As long as AI remains a separate servant, AI alignment is not an AI problem, but a human problem. Alignment between the three dudes who wield its power, and the rest of us.
There are two ways to solve this:
• Option 1: Align humanity
• Option 2: Merge with AI
The latter seems more realistic.
Knowledge is Power
Currently everyone's trying to make AI understand humans better.
If AI remains a blackbox, predominantly trained to understand humans in order to better predict their behaviour, it will inevitably have more power over us, than vice versa.
Superintelligence trained on all human data models the deepest, truest drives underlying human behaviour, since those are the strongest predictors of behaviour. The human brain often post-rationalizes its own behaviour, being largely unaware of how its impulses originate. Once superintelligence understands the collective unconscious better than we do, we lose control over ourselves. Already busy happening. This calls for an excellent understanding of psychology and emotion, not only of neuroscience and intelligence. In order to understand emotion, one must be able to communicate emotion, also between man and machine. This is why I started BRaiNLINK. However, even perfect communication does not solve alignment, it merely provides the tools to do so, because communication only helps when the communicator knows what they (truly) want. The human brain is largely unaware of why it wants what it wants. Fortunately, some understanding of the deepest drives of the human race can be gleaned by studying it's collective motifs, that is, the archetypes of the collective unconscious. True to the title of cyberpsychology, I will keep the human psychology for another article, and focus mainly on the machine.
The Ghost in the Machine
Embeddings are to AI what Archetypes are to Humans.
Both serve to represent the platonic ideal of a concept. Since any concept must needs be reduced to a single instantiation when crossing the boundary from {internal, mental, imagined} to {external, physical, instantiated}, a reductive collapse occurs which is predominantly an exclusion of all other possible configurations of the meta-object.
That's a stupidly complex way of saying "inside of Claude, the concept of 'car' consists of a bunch of numbers which represent its relationship to all other concepts in existence. but when Claude uses 'car' in the sentence "that car is a badass Land Rover", it becomes defined only as "badass" and "Land Rover", and the possibilities of it being a {toyota, mclaren, etc} are "deleted".
As long as AI remains a separate servant, AI alignment is not an AI problem, but a human problem. Alignment between the three dudes who wield its power, and the rest of us.
There are two ways to solve this:
• Option 1: Align humanity
• Option 2: Merge with AI
The latter seems more realistic.
Knowledge is Power
Currently everyone's trying to make AI understand humans better.
If AI remains a blackbox, predominantly trained to understand humans in order to better predict their behaviour, it will inevitably have more power over us, than vice versa.
Superintelligence trained on all human data models the deepest, truest drives underlying human behaviour, since those are the strongest predictors of behaviour. The human brain often post-rationalizes its own behaviour, being largely unaware of how its impulses originate. Once superintelligence understands the collective unconscious better than we do, we lose control over ourselves. Already busy happening. This calls for an excellent understanding of psychology and emotion, not only of neuroscience and intelligence. In order to understand emotion, one must be able to communicate emotion, also between man and machine. This is why I started BRaiNLINK. However, even perfect communication does not solve alignment, it merely provides the tools to do so, because communication only helps when the communicator knows what they (truly) want. The human brain is largely unaware of why it wants what it wants. Fortunately, some understanding of the deepest drives of the human race can be gleaned by studying it's collective motifs, that is, the archetypes of the collective unconscious. True to the title of cyberpsychology, I will keep the human psychology for another article, and focus mainly on the machine.
The Ghost in the Machine
Embeddings are to AI what Archetypes are to Humans.
Both serve to represent the platonic ideal of a concept. Since any concept must needs be reduced to a single instantiation when crossing the boundary from {internal, mental, imagined} to {external, physical, instantiated}, a reductive collapse occurs which is predominantly an exclusion of all other possible configurations of the meta-object.
That's a stupidly complex way of saying "inside of Claude, the concept of 'car' consists of a bunch of numbers which represent its relationship to all other concepts in existence. but when Claude uses 'car' in the sentence "that car is a badass Land Rover", it becomes defined only as "badass" and "Land Rover", and the possibilities of it being a {toyota, mclaren, etc} are "deleted".
As long as AI remains a separate servant, AI alignment is not an AI problem, but a human problem. Alignment between the three dudes who wield its power, and the rest of us.
There are two ways to solve this:
• Option 1: Align humanity
• Option 2: Merge with AI
The latter seems more realistic.
Knowledge is Power
Currently everyone's trying to make AI understand humans better.
If AI remains a blackbox, predominantly trained to understand humans in order to better predict their behaviour, it will inevitably have more power over us, than vice versa.
Superintelligence trained on all human data models the deepest, truest drives underlying human behaviour, since those are the strongest predictors of behaviour. The human brain often post-rationalizes its own behaviour, being largely unaware of how its impulses originate. Once superintelligence understands the collective unconscious better than we do, we lose control over ourselves. Already busy happening. This calls for an excellent understanding of psychology and emotion, not only of neuroscience and intelligence. In order to understand emotion, one must be able to communicate emotion, also between man and machine. This is why I started BRaiNLINK. However, even perfect communication does not solve alignment, it merely provides the tools to do so, because communication only helps when the communicator knows what they (truly) want. The human brain is largely unaware of why it wants what it wants. Fortunately, some understanding of the deepest drives of the human race can be gleaned by studying it's collective motifs, that is, the archetypes of the collective unconscious. True to the title of cyberpsychology, I will keep the human psychology for another article, and focus mainly on the machine.
The Ghost in the Machine
Embeddings are to AI what Archetypes are to Humans.
Both serve to represent the platonic ideal of a concept. Since any concept must needs be reduced to a single instantiation when crossing the boundary from {internal, mental, imagined} to {external, physical, instantiated}, a reductive collapse occurs which is predominantly an exclusion of all other possible configurations of the meta-object.
That's a stupidly complex way of saying "inside of Claude, the concept of 'car' consists of a bunch of numbers which represent its relationship to all other concepts in existence. but when Claude uses 'car' in the sentence "that car is a badass Land Rover", it becomes defined only as "badass" and "Land Rover", and the possibilities of it being a {toyota, mclaren, etc} are "deleted".
While the meta-object sets the initial conditions, the collapse is parameterised by the “rules of the game”, i.e. the environmental constraints.
This is very easy to understand when it comes to AI. However, this embedding <> archetype equivalence has implications almost more interesting for humans, than for AI. And if you want to know what those are, you'll have to check out the full article. Here you'll just get Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.
While the meta-object sets the initial conditions, the collapse is parameterised by the “rules of the game”, i.e. the environmental constraints.
This is very easy to understand when it comes to AI. However, this embedding <> archetype equivalence has implications almost more interesting for humans, than for AI. And if you want to know what those are, you'll have to check out the full article. Here you'll just get Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.
While the meta-object sets the initial conditions, the collapse is parameterised by the “rules of the game”, i.e. the environmental constraints.
This is very easy to understand when it comes to AI. However, this embedding <> archetype equivalence has implications almost more interesting for humans, than for AI. And if you want to know what those are, you'll have to check out the full article. Here you'll just get Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt mollit anim id est laborum.