Strongly disagree. ML is traditionally deployed in low-risk scenarios like image labeling or recommendations where 1% or 5% error rate is totally "routine" and

Strongly disagree. ML is traditionally deployed in low-risk scenarios like image labeling or recommendations where 1% or 5% error rate is totally "routine" and acceptable. Doesn't mean this kind of ML pipeline can translate to cases where errors are literally fatal.