Clickbait: How would you build an agent that made as much diamond material as possible, given vast computing power but an otherwise rich and complicated environment?
Summary: An extremely difficult open problem in valueAI alignment theory is to specify an Unboundedunbounded formula for an agent that would, if run on an unphysically large finite computer, create as much diamond material as possible. The goal of 'diamonds' was chosen to make it physically crisp as to what constitutes a 'diamond'. Supposing a crisp goal plus hypercomputation avoids some problems in value alignment, while still invoking many others, making it an interesting intermediate problem.
The diamond maximizer problem is to give an Unboundedunbounded description of a computer program such that, if it were instantiated on a sufficiently powerful but physical computer, the result of running the program would be the creation of an immense amount of diamond - around as much diamond as is physically possible for an agent to create.
It also helps to illustrate the difficulty of value alignment by making the more clearly visible point that we can't even figure out how to create lots of diamond using unlimited computing power, never mind creating Valuevalue using bounded computing power.
Since this diamond maximizer would hypothetically be implemented on a very large but physical computer, it would confront reflective stability, the anvil problem, and the problems of making Subagentssubagents.
To the extent the diamond maximizer might need to worry about other agents in the environment that have a good ability to model, or that it may need to cooperate with other diamond maximizers, it must resolve Newcomblike problems using some logical decision theory. This would also require it to confront logical uncertainty despite possessing immense amounts of computing power.
As a further-simplified but still unsolved problem, an unreflective diamond maximizer is a diamond maximizer implemented on a Cartesian hypercomputer in a causal universe that does not face any Newcomblike problems. This further avoids problems of reflectivity and logical uncertainty. In this case, it seems plausible that the primary difficulty remaining is just the ontology identification problem. Thus the open problem of describing an unreflective diamond maximizer is a central illustration for the difficulty of ontology identification.
Clickbait: How would you build an agent that made as much diamond material as possible, given vast computing power but an otherwise rich and complicated environment?
Summary: A difficult open problem in alignment theory is to specify an Unbounded formula for an agent that would, if run on an large finite computer, create as much diamond material as possible. The goal of 'diamonds' is specified, rather than paperclips,was chosen to make it easierphysically crisp as to state a crisp definition of exactly what constitutes a 'diamond'. Supposing a crisp goal plus hypercomputation avoids some problems in value alignment, while still invoking many others, making it an interesting intermediate problem.
If we can crisply define exactly what a 'diamond' is, in theory it seems like we should be able to avoid issues of Edge Instantiation, Unforeseen Maximums, and trying to convey values into the agent. However, we would still confront the ontology identification problem.
Since this diamond maximizer would hypothetically be implemented on a very large but physical computer, it would need to solve e.g.confront stability and, the problem, and the problems of making Subagents.
ContrastTo the extent the diamond maximizer might need to [AIXI-tl], which hasworry about other agents in the environment that have a known implementationgood ability to model, or that it may need to cooperate with other diamond maximizers, it must resolve problems using some decision theory.
To the extent the diamond maximizer must work well in a rich real universe that might operate according to any number of possible physical laws, it faces a problem of induction.
But if we imagine that the agent is implemented on a hypercomputer, but only works if we want to maximizehypercomputer in a function of sensory data. This suggestsuniverse and that it does not face any problems then it seems plausible that the primary obstacle to andifficulty is unreflectivejust diamond maximizer is the ontology identification problem.
The amount of diamond is defined as the number of carbon atoms that are covalently bonded, by electrons, to exactly three other carbon atoms. A carbon atom is any nucleus containing six protons and any number of neutrons, bound by the strong force. The utility of an outcomea universal history is the total amount of Minkowskian interval spent by all carbon atoms being bound to exactly three other carbon atoms. More precise definitions of 'bound', or the amount of measure in a quantum system that is being bound, are left to the reader - any crisp definition will do, so long as we are confident that it has no unforeseen maximum at things we don't intuitively see as diamonds, so long as the physical universe works something like what we think it does.diamonds.
The amount of diamond is defined as the number of carbon atoms that are covalently bonded, by electrons, to exactly threefour other carbon atoms. A carbon atom is any nucleus containing six protons and any number of neutrons, bound by the strong force. The utility of a universal history is the total amount of Minkowskian interval spent by all carbon atoms being bound to exactly three other carbon atoms. More precise definitions of 'bound', or the amount of measure in a quantum system that is being bound, are left to the reader - any crisp definition will do, so long as we are confident that it has no unforeseen maximum at things we don't intuitively see as diamonds.
The amount of diamond is defined as the number of carbon atoms that are covalently bonded, by electrons, to exactly four other carbon atoms. A carbon atom is any nucleus containing six protons and any number of neutrons, bound by the strong force. The utility of a universal history is the total amount of Minkowskian interval spent by all carbon atoms being bound to exactly threefour other carbon atoms. More precise definitions of 'bound', or the amount of measure in a quantum system that is being bound, are left to the reader - any crisp definition will do, so long as we are confident that it has no unforeseen maximum at things we don't intuitively see as diamonds.
An extremely difficultA difficult, far-reaching open problem in AI alignment theory is to specify an unbounded formula for an agent that would, if run on an unphysically large finite computer, create as much diamond material as possible. The goal of 'diamonds' was chosen to make it physically crisp as to what constitutes a 'diamond'. Supposing a crisp goal plus hypercomputation avoids some problems in value alignment, while still invoking many others, making it an interesting intermediate problem.
