AGI Can Be Safe
Data Skeptic
00:00
The Future of Artificial Intelligence
Neural networks can self-learn early feature engineering layers in computer vision and end-to-end transcription systems in translation or transcription./nStructures that can be built by hand may not automatically appear in AI despite training, suggesting the importance of putting some priors or symbolic reasoning into AI./nThere is a need to look at hybrid techniques, such as neuro-symbolic, instead of solely relying on deep neural nets./nThere are surprisingly brittle aspects of deep neural nets, which can easily misinterpret human commands regarding human values and emotions./nThe speaker is involved in writing technical standards for AI safety to support the European AI Act.
Transcript
Play full episode
