Artificial Intelligence and Machine Learning to help with Literature Reviews
Artificial Intelligence (AI) and Machine Learning (ML) are hot topics. Entrepreneurs, politicians, and scholars alike draw on these concepts to envision future opportunities that can be made possible through technological advances. At the same time, they also incite fear in those whose jobs may ultimately be automated because of such changes. As educators and researchers, we may believe that our positions are largely shielded from such developments (despite recent advances in neural interface technology which could, one day, enable humans to link directly with computers and learn from them). There are, however, areas where AL and ML already impact our work and could theoretically, at some point, replace our efforts. This subject is addressed in a recent study by Sebastian Robledo, Andrés Mauricio Grisales Aguirre, Mathew Hughes, and Fabian Eggers, entitled ‘“Hasta la vista, baby” – will machine learning terminate human literature reviews in entrepreneurship?’ and published in the Journal of Small Business Management.
To address the question that the article so provocatively pose in its title, the authors set out to conduct a literature review using AI and ML. As Sebastian explained: “we started searching for tools that applied artificial intelligence in literature review and realized that most of them are for medical purposes … so in the beginning that was a big challenge, because natural language processing was just for medical papers, and we wanted to apply it to an entrepreneurial topic.” Mat agreed that “there was a lot of legwork that needed to be done in terms of adapting … [and] training the software. Human beings can only process so much information – we can’t process thousands of papers. Not only can the machine do that, it needs to do that. The machine needs volume in order to learn right what's going on, what to look for and so then it goes through episodes of learning … So, for me that was fascinating because it really opened up our eyes.”
Indeed, the process of developing this paper involved a lot of learning for the authors. As Mat described: “Assumptions that we had, going in about various benefits and advantages [of AI], did not play out the way that we were expecting … the real thing in the back of our mind was ‘Well, this is going to be better than the human being’ and, in reality, actually, we found that it doesn't replace the human being at all, but it can supplement the human being, and it can give reassurance that the review you’ve done is accurate and correct. So, it's a really good application tool in that respect, but it doesn't … terminate us at all.”
Mat elaborated that AI “doesn't replace scholars’ agency. Instead, it is there as a support, and I think it's better to think about it along those lines, rather than thinking ‘I don't need to read the literature, I'm going to let the machine do it for me’ … I mean, yeah you can [do that], but what we saw as well [is] how do you know that the machine will be right?” Sebastian added that AI “cannot replace the human process, but machines can push humans to be more creative because we won't need to spend so much time [reading papers], searching and things like that. We will have more time to reflect on things.”
Given the limitations that the authors experienced with the currently available programs, “there's an opportunity for entrepreneurs to create literature review tools that would be beneficial in management research or maybe more broadly social science research … I think the language complexity requirements are quite extensive in social science and management, because in medical research, you can look for specific types of terms and specific types of conditions or symptoms whereas, you know, in management there are substitutes for ideas, where we will … struggle with construct clarity in many cases … so I think the potential is there for these tools to be improved – there's a commercial opportunity here for developers” (Mat).