# Ω 13

Crossposted from the AI Alignment Forum. May contain more technical jargon than usual.

Status: a brief distillation of Wittgenstein's book On Certainty, using examples from deep learning and GOFAI, plus discussion of AI alignment and interpretability.

"That is to say, the questions that we raise and our doubts depend on the fact that some propositions are exempt from doubt, are as it were like hinges on which those turn."

— Ludwig Wittgenstein, On Certainty

# 1. Deep Learning

Suppose we want a neural network to detect whether two children are siblings based on photographs of their face. The network will received two -dimensional vectors  and representing the pixels in each image, and will return a value  which we interpret as the log-odds that the children are siblings. So the model has type-signature .

There are two ways we can do this.

1. We could use an architecture , where —
•  is the sigmoid function
•  is an  matrix of learned parameters,
•   is a learned bias.
• This model has  free parameters.
2. Alternatively, we could use an architecture , where —
•   is the sigmoid function
•  is an  upper-triangular matrix of learned parameters
•   is a learned bias
• This model has  free parameters.

Each model has a vector of free parameters . If we train the model via SGD on a dataset (or via some other method) we will end up with a trained models , where  is the architecture.

Anyway, we now have two different NN models, and we want to ascribe beliefs to each of them. Consider the proposition  that siblingness is symmetric, i.e. every person is the sibling of their siblings. What does it mean to say that a model knows or belives that .

Let's start with a black-box definition of knowledge or belief: when we say that a model knows or believes that , we mean that  for all  which look sufficiently like faces. According to this black-box definition, both trained models believe

But if we peer inside the black box, we can see that NN Model 1 believes  in a very different way than how NN Model 2 believes .

• For NN Model 1, the belief is encoded in the learned parameters .
• For NN Model 2, the belief is encoded in the architecture itself .

These are two different kinds of belief.

# 2. Symbolic Logic

Suppose we use GOFAI/symbolic logic to determine whether two children are siblings.

Our model consists of three things —

1. A language  consisting of names and binary familial relations.
2. A knowledge-base  consisting of -formulae.
3. A deductive system  which takes a set of -formulae (premises) to a larger set of -formulae (conclusions).

There are two ways we can do this.

1. We could use a system  , where —
• The language  has names for every character and familial relations
• The knowledge-base  has axioms
• The deductive system  corresponds to first-order predicate logic.
2. Alternatively, we could use a system , where —
• The language  has names for every character and familial relations
• The knowledge-base  has axioms
• The deductive system  corresponds to first-order predicate logic with an additional logical rule .

In this situation, we have two different SL models, and we want to ascribe beliefs to each of them. Consider the proposition  that siblingness is symmetric, i.e. every person is the sibling of their siblings.

Let's start with a black-box definition of knowledge or belief: when we say that a model knows or believes that , we mean that  for every pair of closed -terms . According to this black-box definition, both models believe

But if we peer inside the black box, we can see that SL Model 1 believes  in a very different way than how SL Model 2 believes .

• For SL Model 1, the belief is encoded in the knowledge-base .
• For SL Model 2, the belief is encoded in the deductive system  itself.

These are two different kinds of belief. Can you see how they map onto the distinction in the previous section?

# 3. Wittgenstein

In On Certainty, Wittgenstein contrasts two different kinds of belief.

• Humans have free beliefs and hinge beliefs.
• A human's free beliefs are similar to how NN Model 1 and SL Model 1 believe . In other words, these are beliefs encoded in our learned parameters , or in the knowledge-base
• In contrast, a human's hinge beliefs are similar to how NN Model 2 and SL Model 2 believe . In other words, these are beliefs encoded in the architecture itself , or in the deductive system  .
• Here are some of my free beliefs:
• Cairo is the capital of Egypt.
• 101 is a prime number.
• There are eight planets in the Solar System.
• Today is a Thursday.
• Here are some of my hinge beliefs:
• I am currently on Earth.
• Today is not 1943.
• Here is my hand
• The external world exists.
• My memory is at least somewhat reliable over short timespans.
• Let's use LessWrong's favourite analogy — the map and the territory.
• We might say the map knows that Manchester is north of Portsmouth because that's what's shown on the map. This would count as a free belief.
• We might also say the map knows that England is roughly two dimensional — that's also shown on the map. But this would count as a hinge belief, because it's not a free parameter.
• Wittgenstein calls these "hinge beliefs" because they must be fixed, allowing our  world-model to "swing like a door" throughout the rest of the possibilites.
• Hinge beliefs are not like axioms. They aren't foundational, but instead pre-foundational. They are the presuppositions for our conceptual map to connect with the external world whatsoever.
• Hinge beliefs are not subject to rational evalutation or empirical testing, but they can be evaluated in other ways.
• It's somewhat defective to say "I know " or "I doubt " when  is a hinge belief.

