Talk:Machine Learning Underlying Technology and Ethical Issues
Length of article is at roughly 2800 words, without counting the content block. I see that you have a couple subheadings under "Real World Impacts of Bias and Unfairness in Machine Learning" which hasn't been written about yet. The subheadings look like they could be expanded into decent size paragraphs so you should be good on word count then.
The article does contain the 3 major components of a good article. The topic is summarized effectively in the opening paragraph, however I felt that it contained too much information and lacked (nearby) citations. Specifically, the details of gender bias for the Apple Card felt out of place, while the assertion of facts such as "The field of machine learning has grown significantly since the 1970s due to the rapid increase in computational power as well as the decreased costs", and "Conversations about the ethics of machine learning are becoming more widespread and there has been an increase in research into mitigation strategies", among others, seem to lack supporting citations. In particular, the assertion that computational power (and lower costs) was the only reason for growth since the 70s is contradictory to the later section explaining the AI Winter- when researchers/funders lost interest in the subject, and later regained interest (fueling growth) when new research did emerge.
The article is also split into multiple logical sections, but I feel that the subheadings on "Bias in Neural Networks" and "Methods to Limit Bias in Neural Networks" should be under a "Bias in ML" heading rather than under ML algorithms, as these are, based on the article (and from my EECS445 experience), parameter tuning techniques based on the AUROC graph/Grad-CAM and usually applied post-training.
Citations and references are present throughout the article, but feel lacking in some paragraphs, and are absent for some statements, as elaborated above. Most of the references seem to be from reliable journals/websites. Many of the references are repeated, you can name a tag by using < ref name=xyz >< /ref > and refer to the same reference multiple times without repeats showing up at the bottom.
The article is seems to go into too much detail with regards to implementation of algorithms, the usage of machine-learning lingo without definition (e.g. feature vector dimensions, gradient based localization, hyperparameters), and out-of-place examples, such the second sentence till the second last sentence under the supervised learning subheading. In addition, the lack of content in the final heading fails to highlight the importance of the ethical issues. As such, although I believe the issue at stake is conveyed through the article and I personally understand the ethical issues and implications being discussed, I don't think it is done with enough clarity for a layperson to interpret. Also, clarifying the grass and dog breed grad-CAM example as being a "ML Shortcut" would be a good addition, in my opinion.
From the content present as of this review, the reporting on ethical issues regarding machine learning algorithms is, for the most part, objectively reported. An exception to this is under the subheading of "Bias in Neural Networks", where the second half of the paragraph appears to be written from a first-person anecdotal perspective. I would suggest rewriting that portion to be presented from a third person perspective. Aside from that, descriptions of the ethical issues are presented and elaborated on, without the author arguing for a particular side of the ethics. Multiple perspectives are offered in the form of methods to reduce bias, in addition to the discussion of the bias.
Overall I think the article is well written as a draft article.
