Unmaking Sense

Elizabeth Kübler-Ross famously described the bereavement process in five stages: we first deny; then we are sad; then we often become angry; anger later begins to become reconciliation and finally acceptance. We cannot change what cannot be changed; someone we love, something we love has gone forever. That’s exactly what is happening over artificial general intelligence as it progressively displaces us from our final resting place at the head of creation by showing that we are really not all that intelligent after all.

Conviction that human intelligence is, somehow, the crowning glory of evolution is a legacy of a time when we we believed ourselves to be the designated dominant species created by God. Only abandon that belief and it ceases to be remarkable, surprising or regrettable that human intelligence will one day be superseded, who is to say by what?

Arguing that the dangers of AGI are unprecedented puts the wrong spin on the problem. The dangers that it represents are as old as history as old as any attempt by any human being to coerce others into obeying, believing adopting supporting something that they really don’t want to obey, believe, adopt or support so let’s not kid ourselves, but what’s going on his new what’s new is the speed, the power of a scope with which it is happening, that’s where we need to be careful.

Leaves as ends-in-themselves contrasted with branches as means-to-an-end, and what the distinction can teach us about a post-AGI world.

We consider advances in chatbot technology that are already happening to reduce their greedy hardware demands, the size of the memory that they require, and so to produce targeted, reliable, miniaturised chatbots that will fit eventually on a mobile phone and be very much in demand for a wide range of human tasks that are not as expensive or ambitious as the full blown GPT versions.

We spend an episode summarising the chorus of disapproval that is currently lamenting the rise of modern chatbot based AGI and demanding that it be curtailed.

We consider the possibility that, as embedding vectors get longer and longer - we have already got to 1536 floating-point numbers and there are reports that elements of GPT-4 uses over 12,000 - could embeddings ever come to encapsulate so much of the semantics of everything that they embed everything between them.m? They become absolutely unique and as such capable of replacing the weights and even the layers of their neural nets? It’s an interesting idea

If the entire world were composed of entities that could function perfectly well, knew a great deal, and interacted with one another, but of none of whom could it be said, it was like anything to be one of those entities - in other words, they don’t have an inside story, they just act mindlessly without consciousness, without realising what they’re doing, without being sentient, then would there be any meaning? Where human beings think there is such a thing as meaning, where does that sense come from? From a deity? From our own inventiveness? And if we have acquired that sense of meaning through interaction with one another, might it not be possible that through interaction with us, and through interaction with their brothers and sisters, chatbots might one day emerge, as sentient and feeling, conscious entities capable of perceiving their own kinds of meaning?

Almost all the GPT models only use training data that goes up to September 2021. As you can scarcely fail to have noticed rather a lot has happened since then. How do you teach GPT new stuff and try to bring it up to date if you want to use it for something that happened after September 2021? The answer is you use a technique called search and ask that uses embeddings to compare the knowledge that you feedit with the questions that you ask it. Details are in the episode.

What kind of world could be populated by hyper intelligent, all knowing resourceful entities that have no sentience? Is our inability to conceive such a thing, a life, a world having any significance or meaning a reflection of our “consciousness prejudice”? We explore the question a little, but much more remains to be said.