How to discover the nature of sentience, and ethics

post by Gustavo Ramires (gustavo-ramires) · 2024-09-11T17:22:46.076Z · LW · GW · 4 comments

Contents

      Note: This is a very rough draft of a theory. Any comments or corrections are welcome. I will try to give credit to useful observations.
    How can we do that?
  What about ethics?
    Is this practical?
  Risks
None
4 comments

Note: This is a very rough draft of a theory. Any comments or corrections are welcome. I will try to give credit to useful observations.

To discover the nature of sentience and ethics, we can actually use scientific principles even though they are sometimes considered metaphysical.

The great insight though is that consciousness is part of reality. It is a real phenomenon. So I believe it turns out we can study this phenomenon, and assemble essentially complete theories of what it 'is'.

What do I mean by a theory of consciousness (later a theory of ethics, or 'what we must do')?

A theory of physics may, for example, predict the position of particles at a later time given certain initial conditions.

A theory of consciousness can answer a number of questions:
- Given a mind (e.g. a human brain), and its initial state, when we give it a certain input, for example corresponding to the physiological (neural) response to smelling a sweet, what taste does the mind/person feel? (is it really sweet or is this person different and smells what is commonly known as salty?) This is a purely subjective phenomenon, but it is emergent from information and communication in the brain.

In general, we are able to given neural activity describe the internal feelings experienced by the person using common sense words, because we have a theory of neural activity, and a theory of how different feelings manifest neurally in relationship to associated well-enough defined words and concepts (for example emotions, feelings, sensations).

- We understand all qualia and what qualia 'is', that is, we can tell if a certain object with say certain electrical activity manifests qualia (i.e. if it is sentient) or not, for any system.

How can we do that?

The basic principle really is observation of what people say, including with some caveats personal observations.

The second principle is that if there is a singular reality, and if things are a certain way and not another, then it may be (hypothesis) that there should be a single theory that is both logically internally consistent and consistent with reality, that explains both physics and sentience: simply because sentience is an emergent phenomenon hence part of reality. (in particular, there could perhaps be multiple theories giving the same predictions thus being equivalent). This may also be known as an 'principle of explosion'
from mathematics.

The third principle is that of regularity. For example, say for example, in the case of the sweet smell given previously. Say we collect a large sample of people smelling sweets and  salts (could be a blind experiment for the researchers) and their neural patterns. Then we ask people what they tasted, and try to find correlations between the
neural activity and what people reported.

This allows us to circumvent a problem: what if people lie, don't understand or somehow misreport what they feel? Suppose we take a sample with 50 people, 45 people tell the truth and 5 lie. We may be able to detect the outlier case and just tell there is something different about those 5 people. In a statistical model, if a fixed portion of people (say 10%) lie about what they tasted, then we can use the law of large numbers
to not only have a probabilistic notion of what is sweet and isn't, but it may actually converge to an exact model of what is sweet and isn't. Beyond a simple correlation, we should try to find general and simple functions that predict (exactly or approximately) those things like taste while 'understanding' their nature. The theory again of what is the taste known as sweet again, with enough observations, will converge in a singular theory (or set of equally predictive ones).

By understanding I mean we will have a map of what exactly are the necessary conditions, variables and structures for the manifestation of the subjective phenomenon known as tasting sweet.

What about ethics?

By ethics I mean, 'what we *should* do', a general normative, ideal theory of what should be done, what is good, what is purpose, what is meaning.

Because we are all again emergent phenomena of the cosmos, emergent minds from neural structures, and a feature of reality, I think a first observation should be:

- What should be done is a function of sentience, exclusively, because *we are* sentience, we are emergent phenomena of human brains.

I can declare this because to say otherwise is inconsistent, inconsistent with the definition of sentience and also inconsistent with observations of reality.

I hypothesize a curious phenomenon: as we study sentience,  and as soon as we establish with varying rigor what we mean by 'what should be done' (using the principles outlined
in the previous section about sentience), then again a singular theory should emerge, a kind of emergent normativity, about what is ethical, what is meaning, what should be done.

An example is for example take some idea of 'literal physical constructivism': suppose we think the meaning of life is to build structures throughout the cosmos, literal physical structures, and the more we build the better.

But if meaning is about sentience as argued, then it cannot be: consider an universe of non-sentient robots building out large structures. It cannot be the meaning to bring about this universe, because in it there is no sentience. No one can for example experience things that caracterize sentience, therefore there cannot be meaning.

As we simultaneously understand, formalize and build theories of the questions we are interested in, and even theories of all plausible questions, and as we use the principles of singular reality and regularity, we may approach or converge toward notion of truth. In particular, the truths that are most important (ethics), and even the notion
of importance can be understood this way.

Is this practical?

I think so. Simply because simple insights such as that meaning must be a product of sentience seem already very useful. What use is money without a mind to use it and enjoy it? What use is power or wealth, if not just instrumental?

There are many other metaphysical questions that I believe can be answered more or less definitely[1] that are also very practical and very significant. This is only a sketch of this general  plan but I intend to elaborate further in the future. This procedure itself can be meta-improved with the initial (bootstrapping) principles outlined in this article. It is really a key to understanding anything physical or metaphysical, by slightly extrapolation scientific principles to subjective and metaphysical phenomena; it is of course full of hypotheses and if only very weak versions of claim here pan out, I still think we can find out much with good reasons to be confident.

Finally, we may use this to improve our lives and lower the level of conflict in society as a whole.

In particular, I think AI may be of great help analyzing huge quantities of data if we wish  to understand the nature of various qualia, including the 'postive' ones like joy, awe, contentment, etc. Also likely helpful in the effort to theorize, formalize and find inconsistencies in hypotheses.

Risks

Every theory carries some risks and it's important to be aware of them, specially in this case I think. I think the risks posed by this kind of theory largely relate to the crystalization associated with theory-building, and possible associated formalizations and formal use of words including common-sense words. So we could formalize for example a notion of joy and use those words within the context of their theories of joy; but the theory will either likely be significantly incomplete, or the daily usage of the word within society might shift its meaning (hopefully for the better), and over-reliance on some formal system could create a prison where changing is disallowed or discouraged (for example if some reasoning systems, chatGPT-like, that we interact with adopted formal definitions). The defense against it are a few (non-exclusionary): 

(1) Make such systems flexible in that definitions can be changed to match common usage; 

(2) Avoid overreliance or careless usage of formal inflexible systems; 

(3) Create cultural understanding that formal definitions and theories constitute a "closed world", much like some words have specific meanings in science that may not correspond with common-sense notions and intuitions, creating a separation between the two (formalized meanings and daily-usage meanings).

[1] That is, asymptotically definitely perhaps.

4 comments

Comments sorted by top scores.