2017 — Book 21 — “Thinking Machines“
Jul 22, 2017 · 2 min read
The Quest for Artificial Intelligence and Where it’s Taking Us Next
By Luke Dormehl
Interesting book which covers the history, current state and future directions of artificial intelligence.
Some important take-away’s:
- the approach to AI in the 70’s and 80’s is now known as “symbolic AI” — in which expert systems were trained by providing as many rules as possible and having systems deduce from the known rules
- this approach has been superseded by the “deep learning” approach to AI, in which systems are taught only the desired outcomes (a problem to solve or value to be optimized) and left to “learn” or find a solution
- the risk of the deep learning approach is that an AI system will make a decision or come to a conclusion without anyone being able to explain the rationale behind the decision. In some instances, this is fine — “find me the optimal driving route to Amarillo, knowing I have plenty of time, little money and I am a vegetarian who loves Chipotle”; but in other instances, not being explain the reasoning for a decision is unacceptable — “decide whether to give an applicant a mortgage loan”
- the author does an excellent job providing a coherent explanation of the what is meant by “the singularity” — in this explanation, the idea is that an AI can be coded to eventually become better at humans in almost any task; if an AI becomes better than humans at creating AI’s (which create better AI’s), the process will suddenly explode exponentially, very quickly resulting in AI’s with general intelligence eclipsing humans. We would be left very far behind, very quickly.