I have recently read In that article, the traditional non-embedded model has many similarities with PBR. For those interested, t
Quick Review of PBR
This is just a quick review of PBR’s claims. If you are familiar with them feel free to skip this section.
1. Perspective-based reasoning is more fundamental than objective reasoning.
It just means thinking from a perspective (e.g. your first-person perspective) is more basic than thinking perspective-independently. Objective reasoning is developed from perspective reasoning.
2. Perspective is primitive.
Perspective is a reasoning starting point. Like an axiom, it can only be regarded as given, not further analyzed. This is why in anthropics, self-locating probabilities such as “I am this particular person” or “Today is this particular day” are meaningless. Also why the Copenhagen Interpretations, in which there exists a physically irreducible “observer”, is most likely the correct way to think about quantum mechanics.
Also worth pointing out that there can only be one perspective in any reasoning process. An example would be from my perspective “I am a man” is true, whereas from my wife’s perspective “I am not a man” is true. Mixing statements from these two perspectives together would generate inconsistency. It is also the reason for perspective disagreement in anthropics.
Characteristics of Non-Embedded Agency
1. A dualistic agent exists outside of its environment. It primitively carves the world into “agent” and “environment” with an input-output relationship. This is mirrored in PBR regarding the perspective (self) as primitively defined. The input-output relation in the AI context is mirrored by our perception and subjective experience as highlighted on the homepage.
2. Dualists agents don’t tend to model themselves, because the agent is made of different stuff than what the agent reasons about. Figuratively they “can treat himself as an unchanging indivisible atom”. PBR does not self-analyze because the perspective center is considered primitively given. It is why “self-locating probabilities” are invalid. Also why PBR supports the Copenhagen interpretation of quantum mechanics, in which the “observer” is not physically explainable by reductionism is to be expected.
3. Dualistic agents assume a particular way of carving up the world and don’t allow for switching between different carvings. (
In Defence of Dualist Agency
Given I strongly support PBR, I also think the non-embedded AI model is correct. I am not familiar with the AI field (which is also why I have only recently come across the related articles). However, I do think PBR could help to explain some of the supposed shortcomings of the non-embedded model.
It feels like the dualistic agency is an idealist simplification while the embedded agency reflects the real world. I think people feel this way because the embedded agency reflects objective reasoning: it does not consider the self as special while dualistic agency (and PBR) seem to endorse it. This is actually not true. While reasoning from a perspective would consider the “self” as primitively unique, it does not mean this particular perspective is in any way special. Other perspectives are just as valid as mine. And from those perspectives, their “selves” are also primitively unique in each of their respective reasonings. Everyone is special to themselves, yet no one’s perspective is special. (Checkout Consciousness, Free Will and Scientific Objectivity)
Another problem of the dualist agency is that it cannot self-analyze thus cannot self-improve. It is technically true that self-analyze is not possible under PBR. However, as long as one can imagine (or simulate) reasoning from a different perspective then it can analyze itself. Since from a different perspective, the original agent is no longer the “self”. The same idea has been discussed in quantum interpretations: all physical systems are analyzable, it just cannot be done from its own perspective. i.e. we simply need to assume a different “observer” and conduct the analysis from there. For programs, the outsider perspective can be achieved by copying oneself then analyzing and modifying the new copy.
This leads to another point: thinking about what kind of decision I should make and reductively analyzing what decision I would make (e.g. by physically analyzing the brain activities or studying the source code) are two different processes from distinct perspectives. The former is reasoning from the first-person perspective: consider the possible actions I could take, evaluate them by their outcomes, and choose the one deemed optimal. While the latter is reasoning from the perspective of an outsider: treat the agent like a machine, reductively figuring out how it functions, and derive the output it would produce. These two parts should not be used together in the same reasoning process or it could lead to paradoxes. (I feel this deserves more elaboration. I will discuss it in detail in future posts)
Bottomline
I think the non-embedded model would produce the same solution to anthropic paradoxes as PBR does. Reversely, we should recognize our logic and reasoning are based on the same non-embedded model as an AI. To say the very least, this agency study and anthropic paradoxes are deeply connected. For example, one open problem of embedded agency is about worlds that include multiple copies of the agent. The same problem repeatedly appears in anthropic arguments, which itself is an extension of the reference class problem. Finally given I support PBR, I would guess the traditional non-embedded agency would be the better model in the AI field as well.