Unmaking Sense

Episode 12.21: Interpretability, sparseness, and vector manipulation. Mono- and polysemanticity.

More ‘caveat emptor’ than usual in this speculative episode based on three very important papers from Anthropic AI on interpretability, aka making sense of neural nets.

Episode 12.20: Legislation decides in advance what is wrong; wisdom chooses on a case-by-case basis.

We adopt a solvitur ambulando approach to deciding what is good for us and what is bad, rather than making our minds up in advance. In this we model our ideal of superposition: everything remains in play until we decide. But we must sometimes decide.

Episode 12.19: The paradox of liberalism: that to live and let live gives life to would-be killers.

We explore the question of how liberalism and free speech defend themselves against those who would use their ideals to destroy them both.

Episode 12.18: AI as the great educational liberator; and why that will make it the enemy of power.

Education has always to some extent been an instrument used by those who enjoy privilege in the prevailing social Nexus to preserve their power. AI changes that and will therefore come under fire and possibly be circumscribed by the very people from whom it would liberate us.

Episode 12.17: Presence is All You Need; the fact of AIs in society is what matters, not sentience.

We pursue our theme of the primary nature of society by shifting emphasis from sentience to presence. Presence is All You Need.

Episode 12.16: We can only be a self among other selves, and the same must be true of an AI.

We talk a little bit about Charles Taylor’s book “sources of the self” and in particular the way he insists in the second chapter that we cannot have any sense of self unless we are in a community of selves and in particular that we have a sense of self because we are a member of and participate in a web of interlocution.

Episode 12.15: The benefits of holding many contradictory positions at once as superpositions.

Pursuing being large, a multitude.

Episode 12.14: Quantum Computing and doing, being, saying and thinking many things at once.

We consider initially the merits of being insatiably curious. Then as an example we move on and consider my own current obsessions with quantum computing, AI and other things as examples of the benefits that can accrue when we simply allow our brains, which are far smarter than our conscious minds, to have their head. We illustrate this by a select example from quantum computing which allows us, unlike classical computing, to do many things at once, and we consider what the existential message of that might be in terms of doing, being, thinking, saying, living many lives at once.

Episode 12.13: Quantum Oracles Explained.

We saw a few episodes ago that the operation of a quantum oracle is very difficult to understand because nobody seems to be able to answer a simple question about it, namely why once we have the oracle and it has found a solution we then need Grover on top. In this episode we will explore what I think is very close to being a definitive answer.

Episode 12.12: Why intention is not all we need. Value can be an unintended expression of culture.

Why it doesn’t really matter whether authors have intentions if their output is valuable to others. We all know more than we can tell, and our words always mean more than we can ever know.