The Doctor and the Algorithm by Graham S. Scott;

Author:Graham, S. Scott;
Language: eng
Format: epub
Publisher: Oxford University Press, Incorporated
Published: 2022-08-15T00:00:00+00:00

Figure 5.3. Shared control estimation plot for promotional percent and ground truth approach.

However, there were many examples of forged ground truth in the Training Set used for HypeDx. The 82 articles were 7.32% dredged, 53.7% engineered, and 39.0% forged. Based on the percentage of promotional language and the highest reported AUC scores, 43.18% of engineered articles and 68.75% of forged articles were in the zone for high risk of hype. This difference is statistically significant.40 At the same time, the average max AUC of forged articles was notably higher than those for engineered articles with a mean difference of 0.0632.41 See Figure 5.4 for more details. So forged articles were both more likely to return a higher AUC and more likely to be in the high-risk zone, making them especially dangerous. This follows from concerns in the last chapter about the bearing capacity limitations of forged ground truth. It’s simply much easier to effect the appearance of a solid foundation for model training when the ground truth is forged rather than dredged or engineered. The good news is that forged groundtruthing does not appear to be all that common in the random sample of articles. However, it is overrepresented in Lieu et al.’s meta-study. If meta-studies evaluating the overall performance of AI in health are built significantly on analysis of articles that lack firm bearing capacity, then this raises concerns for the quality of health AI meta-discourse.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.