--Original published at JVershinski's Blog
Getting recommendations from someone is not something we put a lot of thought into. You’re looking for a suggestion to see if anyone else knows someone or something that could help to solve some kind of problem you have. Some people value others’ suggestions, and some people wave them aside. However, it is now being tested that a machine may give better suggestions than other people. New research indicates that machines do indeed give better recommendations in a field that is usually very human-like area, humor.
Participants in this study were given a series of jokes and asked to rate them on a scale from -10 to 10, with -10 being the least funny and 10 being the most funny. However, each experiment in this overall study had a slight difference in how the assessment was made as well as what additional questions were asked. For example, one of the experiments allowed the participants to know whether the recommendations they were being given were human or machine, while a different study told the participants the opposite source of where the joke truly came from. (Ex. If the joke came from a machine, the participants were told it came from a human and vice versa)
Participants from this study were selected from two different locations: the Museum of Science and Industry or Amazon.com’s Mechanical Turk. Participants were randomly selected, but some of the participants responses were dropped due to various reasons including: incomplete responses, failure to comply with instructions, failure to pass manipulation check, or giving the same rating to every joke. In total, there were about 3,750 participants recruited to partake in the study, but only 3,647 of the responses were recorded.
These results should have the ability to be applied to the general public because the participants were selected from the general public. I do not think that these results should be specified to only those from the museum or from the Turk because the people there are no different from the everyday person.
The results of all of the studies were quite interesting. First, it was determined that computer algorithms can generate better recommendations than actual people. This was based off of personal preference of the subjects. Second, the computer generated recommendations can be created from very little information. In the study, the computers that were generating the recommendations did not have any personal information about the person, beyond the ratings which they gave to the jokes. Finally, it was concluded that people preferred human recommendations over machine generated ones, despite the accuracy of the machines being higher. People seem to value human recommendations higher, which influences their opinion of the recommendation. This goes to show that even though it may not be as accurate as a machine, there could still be some value to human recommendations.
Reflection
As I went through the article and wrote my summary of it, I decided to include some of the 5 critical questions, but not all of them. I thought it would be important to include the first two questions about selecting participants and operationalizing values because I thought that it would be useful information the reader should know. However, because the groups chosen were already the groups that would be partaking in the study, there was nothing to write about how the individuals were separated into groups. I also decided to not include whether or not the study allows for causal claims, because I think that based off of the information given, that answer can be figured out by the reader. Finally, I did include the ability of the study to generalize to the public. I think that this is important because it was not included in the study, but it is important because machine recommendations are already working their way into the life of millions of people, so I thought that it was important to point out.
Compared to the article, I think that my summary shares the most prevalent of the 5 critical questions. Both my summary and the article talk about how the study operationalized the values for the subjects, but that was the only question that was included in both. I decided to also share how the study could be generalized to the right population as well as how the participants were selected because I thought it was important in the study. Neither my summary or the article talked about how the participants were split into groups or if the study allowed for causal claims. I think this was a good move because the reader can determine from the information given the answer to both of those questions.
The pop culture article was written with what seemed to be the reader in mind. It had a small intro that allowed for a smooth transition between the topic of the research and the research itself. The scholarly article seemed to be written with the research in mind, and the reader secondary. It focused on the research very thoroughly and was not written to be read for entertainment. My summary was more like the pop culture article I think. I think I wrote my summary to tell the reader about the study, but also to be a little entertaining. It was more a strict summary of the research so that the reader could read my summary, and know what the study was about and the results of it. Three different approaches for the same study, but each with their own unique way of conveying the information across for different purposes.
Sources:
“Why Ask a Machine for a Recommendation?” Psychology Today, Sussex Publishers, http://www.psychologytoday.com/us/blog/choice-matters/201905/why-ask-machine-recommendation.
Yeomans, M., Shah, A., Mullainathan, S., & Kleinberg, J. (2019). Making sense of recommendations. Journal of Behavioral Decision Making, Advance online publication. https://doi.org/10.1002/bdm.2118