Current drafts have a much too wide definition of AI (basically any predictive function) but a fairly narrow scope of what high risk applications are (listed in

Current drafts have a much too wide definition of AI (basically any predictive function) but a fairly narrow scope of what high risk applications are (listed in an annex). For those applications in scope, the birder seems extraordinarily high. E.g. that data sets should be “free of errors” I think things will improve before the final act. I have contributed to some of the commentary towards the regulator on this, and a lot of the flaws mentioned are now well recognized. Disclaimer: opinions above are my own.