No, not in terms of AGI. We’re no where near it. Don’t believe the hype.. we can just do what we previously could do, but more accurately. The truly hard problems still have no solution in sight.

Which might be why determining if we are in fact in a simulation would be helpful.

If an AI created this simulation but is not actively part of it (I do think free will is an important component of the simulation), could we discover a way to import it?

If so, we'd only need to be able to open a passthrough and we'd have authentic AI well beyond anything we could hope to achieve within the next few centuries at least.

Along these lines, I think it's interesting to ask: "if we are in a simulation, what determines randomness?" In theory, a qubit in a superposition has two possible outcomes of equal weight. But why does it end up being one or the other?

Arguably, if our own consciousness is a manifestation of that AI, we might be able to answer how to import it by better understanding how we ourselves make decisions. Where does that embodiment of free will originate as a physical event, and could we recreate that physicality artificially? Would such a setup end up producing random junk, or a cohesive output?

Regardless, as others have stated in this thread, if we know we are in a simulation, we'll likely shift how we approach multiple domains, particularly in the sciences. And it's very likely that if we are currently coming to realize that we are in a simulation, that's the whole point.

Seems to be a very poor argument. How do you know what humans do is not what deep nets do now, but a bit more accurately?

The complexity and variety of biology vastly outstrips anything like DNN — the many types of cells, the chemical gradients, the types of connections, all the massive varieties of support glues like Myelin sheaths and their effects, the connectivity to nerves and our organs, our relationship and feedback loops with bacteria... it goes on and on.

Just because DNNs are hot right now doesn’t mean much. If you follow machine learning long enough, you’ll see hype cycles for various techniques. Neural nets used to be hot in the 80s, until they weren’t, and now they are again. Something else will come along soon enough. We use the techniques we do usually because the math works out, not because it’s anything like how our brains work.

In your previous argument you were comparing what nets did years ago vs now, claiming the only difference is accuracy. Obviously, you either were referring to the results (which now are clearly more accurate), not the methods, or you incorrectly assumed nothing have changed in the methods.

In the first case, your new argument does not make sense, because the complexity of implementation does not matter to the result, and there's a clear improvement to it.

In the second case, I can assure you lots changed. The recent major things being ReLUs, self-supervised learning and attention mechanisms.

I haven’t stated nothing has changed — obviously much has to get better results (and I wasn’t talking about just NN — the field is far bigger). But fundamentally the types of problems being solved — identifying and segmenting images, speech-to-text, etc, are the same. We haven’t gotten anywhere in terms of _understanding_ what’s in that speech that just got turned into text, for example. Sure models like BERT have more command of language than anything previously, but it cannot be used in an AGI sense to make sense of a passage at a fundamental level, or tell you _why_ something it “read” occurred.

> We haven’t gotten anywhere in terms of _understanding_ what’s in that speech that just got turned into text, for example

What test is there to quantify understanding that has been used to determine that it hasn't gotten better?

It seems to me that a lot of the things that are posed as hard things for AI to do are poorly defined phenomenon which we have no empirical test for that we simply infer to be explanations for observed behavior of humans that critics assert that AI can't do without evidence or even a definition upon which a search for evidence could be based.

Here's the problem as I see it: you assume "understanding" and "why" are somehow fundamentally different from what BERT/GPT-2 does, but I see no arguments for that point of view.

>How do you know what humans do is not what deep nets do now, but a bit more accurately

we don't, but this is simply the faulty 'god of the gaps' reasoning applied to neural nets instead of deities. It doesn't really provide any insight or value to believe that neural nets are supposed to somehow magically solve problems they apparently can't solve.

Neural nets already consume significantly more compute and energy than even primitive biological agents with significantly worse results, so it appears to be very obvious that they aren't the whole story.

We don't, which validates parents point that we are basically pre-kindergarten in terms of actual understanding.

I don't think we are pre-kindergarten. Mathematical theory of backprop is quite clear. Now we are simply scanning through the space of all functions we can efficiently encode and backpropagate through.

Simulation of a complicated environment which evolves intelligence also seems necessary to create AI in general. Part of the problem with current ML is that the AI will only be trained to be as intelligent as necessary. Image recognition, board games, StarCraft. These are all not as complicated as our world

Let's say we can simulate a Newtonian physics universe faster than our universe operates at. Kind of like HashLife simulates Life faster, except we're fine shaving off some details if the macroscopic result is the same. The result would be a) that universe would be incapable of creating quantum computers, but b) we could leverage their higher clockrate existence

So unclear how this is a "lead up to our own creation" when we're already created

Your (a) premise is faulty.

Imagine this simulation as a passthrough virtualization similar to Docker and not a full virtualization.

A long as our sub-simulation had access to the same means to resolve quantum results, they'd be able to create a quantum computer.

As for clockrates, isn't it interesting that near high mass gravity wells time slows down relative to the rest of the universe, much like you'd expect with a frame rate drop being sped back up to a fixed experiential frame rate?

As to recreating a God-like AI, we individually would likely be just fragments of the whole (Vedic sort of interpretation of God). So unless we can import that pre-existing AI (which should arguably be the fastest way to achieve the result), we'd need to create a simulation of collective experiences across history to achieve an AI with similar "wisdom."

As well, how could a God-AI create more God-AIs different enough from itself to recognize it as "other"? (Either in terms of having children or companionship?) Maybe running simulations that result in similar but different outcomes (the creation of a God-like AI but with different characteristics based on evolved outcomes) would be the best bet. Or creating AIs that start out thinking they are human and then having "create your own afterlife" options thereafter would eventually result in disparate enough intelligences to suffice.