"ML promises to revolutionize 'X' because of the explosion of data in the modern era."
Outside of some singularity whack-jobs, that's always been the promise. The explosion of data in the field is a necessary requirement.
Healthcare fields make it nigh impossible to access data in way that will allow for fast prototyping or detailed experimentation. This isn't just about privacy either. Each Hospital treats even anonymized samples as a prospective source of income and a competitive advantage. I understand why they do it from a profit motive perspective, but it is certainly being traded off against prospective decreases in healthcare prices and significantly improved diagnostics.
ML revolutionized Vision because of Imagnenet and Coco. ML revolutionized Language when Google scraped the entire internet for BERT. Graph neural networks have started working now that they're being run on internet sized knowledge graphs. Even self-driving companies know that the key to autonomous diving mecca lies in the data and not the models. (Karpathy goes into intricate detail here during his talks)
If a field wishes to claim that ML has failed to revolutionize it, I would ask it to first meet the one requirement ML needs satisfied: Large-scale publically-ish available labelled data. The sad thing is that Healthcare is not incapable of providing this. It's just that the individual players do not want to cooperate to make it happen.
Outside of some singularity whack-jobs, that's always been the promise. The explosion of data in the field is a necessary requirement.
Healthcare fields make it nigh impossible to access data in way that will allow for fast prototyping or detailed experimentation. This isn't just about privacy either. Each Hospital treats even anonymized samples as a prospective source of income and a competitive advantage. I understand why they do it from a profit motive perspective, but it is certainly being traded off against prospective decreases in healthcare prices and significantly improved diagnostics.
ML revolutionized Vision because of Imagnenet and Coco. ML revolutionized Language when Google scraped the entire internet for BERT. Graph neural networks have started working now that they're being run on internet sized knowledge graphs. Even self-driving companies know that the key to autonomous diving mecca lies in the data and not the models. (Karpathy goes into intricate detail here during his talks)
If a field wishes to claim that ML has failed to revolutionize it, I would ask it to first meet the one requirement ML needs satisfied: Large-scale publically-ish available labelled data. The sad thing is that Healthcare is not incapable of providing this. It's just that the individual players do not want to cooperate to make it happen.