Artificial intelligence (AI) is changing industries and everyday life, and it is developing in ways that looked like science fiction only a few years ago. But not all AI is the same. It is usually divided into three types according to what it can do: narrow AI, general AI and superintelligent AI. Knowing the difference helps you see where AI stands today and where it might go next.

New to the subject? Start with An Introduction to Artificial Intelligence.

Narrow AI (weak AI)

Narrow AI is the AI we use today, and it is all around us. It is called “narrow” or “weak” because it is designed for a single task or a set of closely related tasks. It works within a limited range of abilities and cannot do anything outside them.

Examples of narrow AI

  • Voice assistants such as Siri and Alexa answer questions, play music and control smart devices.
  • Recommendation systems on platforms such as Netflix and YouTube suggest films and videos based on what you have watched.
  • Image recognition software identifies faces in photos and detects tumours in medical images.
  • Spam filters separate unwanted email from the messages you want.

What defines narrow AI

  • Specialised: It excels at specific tasks and fails outside them.
  • No real understanding: It does not “understand” its task; it is trained to recognise patterns in data.
  • Widespread: It powers most of the AI-driven technology we use today.

Is narrow AI really “weak”?

Despite the label, narrow AI can be remarkably powerful within its domain. In medicine, for example, some systems analyse medical images faster than human experts and, in some cases, more accurately. But these systems cannot do anything outside the purpose they were built for.

General AI (strong AI)

General AI, also called strong AI or artificial general intelligence (AGI), is a theoretical form of AI that could perform any intellectual task a person can. Where narrow AI is limited to specific tasks, general AI would understand, learn and apply intelligence across a broad range of them, much as a human does.

What general AI would look like

  • Flexible and adaptable: It would switch between tasks with ease. It could learn to play a musical instrument, solve a maths problem or hold a meaningful conversation, as a person can.
  • Human-like understanding: It would have reasoning, understanding and a sense of self-awareness, allowing it to interpret context and emotion.
  • Learning across domains: It would not need retraining for each new problem; it would learn and adapt on its own across different fields.

How close are we?

General AI does not exist yet. Today’s most capable systems, the large language models behind chatbots such as ChatGPT, can write, translate, summarise and produce code, which makes them far broader than earlier AI. Whether they are a step towards general AI or a very capable form of narrow AI is an open debate, not least because researchers do not agree on how AGI should be defined or measured. Reaching human-level understanding and flexibility would probably take further breakthroughs in reasoning, in learning beyond the patterns found in large datasets and, some argue, in our understanding of consciousness.

What stands in the way

  • The complexity of human intelligence: Reproducing the abilities of the human brain, including reasoning, emotion and creativity, is an enormous challenge.
  • Ethics and safety: General AI raises serious concerns, from job displacement to the risk of systems acting in ways nobody intended.

Superintelligent AI

Superintelligent AI is the idea of an AI that surpasses human intelligence in every field, from creativity and problem-solving to emotional intelligence. It would be more capable than any person in virtually every domain, and it could reshape society in ways that are hard to predict.

What superintelligence would mean

  • Beyond human ability: It would be smarter than people in every respect, including scientific reasoning, social skills and creativity.
  • Self-improving: It might be able to improve and evolve on its own, leading to exponential gains in capability.
  • Radical change: With abilities beyond human comprehension, it might solve complex problems such as disease, climate change and poverty. It could also pose existential risks.

Is superintelligence possible?

Superintelligent AI is speculative, but prominent figures have taken the idea seriously, among them the late physicist Stephen Hawking and the entrepreneur Elon Musk. Some experts believe that general AI, once achieved, could eventually lead to superintelligence through continuous self-improvement. Whether and when that might happen is much debated.

Risks and opportunities

  • Opportunities: Superintelligent AI could deliver breakthroughs in science and medicine that are beyond our reach today.
  • Risks: It could become uncontrollable, with unpredictable consequences. Some researchers argue for strict controls and ethical guidelines to manage that risk.

Where we stand today

AI today is primarily narrow AI: systems built for specific tasks such as recognising images, translating languages and recommending content. General AI and superintelligent AI remain theoretical, and much work lies ahead before they become reality, if they ever do. Understanding the difference helps us see both what AI can do now and what it might do in future, along with the ethical and safety questions that come with each step forward.

AI has already brought far-reaching change, and its future holds possibilities that are exciting and sometimes daunting. It is worth staying informed about where we are and where we might be going.