AI Systems: Foundations for Artificial Minds
summary
Argues that creating an artificial mind requires a systems approach that stitches symbolic rules to learned representations, using Clojure plus Datomic data provenance to tame ML technical debt. Demonstrates empirical pieces: convolutional nets for imagery, word embeddings (word2vec/GloVe) that support analogy like puppy - dog + cat = kitten, and TensorFlow graphs for deployment. Claims hybrid systems (as in AlphaGo and self‑driving stacks) are practical today, and shows a provenance-backed pipeline that reproduces and debugs model inputs and failures.
AI-written summary · about
chapters (11)
- 0:00Intro
- 0:17Annotated Slides
- 8:04Solution Components
- 8:35Machine Learning Advances
- 11:56Good Representations
- 14:38Why do humans do it?
- 23:48Examples of Hinges
- 33:32Affordances when in Mixed Systems
- 35:20Machine Leaming and its Technical Debt
- 37:18The machine learning world is trying to solve these problems
- 38:51Datomic Wins
related talks
42mReasonable AI
33mHow Might Artificial Intelligence Come About?
21mSentient AI – Can We Create Artificial Consciousness?
54mArtificial Intelligence seen from the software development lifecycle perspective
41mFrom AI Software Engineers to AI Knowledge Workers
21mLearning from History: The Past as a Compass for Systems Engineers in the Age of AI
13mArtificial General Intelligence: Why aren't we there yet?
46mAI vs Humans Rewriting the Rules of Success