems possess them to an adequate degree.

Other capabilities are considered desirable in intelligent systems, as they may affect intelligence or aid in its expression. These include:^[32]

• the ability to sense (e.g. see, hear, etc.), and
• the ability to act (e.g. move and manipulate objects, change location to explore, etc.)

This includes the ability to detect and respond to hazard.^[33]

Although the ability to sense (e.g. see, hear, etc.) and the ability to act (e.g. move and manipulate objects, change location to explore, etc.) can be desirable for some intelligent systems,^[32] these physical capabilities are not strictly required for an entity to qualify as AGI—particularly under the thesis that large language models (LLMs) may already be or become AGI. Even from a less optimistic perspective on LLMs, there is no firm requirement for an AGI to have a human-like form; being a silicon-based computational system is sufficient, provided it can process input (language) from the external world in place of human senses. This interpretation aligns with the understanding that AGI has never been proscribed a particular physical embodiment and thus does not demand a capacity for locomotion or traditional “eyes and ears.”^[34]

Several tests meant to confirm human-level AGI have been considered, including:^[35][36]

The Turing Test (Turing)

Proposed by Alan Turing in his 1950 paper "Computing Machinery and Intelligence," this test involves a human judge engaging in natural language conversations with both a human and a machine designed to generate human-like responses. The machine passes the test if it can convince the judge it is human a significant fraction of the time. Turing proposed this as a practical measure of machine intelligence, focusing on the ability to produce human-like responses rather than on the internal workings of the machine.^[38]

Turing described the test as follows:

The idea of the test is that the machine has to try and pretend to be a man, by answering questions put to it, and it will only pass if the pretence is reasonably convincing. A considerable portion of a jury, who should not be expert about machines, must be taken in by the pretence.^[39]

In 2014, a chatbot named Eugene Goostman, designed to imitate a 13-year-old Ukrainian boy, reportedly passed a Turing Test event by convincing 33% of judges that it was human. However, this claim was met with significant skepticism from the AI research community, who questioned the test's implementation and its relevance to AGI.^[40][41]

More recently, a 2024 study suggested that GPT-4 was identified as human 54% of the time in a randomized, controlled version of the Turing Test—surpassing older chatbots like ELIZA while still falling behind actual humans (67%).^[42]

The Robot College Student Test (Goertzel)

A machine enrolls in a university, taking and passing the same classes that humans would, and obtaining a degree. LLMs can now pass university degree-level exams without even attending the classes.^[43]

The Employment Test (Nilsson)

A machine performs an economically important job at least as well as humans in the same job. AIs are now replacing humans in many roles as varied as fast food and marketing.^[44]

The Ikea test (Marcus)

Also known as the Flat Pack Furniture Test. An AI views the parts and instructions of an Ikea flat-pack product, then controls a robot to assemble the furniture correctly.^[45]

The Coffee Test (Wozniak)

A machine is required to enter an average American home and figure out how to make coffee: find the coffee machine, find the coffee, add water, find a mug, and brew the coffee by pushing the proper buttons.^[46] This has not yet been completed.

The Modern Turing Test (Suleyman)

An AI model is given $100,000 and has to obtain $1 million.^[47][48]

A problem is informally called "AI-complete" or "AI-hard" if it is believed that in order to solve it, one would need to implement AGI, because the solution is beyond the capabilities of a purpose-specific algorithm.^[49]

There are many problems that have been conjectured to require general intelligence to solve as well as humans. Examples include computer vision, natural language understanding, and dealing with unexpected circumstances while solving any real-world problem.^[50] Even a specific task like translation requires a machine to read and write in both languages, follow the author's argument (reason), understand the context (knowledge), and faithfully reproduce the author's original intent (social intelligence). All of these problems need to be solved simultaneously in order to reach human-level machine performance.

