Reminds me of "Dolly" the genetically engineered sheep back in 1996 who was cloned from a single adult somatic cell. At the time it was a fearsome break through in technology, but the theory that people could be cloned in a similar fashion just did not happen, largely due to technological limitations. It has given rise to some cell therapies which have only proven to be marginally successfull for all the $$ dumped in to them and the market hype but we are not yet at a place where just whole organs can be generated from a single cell, never mind a entirely cloned human body.

Whether it's machine learning or polymerase chain reaction in the end being employed to create protiens, it is all still information - sophisticated information to be sure - but still just information as a strand. There are 3-d crosslinkages and linear internal bondings which are present most often that define the spatial necessity of the molecules which in turn become proteins

If information integrity in-process cannot be perfectly assured errors will result in failures, not new and improved organisms (or mehanisms created through machine learning of any kind.) It is why Darwinian evolution by random chance is such a deeply flawed concept.