Artificial General Intelligence: Why aren't we there yet?
summary
Argues that modern AI, dominated by convolutional networks and deep learning, excels at narrow perception but fails on common-sense reasoning and safety. Gives concrete failures: Google image captioner calls a parking-sign-covered car a refrigerator, vision nets hallucinate school buses or digital clocks, and a 97.5% pedestrian detector still kills when scaled to fleets. Contrasts AlphaGo’s infinite self-play data with infant and ibex zero-shot learning and urges embedding structured knowledge rather than blank-slate learning.
AI-written summary · about
chapters (4)
related talks
04mFrom big data to human-level artificial intelligence
45mThe Curious Incident of Developing Artificial General Intelligence
48mArtificial Intelligence. Our journey towards singularity
52mArtificial, yes; Intelligent, no.
21mWhen the machines out-smarts us
18mTrinity of Artificial Intelligence
39mThe Pocket Guide to Artificial Intelligence
26mArtificial Intelligence? - More like Artificial Stupidity!