However, many of these tasks can now be performed by modern large language models. According to Stanford University's 2024 AI index, AI has reached human-level performance on many benchmarks for reading comprehension and visual reasoning.^[51]

Modern AI research began in the mid-1950s.^[52] The first generation of AI researchers were convinced that artificial general intelligence was possible and that it would exist in just a few decades.^[53] AI pioneer Herbert A. Simon wrote in 1965: "machines will be capable, within twenty years, of doing any work a man can do."^[54]

Their predictions were the inspiration for Stanley Kubrick and Arthur C. Clarke's character HAL 9000, who embodied what AI researchers believed they could create by the year 2001. AI pioneer Marvin Minsky was a consultant^[55] on the project of making HAL 9000 as realistic as possible according to the consensus predictions of the time. He said in 1967, "Within a generation... the problem of creating 'artificial intelligence' will substantially be solved".^[56]

Several classical AI projects, such as Doug Lenat's Cyc project (that began in 1984), and Allen Newell's Soar project, were directed at AGI.

However, in the early 1970s, it became obvious that researchers had grossly underestimated the difficulty of the project. Funding agencies became skeptical of AGI and put researchers under increasing pressure to produce useful "applied AI".^[c] In the early 1980s, Japan's Fifth Generation Computer Project revived interest in AGI, setting out a ten-year timeline that included AGI goals like "carry on a casual conversation".^[60] In response to this and the success of expert systems, both industry and government pumped money into the field.^[58][61] However, confidence in AI spectacularly collapsed in the late 1980s, and the goals of the Fifth Generation Computer Project were never fulfilled.^[62] For the second time in 20 years, AI researchers who predicted the imminent achievement of AGI had been mistaken. By the 1990s, AI researchers had a reputation for making vain promises. They became reluctant to make predictions at all^[d] and avoided mention of "human level" artificial intelligence for fear of being labeled "wild-eyed dreamer[s]".^[64]

In the 1990s and early 21st century, mainstream AI achieved commercial success and academic respectability by focusing on specific sub-problems where AI can produce verifiable results and commercial applications, such as speech recognition and recommendation algorithms.^[65] These "applied AI" systems are now used extensively throughout the technology industry, and research in this vein is heavily funded in both academia and industry. As of 2018, development in this field was considered an emerging trend, and a mature stage was expected to be reached in more than 10 years.^[66]

At the turn of the century, many mainstream AI researchers^[67] hoped that strong AI could be developed by combining programs that solve various sub-problems. Hans Moravec wrote in 1988:

I am confident that this bottom-up route to artificial intelligence will one day meet the traditional top-down route more than half way, ready to provide the real-world competence and the commonsense knowledge that has been so frustratingly elusive in reasoning programs. Fully intelligent machines will result when the metaphorical golden spike is driven uniting the two efforts.^[67]

However, even at the time, this was disputed. For example, Stevan Harnad of Princeton University concluded his 1990 paper on the symbol grounding hypothesis by stating:

The expectation has often been voiced that "top-down" (symbolic) approaches to modeling cognition will somehow meet "bottom-up" (sensory) approaches somewhere in between. If the grounding considerations in this paper are valid, then this expectation is hopelessly modular and there is really only one viable route from sense to symbols: from the ground up. A free-floating symbolic level like the software level of a computer will never be reached by this route (or vice versa) – nor is it clear why we should even try to reach such a level, since it looks as if getting there would just amount to uprooting our symbols from their intrinsic meanings (thereby merely reducing ourselves to the functional equivalent of a programmable computer).^[68]

The term "artificial general intelligence" was used as early as 1997, by Mark Gubrud^[69] in a discussion of the implications of fully automated military production and operations. A mathematical formalism of AGI was proposed by Marcus Hutter in 2000. Named AIXI, the proposed AGI agent maximises "the ability to satisfy goals in a wide range of environments".^[70] This type of AGI, characterized by the ability to maximise a mathematical definition of intelligence rather than exhibit human-like behaviour,^[71] was also called universal artificial intelligence.^[72]

The term AGI was re-introduced and popularized by Shane Legg and Ben Goertzel around 2002.^[73] AGI research activity in 2006 was described by Pei Wang and Ben Goertzel^[74] as "producing publications and preliminary results". The first summer school in AGI was organized in Xiamen, China in 2009^[75] by the Xiamen university's Artificial Brain Laboratory and OpenCog. The first university course was given in 2010^[76] and 2011^[77] at Plovdiv University, Bulgaria by Todor Arnaudov. MIT presented a course on AGI in 2018, organized by Lex Fridman and featuring a number of guest lecturers.

As of 2023, a small number of computer scientists are active in AGI research, and many contribute to a series of AGI conferences. However, increasingly more researchers are interested in open-ended le