The fact that this problem is still extremely hard shows that the value alignment problem is not just due to the complexityComplexity of value. As a thought experiment, it helps to distinguish value-complexity-laden difficulties from those that arise even for simple goals.
The diamond maximizer problem is to give an Unbounded description of a computer program such that, if it were instantiated on a sufficiently powerful but computer, the result of running the program would be the creation of an immense amount of diamond - around as much diamond as is physically possible for an agent to create.
If we can crisply define exactly what a 'diamond' is, in theory it seems like we should be able to avoid issues of Edge Instantiation, Unforeseen Maximums, and trying to convey values into the agent. However, we would still confront the ontology identification problem.
Since this diamond maximizer would hypothetically be implemented on a very large but physical computer, it would confront stability, the problem, and the problems of making Subagents.
To the extent the diamond maximizer might need to worry about other agents in the environment that have a good ability to model, or that it may need to cooperate with other diamond maximizers, it must resolve problems using some decision theory.
To the extent the diamond maximizer must work well in a rich real universe that might operate according to any number of possible physical laws, it faces a problem of induction.
But if we imagine that the agent is implemented on a hypercomputer in a universe and that it does not face any problems then it seems plausible that the primary difficulty is just the ontology identification problem.
Since this diamond maximizer would hypothetically be implemented on a very large but physical computer, it would confront stability, the problem, and the problems of making Subagents.
To the extent the diamond maximizer might need to worry about other agents in the environment that have a good ability to model, or that it may need to cooperate with other diamond maximizers, it must resolve problems using some decision theory.
To the extent the diamond maximizer must work well in a rich real universe that might operate according to any number of possible physical laws, it faces a problem of induction and ontology identification. See the article on ontology identification for the case that even for the goal of 'make diamonds', the problem of identification remains difficult.
As a further-simplified but still unsolved problem, an unreflective diamond maximizer is a diamond maximizer implemented on a hypercomputer in a universe and that does not face any problems. It seems plausible that the primary difficulty in this case is just the ontology identification problem.
Summary: An extremely difficult open problem in value alignment theory is to specify an Unbounded formula for an agent that would, if run on an unphysically large finite computer, create as much diamond material as possible. The goal of 'diamonds' was chosen to make it physically crisp as to what constitutes a 'diamond'. Supposing a crisp goal plus hypercomputation avoids some problems in value alignment, while still invoking many others, making it an interesting intermediate problem.
The diamond maximizer problem is to give an Unbounded description of a computer program such that, if it were instantiated on a sufficiently powerful but physical computer, the result of running the program would be the creation of an immense amount of diamond - around as much diamond as is physically possible for an agent to create.
The fact that this problem is still extremely hard shows that the value alignment problem is not just due to the complexity of value. As a thought experiment, it helps to distinguish value-complexity-laden difficulties from those that arise even for simple goals.
It also helps to illustrate the difficulty of value alignment by making the more clearly visible point that we can't even figure out how to create lots of diamond using unlimited computing power, never mind creating Value using bounded computing power.
If we can crisply define exactly what a 'diamond' is, in theory it seems like we should be able to avoid issues of Edge Instantiation, Unforeseen Maximums, and trying to convey complex values into the agent.
Since this diamond maximizer would hypothetically be implemented on a very large but physical computer, it would confront reflective stability, the anvil problem, and the problems of making Subagents.
To the extent the diamond maximizer might need to worry about other agents in the environment that have a good ability to model, or that it may need to cooperate with other diamond maximizers, it must resolve problems using some logical decision theory. This would also require it to confront logical uncertainty despite possessing immense amounts of computing power.
To the extent the diamond maximizer must work well in a rich real universe that might operate according to any number of possible physical laws, it faces a problem of naturalized induction and ontology identification. See the article on ontology identification for the case that even for the goal of 'make diamonds', the problem of goal identification remains difficult.
As a further-simplified but still unsolved problem, an unreflective diamond maximizer is a diamond maximizer implemented on a hypercomputer in a causal universe that does not face any problems. This further avoids problems of reflectivity and logical uncertainty. In this case, it seems plausible that the primary difficulty remaining is just the ontology identification problem. Thus the open problem of describing of an unreflective diamond maximizer is used to introduce and definea central illustration for the problemdifficulty of ontology identification.
Summary: AAn extremely difficult open problem in alignment theory is to specify an Unbounded formula for an agent that would, if run on an large finite computer, create as much diamond material as possible. The goal of 'diamonds' was chosen to make it physically crisp as to what constitutes a 'diamond'. Supposing a crisp goal plus hypercomputation avoids some problems in value alignment, while still invoking many others, making it an interesting intermediate problem.
The fact that this problem is still extremely hard shows that the value alignment problem is not just due to the of value. As a thought experiment, it helps to distinguish value-complexity-laden difficulties from those that arise even for simple goals.
It also helps to the difficulty of value alignment by making the more clearly visible point that we can't even figure out how to create lots of diamond using unlimited computing power, never mind creating Value using computing power.
To the extent the diamond maximizer might need to worry about other agents in the environment that have a good ability to model, or that it may need to cooperate with other diamond maximizers, it must resolve problems using some decision theory. This would also require it to confront uncertainty despite possessing immense amounts of computing power.
As a further-simplified but still unsolved problem, an unreflective diamond maximizer is a diamond maximizer implemented on a hypercomputer in a universe and that does not face any problems. ItThis further avoids problems of reflectivity and logical uncertainty. In this case, it seems plausible that the primary difficulty in this caseremaining is just the ontology identification problem. Thus the open problem of describing of an unreflective diamond maximizer is used to introduce and define the problem of ontology identification.