This paper [1] is based on a study conducted to understand the perspective of people involved in the technical data science work. The research questions that the authors have addressed are regarding the perception of AutoAI technology- Is is looked at as a tool that aids current practices or can have a more significant role in practice, where it can replace humans. They propose that their study be a starting point for future development in HCAI for data science or the "design of data". The “What is AutoAI” heading explains well the technology. As the paper flows, the research uncovers an account of the informants' practices in the workplace. There is a detailed account of the widespread roles the company IBM offers in data science and the expectations from each of them. These are accompanied by the challenges that tag along with each role. The discussion is then taken towards AutoAI, informants' past experiences with the technology, and its perceived benefits and challenges. Trust is a keyword used in the discussion. It is interesting to ask people if they trusted AI for the work that they were doing or not. AutoAI has also been analysed for its fit in different contexts. These are the work that the informants are involved in - teaching their skill, collaborating with the team and clients, automation of data cleaning and manipulation. The paper concludes with a discussion between the two possible directions that data scientists at IBM see it taking - augmentation or automation. Augmentation is the more favoured option and is hence suggested.
The authors read literature to establish a thought process revolving around the current perceptions regarding AutoAI. The description of related work is systematic and has a good flow of thought. It covers different headings that together provide a holistic view of data sciences and AI intervention. The research method used for this study is semi-structured interviews of 20 professionals hired in IBM. The demographics of the informants and selection method is not commented upon. The sampling method adopted was snowball sampling. Since the authors are all from the United States, sampling may have happened in the same country, making the group homogenous. If that were the case, the authors should have been specific and mentioned that the ideology is biased towards IBM USA employees. The sex of the informants were the two binary genders only, but both were almost equally represented. The study focused on the perception of data scientists involved in technical jobs in different domains, in varied power positions, so recruitment of others was unnecessary and hence avoided. The interviews were analysed using open coding. This increased the scope for finding insights since codes are not pre-decided but dependent on the findings.
The interview results reported are incredibly biased towards AutoAI, and painting a positive picture for it. Wang et al. do talk about peoples' scepticism towards certain features and potential uses of the technology. However, everyone seems to agree upon the fact that it is the future and is willing to embrace it. This might not always be the case. Recruiting from a company like IBM that is technologically updated and promotes innovations might be a more significant limitation than what is led by the authors in the paper, where they talk about similar work practices. Common workplace ideology affects personal opinions, which is what semi-structured interviews are built upon. Another possible explanation for the evident bias in the flow of answers is that that questions asked by the authors were biased and leading. Leading questions often lead to similar answers as informants may not contradict and not give their actual opinions.
The research is novel; the authors mention the novelty factor in the paper itself. The title is self-explanatory and complements the research. The paper ends in the recommendation that AutoAI work in the future should be based on augmenting current practices rather than replacing data scientists themselves. The semi-structured interviews, however, had a more extensive scope. Since nine authors took 20 interviews and their analysis makes the per person contribution less. It seems unjustified. The data could be put to use to increase the same. All the discussed problems could be turned into a list of recommendations - a do's and do not's list to be adopted in the company. This could be a possible extension of the research. The recommendations could be tried in the same settings the interviews were conducted, to make practices in IBM better.
Reference