Narrow AI vs General AI (AGI)
Level 3 · Kinds of AI
Why all AI in use today is narrow AI, and what AGI and a mind far past ours would mean. Why experts disagree on when, or if, they come.
Course contentsLesson 5 of 10
Level 1 · What AI is
Level 2 · AI built from rules
Level 3 · Kinds of AI
Level 4 · AI history
Level 5 · Big questions
After this lesson you can
- Tell narrow AI from AGI
- Explain what superintelligence would add
- Say why no one can put a date on AGI
11 mincovers 3 glossary entries
Worth reading first: Artificial Intelligence (AI)
Narrow AI, AGI and superintelligence: what each one means
Narrow AI is AI built for one kind of job, or a small set of them. Every AI in use today is narrow. A chess program can beat any human, yet it cannot sort an email. Chat apps reach much further than that, but they still fall short of a person's full range.
One way to see the three is by range. Put the tasks a mind can learn on a line, from one task up to every task. Narrow AI sits near the start. AGI would sit at the human mark. Superintelligence would sit far past it.
Narrow AI is very good inside its lane. A spam filter, a face finder and a route planner each beat people at their own task. None of them can move its skill to a new kind of task. The program that beat the best Go player could not play a game of chess.
Artificial general intelligence, or AGI, would be AI that can learn almost any thinking job a person can. It would handle new problems it was never trained for. It does not exist yet. Several big AI labs say building it is their goal, and no one agrees on a test that would prove it is here.
Superintelligence goes one step further: AI far smarter than the best people at almost everything. It is an idea about the future, not a product. Some experts think it could follow soon after AGI. Others doubt it will ever come. The worry about it drives much of the work on keeping AI safe.
Experts do not agree on dates. In a 2023 survey of 2,778 AI researchers, the middle guess gave a 50% chance that machines beat people at every task by 2047. That was 13 years sooner than the same survey a year before. Guesses that swing this much show how unsure they are.
People often take a smooth chat as a sign of AGI. Sounding human is not the same as being able to learn any job. Others swing the other way and say narrow AI is not real AI at all. Both miss the point: narrow AI is what exists, and it is already very useful.
The three kinds side by side
| Kind | What it can do | Does it exist? |
|---|---|---|
| Narrow AI | One kind of task, often better than people | Yes, all AI today |
| AGI | Almost any thinking task a person can do | Not yet |
| Superintelligence | Almost everything, far better than people | No, it is an idea |
Worked example
Emma tests her chess app
Emma's chess app beats her in every game. It would beat the world champion too.
She asks it to plan a weekend trip, and it cannot even start. All its skill sits in one lane, which makes it narrow AI.
The words for this lesson3
Check yourself
Is any AI in use today an AGI?
Finished this lesson?
Marking it complete also marks its 3 glossary entries learned, which updates your roadmap and your saved page. 0 are already marked.