# 4. Alignment relevance

• Depending on the architecture, randomly initialised neural networks will "know" things.
• Determining which hinge beliefs are induced by a neural network architecture is (in general) non-trivial.
• Whether a belief is a hinge belief or a free belief will affect —
• Capabilities
• Safety
• Interpretability
• The general trend of ML over the past ten years has been towards free beliefs rather than hinge beliefs. If there are less hinges, then the door can swing through a wider space, i.e. the model is more general.
• Nonetheless, even the most general architecture must induce some hinge beliefs, because otherwise the models couldn't correspond to any external territory whatsoever.
• As a rough rule-of-thumb, I expect that swapping free beliefs with hinge beliefs would make AI more safe and less capable. I'm not sure whether this would be worthwhile on the safety-capabilities trade-off, and I'm not sure whether it would make AI more interpretable (but my guess is slightly yes).
• If mechanistic interpretability goes well, then we should be able to take a trained neural network with free beliefs, identify certain symmetries/regularities within the parameters, and then convert the model into an equivalent model where those beliefs are now hinges. In other words, we should be able to turn knowledge stuck in the parameters to knowledge stuck in the architecture.

# Ω 13

New Comment

Could you elaborate on "For NN Model 1, the belief is encoded in the learned parameters . For NN Model 2, the belief is encoded in the architecture itself "?

If (i.e. is symmetric), then . The first model would (we suppose) learn a symmetric , because in reality siblingness is symmetric. The second model uses a matrix that will always be symmetric, no matter what it's learned.

(In reality the first model presumably wouldn't learn an exactly-symmetric matrix, but we could talk about "close enough" and/or about behavior in the limit.)

Yep, exactly!

Two things to note:

(1)

Note that the distinction between hinge beliefs and free beliefs does not supervene on the black-box behaviour of NNs/LLMs. It depends on how the belief is implemented, how the belief is learned, how the belief might change, etc.

(2)

"The second model uses a matrix that will always be symmetric, no matter what it's learned." might make it seem that the two models are more similar than they actually are.

You might think that both models store an  matrix , and the architecture of both models is , but Model 1 has a slightly symmetric matrix  whereas Model 2 has an exactly symmetric matrix . But this isn't true. The second model doesn't store a symmetric matrix — it stores an upper triangle.

I do not think that "101 is a prime number" and  "I am currently on Earth" are implemented that differently in my brain; they both seem to be implemented in parameters rather than architecture.  I guess they also wouldn't be implemented differently in modern-day LLMs.  Maybe the relevant extension to LLMs would be the facts the model would think of when prompted with the empty string vs. some other detailed prompt.

The proposition "I am currently on Earth" is implemented both in the parameters and in the architecture, independently.

How can "I am currently on Earth" be encoded directly into the structure of the brain?  I also feel that "101 is a prime number" is more fundamental to me (being about logical structure rather than physical structure) than currently being on Earth, so I'm having a hard time understanding why this is not considered a hinge belief.

It feels like if the agent is generally intelligent enough hinge beliefs could be reasoned/fine-tuned against for the purposes of a better model of the world. This would mean that the priors from the hinge beliefs would still be present but the free parameters would update to try to account for them at least on a conceptual level. Examples would include general relativity, quantum mechanics and potentially even paraconsistent logic for which some humans have tried to update their free parameters to account as much as possible for their hinge beliefs for the purpose of better modelling the world (we should expect this in AGI as it is an instrumentally convergent goal). Moreover, a sufficiently capable agent could self-modify to get rid of the limiting hinge beliefs for the same reasons. This problem could be averted if the hinge beliefs/priors were defining the agent's goals but goals seem to be fairly specific and about concepts in a world model but hinge beliefs tend to be more general eg. how those concepts relate. Therefore, I'm uncertain how stable alignment solutions that rely on hinge beliefs would be.

LessWrong's favourite analogy — the map and the territory.