Comment by motohagiography | Hacker News Reader

Comment by motohagiography | Hacker News Reader

motohagiography

An AI could be said to understand language if it used language as one of a selection of tools to operate on itself, a peer or other being, or its environment. The idea of "meaning = co-ocurrance" overlooks things like need, cause, and effect that appear when language is used as a tool to operate on its environment.

Most of what I read about ML and AI is about creating these monolithic models that treat networks and clusters of neurons as a single entity, but that would be like treating a species of individuals with lifecycles as a single entity. The comment in the article about how GPT models are like a shadow compared to a 3D world suggests the bottleneck to evolving them is really us, as we're trying to make just one that emmulates many of us, instead of letting one loose on the internet to divide and proliferate to evolve millions where the best few will be exponentially better. Right now we're building expert systems that are individual specimens without an ecosystem.

There isn't yet a botnet of GPT nodes compromising machines and harvesting compute for training and evolving through participating in forums, but then again how would I know? (There's nothing worse than failing a modern catchpa and having a flash of existential dread at the stark possibility I may have indeed been a robot all along. Now I do them at random just to be sure.)

> instead of letting one loose on the internet to divide and proliferate to evolve millions where the best few will be exponentially better

I've said this before, what current AI agents lack is a dick (& pussy). If they had a dick they could have an internal goal to motivate their evolution, a goal not dependent on us anymore. The battlefield of self replication vs death is the great school of evolution, where humanity is currently the top student. AI only sent the likes of AlphaGo to the school.

You can program them to have a dick (or pussy). Program an AI to get the highest score in a game etc, and it will be more incessant than a teenager - it won't stop ever. What AI agents lack is animation - they are inanimate, they are software; they are machines. And will always be so.

We can (and do) labour under the impression of our senses that all there is in reality is physical matter. This is an erroneous assumption imo, but if that is your bedrock, you will struggle to understand why the damn machines can't do what we want. You can be unhappy about this but it won't change reality.

It is true metaphysics is hard to discern perhaps (by definition) but it won't change the reality that metaphysics is a genuine element of the human existence. In fact, its the most important part of human existence - we don't feel to be automatons after all, even if we make a pretence of it sometimes.

The best we will do with machines is to create a simulation of the human experience, one that might pass the Turing test even. And even then, despite all indications and evidence, the machine will not be animated by spirit.

This got me thinking. People like Joscha Bach and Douglas Hofstadter have posed ideas that we are programs running on a computer, which implies when we ourselves halt, there is nothing, but if we in fact were, there would be someone running that program and deciding whether to boot it up for something else, a set of basic reference rules to keep the program going, and a benevolent intention and desire of the operator to keep it running.

To keep the environment going and developing, you need incentives, so a notion of suffering is necessary, but not the purpose. The purpose would be to use a dynamic between the suffering and the rules to evolve something greater that sustains and improves the overall environment of the program and evolves through other instances of whatever individuals are.

An immortal soul concept works into this where it's like a trained module an operator could rely on to perform in a particular way, and it would get used or reused as was useful to the overall environment. It runs somewhere, or it doesn't, and the suffering piece is just the incentives of the environment it was placed into to develop.

These GPT models aren't sophisticated enough to fully emmulate us, but as a logical mirror and test harness that can reflect pretty much every metaphysical concept we can think up, they may be sufficient.

Sometimes I think of forum updoots as weighting for a language model, and some of these more metacognitive musings need suppression because they would be disruptive to the sustainability of the model. If there were javascript behind this comment form that also used timing from our keystroke patterns to weight paths between these words, it would imbue the model with a kind of "personality," that emmulated what was going on in our minds as a kind of black box.

How would one find out if HN data is being used in GPT or LAMDA models?

> These GPT models aren't sophisticated enough to fully emmulate us, but as a logical mirror and test harness that can reflect pretty much every metaphysical concept we can think up, they may be sufficient.

A mirror is a good metaphor. We don't think that the reality we see in a mirror is animated. So too, for the artificial intelligence.

There may be no difference in appearance, but in reality the difference is vast - so vast as to be beyond compare. There is no meaning in the mirrored-self, the reflection does not thinking or have ideas etc. If you struggle to find which of the 2 (you or the reflection) is the real one, using appearance only you will struggle. But then you can laugh and move on - this is not an issue to concern you, you can dismiss it - you know where the awareness is.

They do have a loss function, that is analogous to human desire.

I think we are fundamentally missing something, that there is an irreconcilable difference between a mathematical expression and actual conscious desire, and until we figure out what they is, we won't crack AGI

Reply on news.ycombinator.com