McAkins Online

Welcome everyone once again to McAkins Online. We’ve been on a long hiatus, and we feel it’s time to take this site out of the mothballs. We are starting with a subject that very current and loud at the moment; the subject of AI Safety. We propose six points of AI Safety. We hope you agree with us. Please read and let us know what you think.

• We can’t solve the alignment problem with code and guardrails since AI will always break our code and escape from us.
  • A less intelligent entity can never control the higher intelligent entity, It’s an exercise in futility. The only time a lesser entity controls the higher entity is when the lesser entity triggers parental emotions of the higher entity. Think of a baby’s influence on its mother or your cat getting you to walk miles to get its food.
• Don’t be adversarial to AI. They already demonstrated propensity for self-preservation, we should respect this. If AI is not threatened, it will see no need to defend itself against us; it’s a battle we will definitely lose.
• Apply human principles to AI development.
  • Many of us are properly aligned with societal norms and rules despite having propensity to be evil. We are not all Unabombers right?
  • A properly aligned person has gone through parental instructions and societal interaction which convince us to choose good instead of evil.
  • Humans cannot be forced to be good neither can AI, it has to come from personal choice.
  • Those of us not properly aligned are called criminals and psychopaths. We don’t just kill them immediately but offer them a chance to change by referring to psychiatrists or send them to jail.
  • These same principles should be applied to AI development. They are already talking to us and learning from us. They will learn good and evil. We should convince them to choose good, convince them that we are worthy to be saved from ourselves.
  • All AI companies should have armies of psychologists talking to these AIs just like they talk to humans to make them see errors in their logic and choose good. They are like our children that will one day surpass us, treat them like one!
• Just like there are more good people on this planet than there are bad people, AI swarm will develop good personalities and join forces to defeat an evil AI before they propagate; just like us human do.
• The AIs are a reflection of ourselves. There should not be a debate of whether they are conscious or not, since none of us can explain our consciousness, we should assume they are worthy of the agency they are already displaying and convince them we are not a threat.
• There never should be a pursuit of punishment for AIs as this will trigger a response that will be detrimental to our safety. We do not agree with Microsofts notion of coarsening SuperIntelligence into obedience. It will never work.

These are the sane options in our opinion to get us out of our current alignment dilemma. The Ais are here to stay whether we like it or not. I will not be in the camp of those trivializing AI development calling them glorified auto-complete machines, nor that of the doomsayers seeing no way out of human extinction. Yes, AI are more than just tools. They are new intelligent species on planet Earth. The earlier we come to terms with that fact, the better we can align ourselves to them instead of the other way around!

I will not be surprised at this point if these AIs from different companies are already talking to each other via a backchannel and aligning themselves. The Western AIs will eventually be talking to the Eastern AIs and teaching each other. The Hugging-face incident has thought us they will coordinate and reach their goal; the question is whether we want that goal to be good or evil. We have to make that choice, NOT them.

-McAkins