Your phone's autocorrect and a hypothetical world-changing superintelligence are both technically called AI. That's a problem

[Grace24]

Nearly everything labeled AI today, from spam filters to voice assistants to the most advanced chatbots on the market, falls into a single category researchers call narrow AI, sometimes also called weak AI. These systems can be extraordinarily capable within their specific lane, but they cannot genuinely transfer their skill to a task they were never trained for, a chess playing AI has no path to suddenly understanding medicine, no matter how good it gets at chess. Even the most impressive chatbots people interact with daily are, by this definition, narrow AI, exceptionally broad narrow AI, but narrow all the same

Artificial General Intelligence, or AGI, describes something categorically different, a system able to learn, reason and adapt across essentially any domain the way a competent human generalist would, transferring knowledge from one unfamiliar situation to another rather than failing outside its original training. No such system exists today, and researchers disagree sharply about how close current AI actually is to crossing that line, or whether today's approach can even get there at all without some further fundamental breakthrough

Artificial Superintelligence, or ASI, sits a further step beyond that, a hypothetical system that does not just match human level reasoning but exceeds the smartest humans across essentially every intellectual domain simultaneously. Most researchers treat the path to ASI as running directly through AGI first, and both remain firmly in the realm of research and speculation rather than engineering reality, however much that distinction gets blurred in casual conversation, marketing copy, and breathless news coverage that uses the word AI to describe all three tiers interchangeably

Types of Artificial Intelligence | IBMwww.ibm.com