Michael Hoffman
101d ago
1/Now in @nature: our Matters Arising letter describing how the @GoogleHealth study claiming an AI system can outperform radiologists at predicting cancer violates basic scientific standards of transparency and reproducibility. @bhaibeka @hugo_aerts et al.
9
We said some things recently, and they got published. I just want to highlight the point that while there are important privacy implications around sharing models and data, we need to expect that source code will be shared.
nature.com/articles/s4158β¦ pic.twitter.com/xAtfc4xEV5
nice piece on (lack of) transparency and reproducibility in a recent artificial intelligence paper
"We have high hopes for the utility of AI methods in medicine. Ensuring that these methods meet their potential, however, requires that these studies be scientifically reproducible"
from "Transparency and reproducibility in artificial intelligence"
A DL-based breast cancer screening method was published in Nature but w/o code.
An objection to this cites transparency & reproducibility: nature.com/articles/s4158β¦
The authors' response cites potential for abuse: nature.com/articles/s4158β¦
Should the authors release their code?
It remains bizarre to me that it is possible to publish a deep learning paper in a prestigious journal without releasing code
Transparency and reproducibility in artificial intelligence - Nature nature.com/articles/s4158β¦ #artificialintelligence #machinelearning #informatics #bioinformatics
1/3 Important and thoughtful debate on transparency and reproducibility in artificial intelligence. Letter at
AI research needs more transparency and reproducibility:
"Journals have an obligation to hold authors to the standards of reproducibility that benefit not only other researchers, but also the authors themselves."
@DigEconLab
@StanfordHAI
#AI
