Is AI better than humans in real life?

Waymo's self-driving cars have 82% fewer injury crashes than human drivers (June 2026), and AI now forecasts world events as well as top human forecasters. Robot AI still trails people steering the robot: it finishes 13% of 18 real-world tasks, people 100% (July 2026).

Updated

Where AI matches or beats people

Drive a car271 million miles with no one at the wheel.Waymo · Jun 2026
Waymo18
Human drivers100
Injury crashes over the same distance, human drivers = 100
AI ahead 82% fewer injury crashes
Predict world eventsNew questions every two weeks, scored once the real answer is known.Forecasting Research Institute · Oct 2026
Google DeepMind69
Superforecasters (median)69
100 = every call right, 50 = no better than a coin flip
Level

Where people still win

Do chores with a robot18 real tasks on real robots, 10 tries each.RoboDojo · Jul 2026
π0.5 (robot AI)13%
People steering the robot100%
People ahead by 87 points
Translate a novelReaders compared both translations of 15 new novels without knowing which was which.arXiv study · Sep 2026
Preferred the AI one11
Preferred the human one19
Out of 30 book excerpts
People ahead 19 of 30
Read messy handwritingNotes, tables and formulas in four languages. Tables are the hardest.WildHandBench · Aug 2026
Gemini 3.1 Pro72%
People77%
People ahead by 5 points
Draw a floor plan from photosAbout 20 photos of a flat in, a floor plan out. Andon Labs' own estimate: AI catches people around Nov 2026.Andon Labs · Oct 2026
Gemini 4 Argon54
People59
100 = a perfect plan, 0 = random
People ahead by 5 points

How often does AI make things up?

1 in 15

summaries by GPT-6 Sol add something the document never said. It is the lowest of the 5 models shown.

Vectara · Oct 2026

GPT-6 Sol7%
DeepSeek V4 Pro9%
GPT-6 Astra9%
Claude Opus 4.712%
Gemini 4 Argon12%

Share of summaries with a claim the document doesn't contain. Bars run from 0 to 20%.

At work: AI against professionals, field by field

Sources (7)
  1. Waymo, waymo.com/safety/impact
  2. Forecasting Research Institute, forecastbench.org/leaderboards
  3. arXiv, arxiv.org/abs/2607.04434
  4. arXiv, arxiv.org/abs/2606.26040
  5. arXiv, arxiv.org/abs/2608.22959
  6. Andon Labs, andonlabs.com/evals/blueprint-bench-2
  7. GitHub, github.com/vectara/hallucination-leaderboard