How Important is that Machine Learning Model be Understandable? We analyze poll results
About 85% of respondents said it was always or frequently important that Machine Learning model be understandable. This was is especially important for academic researchers, and surprisingly more in US/Canada than in Europe or Asia.
The previous KDnuggets Poll asked
The overall results, based on over 500 votes, show that for about 37% of voters, it was always important, and another 48% said it was frequently important, so for about 85% of respondents it was always or frequently important. Only 2% said it was never important.
Fig. 1: How often was it important that Machine Learning Model be Understandable
The poll also asked about employment type, and overall breakdown here was
Fig. 2: How often was it important that Machine Learning Model be Understandable, by employment
Color indicates importance: orange: "Always", green: "Frequently", light grey: "Rarely", dark grey: "Never".
We note that respondents working for company/self had the highest number of "frequently" answers - 51.6%, while students had the lowest - only 36.7%.
Academic researchers said "Always" more than any other group, suggesting that Machine Learning understanding is an active area of research.
However, overall, understandability was always or frequently important for all groups over 80% of the time. Not surprisingly, the only exception were students, for where it was a little less - only 76%.
The overall breakdown by region was
Is understanding of Machine Learning especially important in Europe, given that GDPR went into effect in there on May 25, 2018?
Fig. 3: How often it is important that Machine Learning Model be Understandable, by region
Surprisingly, the region with the highest level of concern for understandability is not Europe but US/Canada. Combining "Always" and "Frequently" answers, we see that US/Canada respondents had the highest concern, with 88.5%, followed by Europe: 81.9%, Asia: 81.0%, and Latin America: 78.2%. The number of respondents in the other 2 regions is too small for a statistical analysis.
Related:
When building Machine Learning / Data Science models in 2018, how often was it important that the model be humanly understandable/explainable?
The overall results, based on over 500 votes, show that for about 37% of voters, it was always important, and another 48% said it was frequently important, so for about 85% of respondents it was always or frequently important. Only 2% said it was never important.
Fig. 1: How often was it important that Machine Learning Model be Understandable
The poll also asked about employment type, and overall breakdown here was
- Company or self, 69%
- Government/ non-profit, 5%
- Student, 15%
- Academia/ University 9%
- Other, 2%
Fig. 2: How often was it important that Machine Learning Model be Understandable, by employment
Color indicates importance: orange: "Always", green: "Frequently", light grey: "Rarely", dark grey: "Never".
We note that respondents working for company/self had the highest number of "frequently" answers - 51.6%, while students had the lowest - only 36.7%.
Academic researchers said "Always" more than any other group, suggesting that Machine Learning understanding is an active area of research.
However, overall, understandability was always or frequently important for all groups over 80% of the time. Not surprisingly, the only exception were students, for where it was a little less - only 76%.
The overall breakdown by region was
- US/Canada, 36%
- Europe, 34%
- Asia, 18%
- Latin America, 6.0%
- Africa/Middle East, 3.7%
- Australia/NZ, 2.8%
Is understanding of Machine Learning especially important in Europe, given that GDPR went into effect in there on May 25, 2018?
Fig. 3: How often it is important that Machine Learning Model be Understandable, by region
Surprisingly, the region with the highest level of concern for understandability is not Europe but US/Canada. Combining "Always" and "Frequently" answers, we see that US/Canada respondents had the highest concern, with 88.5%, followed by Europe: 81.9%, Asia: 81.0%, and Latin America: 78.2%. The number of respondents in the other 2 regions is too small for a statistical analysis.
Related:
- Using Uncertainty to Interpret your Model
- Holy Grail of AI for Enterprise — Explainable AI
- Will GDPR Make Machine Learning Illegal?
