There are several new conceptions for me in the readings of this week. From Alpaydin and Kelleher, we know that Deep Learning intimates human brains to build neurons and set false neural networks with several layers so that the learning algorithm could develop recognitions on what has been learned/processed. In the meantime, from the Introductory Essay written by Prof. Irvine, we learned that how modern computers work on the basis of the binary system (0&1), which makes DL even more magical since mentioned by Alpaydin in Machine Learning: The New AI, it studies with “hidden layer combining the values (which are not 0 or 1 but continuous allows a finer and graded representation of similar inputs).” In other words, it looks like DL is based on the binary system but surpasses that, which means DL supports the development of AI to learn more abstract things (more like humans).  

In addition, about the discourse in AI and ML, I also learned about several clarifications. According to the video- Techniques for Interpretable Machine Learning released by Association for Computer Machinery, “powered by complex models and deep neural networks, interpretable ML is progressing at an astounding rate; however, despite the successes, ML has its limitations and drawbacks- some decisions made by algorithms with ML are hard to interpret.” This fact relates to Reframing AI Discourse. ‘Machine autonomy’ is not equal to human autonomy. Although designers set patterns for the AI system, the AI will become an entity (run by rules that may be unexpected when encountering real problems). This kind of entity does not mean AI can determine where it will go by itself but become an independent program if there is no intervention. However, this also exposes practical problems: how should we make the regulations for an AI system; how could we evaluate the purpose of designers (whether we can get help from six principles proposed by Alpaydin); how should we make the guide for AI practice? Assumptions are assumed for us to predict the future, but questions are asked to solve. Although Johnson and Verdicchio said that the popular AI concepts are futuristic and too hard to achieve, we need to plan for the future.