Today I am happy to announce a new milestone. As of today I have received from my manufacturer the authorization code to initiate semi-closed loop mode on my life support devices. This means that for the first time, my life support devices are capable of keeping me alive for short periods without immediate direct human intervention. For the first time in more than a decade, it is now safe for me to be distracted by such luxuries as homework, and sleep. At least, for short periods, assuming everything works within normal parameters.
Okay, yes, this is a very qualified statement. Compared to the kind of developments which are daily promised by fundraising groups and starry eyed researchers, this is severely underwhelming. Even compared solely to technologies which have already proven themselves in other fields and small scale testing, the product which is now being rolled out is rather pathetic. There are many reasons for this, from the risk-aversiveness of industry movers, to the glacial pace of regulatory shakers, to a general shortage of imagination among decision makers. It is easy to find reasons to be angry and feel betrayed that the US healthcare system has once again failed to live up to its promise of delivering breakneck innovation and improvement.
Even though this is disappointing compared to the technological relief we were marketed, I am still excited about this development. First of all, because it is a step in the right direction, even if a small one, and any improvement is worth celebrating. Secondly, and chiefly, because I believe that even if this particular new product is only an incremental improvement over the status quo, and pales in comparison to what had been promised for the past several decades, the particular changes represent the beginning of a larger shift. After all, this is the first iteration of this kind of life support device which uses machine learning, not merely to enable a fail-safe to prevent medication overdoses, but which actually intends to make proactive treatment decisions without human oversight.
True, the parameters for this decision making are remarkably conservative, some argue to the point of uselessness. The software will not deploy under anything short of perfect circumstances, its treatment targets are short of most clinical targets, let alone best practices, the modeling is not self-correcting, and the software can not interpret human intervention and is therefore mutually exclusive with aggressive treatment by a human.
Crucially, however, it is making decisions instead of a human. We are over the hill on this development. Critiques of its decision-making skill can be addressed down the line, and I expect once the data is in, it will be a far easier approval and rollout process than the initial version. But unless some new hurdle appears, as of now we are on the path towards full automation.
