Narrow AI vs AGI
Today's most capable systems are still narrow in important ways. AGI describes a system that is general across the dimensions below - not just powerful on any one of them.
Note: the "AGI" column describes the target capability profile researchers associate with the term. Today's frontier systems sit on a spectrum between these two columns and have made measurable progress on reasoning, transfer, and autonomy in the past three years.
Human intelligence vs AI, and where machine reasoning fits
The comparison AGI vs AI is really a comparison of generality. Today's artificial intelligence is a collection of specialised AI systems; artificial general intelligence is one integrated system that behaves competently across the same range of tasks a person can. The deeper contrast - human cognition vs machine intelligence - involves embodiment, motivation, social grounding, and lifelong learning, not just benchmark scores.
Human reasoning is slow, energy-efficient, and grounded in a body and a culture. Machine reasoning is fast, parallelisable, and ungrounded by default. Recent AI reasoning models close part of the gap by externalising chains of thought, but AI limitations remain real: brittle out-of-distribution behaviour, weak long-horizon memory, factual hallucination, and an absence of genuine goals. AGI capabilities, by contrast, are defined by the absence of those limitations - which is precisely why no current system qualifies.
On creativity, the picture is mixed. Human creativity vs AIis not a contest: humans set problems and stakes, while machine creativityexpands the search space of forms, drafts, and variations. Future AI systemsand general intelligence systems will likely amplify human creative work long before they replace it - the pattern already visible in design, scientific writing, and software engineering.