Artificial general intelligence

Artificial general intelligence, or AGI, is a theoretical form of artificial intelligence. It would match or even surpass human capabilities across almost all cognitive tasks. This is very different from artificial narrow intelligence, or ANI. Narrow intelligence is confined to specific, well-defined tasks. An AGI system would be able to generalize knowledge. It could transfer skills between different domains. It could also solve novel problems without needing task-specific reprogramming. Some people also discuss artificial superintelligence, or ASI. ASI is a hypothetical type of AGI that would outperform the best human abilities across every domain by a wide margin.

Researchers have proposed various ways to define the intelligence of these machines. There is no single agreed-upon definition for computer intelligence. Computer scientist John McCarthy noted in 2007 that we cannot yet characterize all the computational procedures we call intelligent. However, many researchers believe an AGI must possess specific traits. These include reasoning, using strategy, and solving puzzles. A system should also make judgments under uncertainty and represent knowledge, including common sense. It must be able to plan, learn, and communicate in natural language. Some experts also consider imagination and autonomy to be important traits.

To classify how close we are to AGI, researchers at Google DeepMind proposed a framework in 2023. They define five performance levels: emerging, competent, expert, virtuoso, and superhuman. A competent AGI is defined as an AI that outperforms 50% of skilled adults in many non-physical tasks. A superhuman AGI, which is an ASI, would reach a threshold of 100%. Currently, researchers consider large language models like ChatGPT or LLaMA 2 to be instances of emerging AGI. They also define five levels of autonomy. These range from a "tool," which is fully under human control, to an "agent," which is fully autonomous.

Scientists use several famous tests to see if a machine has reached human-level intelligence. The Turing test was proposed by Alan Turing in 1950. In this test, a human judge has natural language conversations with both a human and a machine. If the machine convinces the judge it is human, it passes. In 2025, a study showed GPT-4.5 passed this test by being judged human in 73% of conversations. Another test is the "Ikea test," or the Flat Pack Furniture Test. This involves a robot attempting to assemble furniture from instructions. In 2013, MIT's IkeaBot assembled a table in ten minutes without human intervention.

Other tests focus on real-world interaction and problem-solving. Steve Wozniak proposed the "coffee test." A machine must enter a home and figure out how to make coffee by itself. In 2024, Figure AI's Figure 01 humanoid learned to use a Keurig machine by watching videos. In 2025, researchers at the University of Edinburgh demonstrated a robotic arm that makes coffee in dynamic kitchens. Another idea is Suleyman's test, which asks an AI to turn $100,000 into $1 million. Finally, some suggest using video games to test intelligence. This would involve an AI succeeding in games that its developers have never seen before.

The history of AGI research began in the mid-1950s. Early researchers were very optimistic about how quickly AGI would arrive. In 1965, Herbert A. Simon predicted machines could do any human work within twenty years. Marvin Minsky also believed the problem would be solved within a single generation. These predictions even inspired the fictional character HAL 9000. However, by the early 1970s, it became clear that researchers had underestimated the difficulty. Funding agencies became skeptical, and researchers faced pressure to create "applied AI" instead. This led to a period in the 1990s where researchers avoided talking about human-level AI to avoid being called dreamers.

Today, the pursuit of AGI is a major goal for many technology companies. OpenAI, Google, xAI, and Meta are all working toward this technology. A 2020 survey identified 72 active AGI research and development projects across 37 countries. This work brings up significant debates about safety and the future of humanity. Some experts argue that mitigating the risk of human extinction from AGI should be a global priority. They view AGI as a potential existential risk. Other experts believe that the development of AGI is still too remote to present such a danger. The topic remains a central focus of both science fiction and serious scientific study.