The article name is: Automating Software Developments with Deep Learning -- https://www.infoq.com/presentations/automated-software-dl/
Why I Found It Interesting: I found the article especially interesting becuase it provides a new insight in viewing software development as a data problem rather than as a human problem. These years, artificial intelligence and automation has became increasing popular and trending. Lots of products are labeled with the word intelligence while computer science and machine learning are now the popular major college students select. Nevertheless, although I have been in self-driving tesla and learned the competitions of Chess between human and Alphago, it's the first time for me to have real outlook and think about the feasibility of automation of software developments reading this particular article. In the article, the author points out how the traditional approach of hard-coded logic is currently not as efficient compared to the emergnig gradient-based techniques. Instead of writing logic codes, scientists now trained and teached models with datasets. This reminds of the React.js we learned in class where it is used to auto-generate HTML, CSS and Javascript so that human-beings no longer needs to write functions that shares similar features. Although it does not fall into the exact category of data-driven automation, it shares similar ideas. The development from Software 1.0 to Software 2.0 then draws a blurry boundaries between software engineering and data science for which software engineers now focus on datasets and pipelines as opposed to crafting logic. This let me be confused about the distinction and wonder if ultimately, every technology will inclined into data-driven models. As I read through the article, I learned about the disadvantage -- strong precision required for coding, compared to that for natural language. In my opinion, software automaiton is convenient and time-saving, however, the precision cannot always be guaranteed so that if everything is automated and driven by data, models might collapse in some occasions which could craete dangers if it is incoprated into daily lives. Therefore, I advocate for the Software 2.5 approach to incoprates both rule-based logic and gradient based techniques. Overall, as a conclusion, I am fascinated by the crossroads of software developments that with deep learning. It beckons developers to be aware of the shifts and to anticipate the trajectory of software development's future. I truly hope to learn more about this field.