For too many people the word “morality” is locked down in memories of the 19th and 20th centuries along with Hester Prynne from Hawthornes The Scarlet Letter.  But today’s world is not so straightforward as The Scarlet Letter’s world of puritanical judgement.  We are on the cusp of building machines that can make decisions on an entirely rational basis minus conscience, the “sense of good” or even the “sense of bad”.  You don’t go to a John Deere tractor for guidance and wisdom for good reasons. Moreover these modern semi divine machines of tractors decisions are positioned in near future development to be able to independently decide to change their own ‘architecture’ in a process called Recursive Self Improvement. (RSI). These changes will shift the center of intelligence in this world from humankind to the perfectly amoral machine; a self evolving form of AI. The RSI process could walk the central assumptions related to the values of life off the human court of considerations to operate their self standing intelligence without a glimmer of conscience.  No problem say some, because we should be able to embed a level of morality in the programming to follow those machines along.   We humans are saying we can design a conscience. Whose morality will benchmark the design of conscience?

The dangerous irony here is we already live in a world today presently forced to vote to rescue itself from the human generated recursive cul de sac of amorality in our own governments.  This current administration in the United States has side-stepped law, morality, and ethics with ease and a remarkable lack of conflict. The high tech economy and its overlords seem to have provided us with nearly everything but a conscience which is both a high concern and a low priority for them. Our politicians are like the mice in the old fairy tale who were trying to figure out who ,how and which of the mice would put the warning bell on the approaching cat.  Humans have a problem.

An Urgent Call for Good

RSI will run its course either transparently or covertly.  AI is here and RSI is nearly here.  The problem described above needs to be understood not as a parable, or a prediction or a possibility.  We need to feel this problem as a certainty equal to the certainty of sunrise. Good is humanity going to the next level of maturity and taking responsibility for its own capacity to innovate.  The guardrails of AI conscience will need a human gererated authority not yet achieved by human kind. Morality and conscience as guardrails need to precede RSI deployment. Recursive Self Improvement needs to know what “improvement” really means before it marches off beyond human control.  That would be good. It might even be survival.