Hacker News Reader

Comment by 011011100 | Hacker News Reader

"Once the brain is understood well enough to be modeled digitally, it stands to reason that the complete set of data representing an individual person (at a particular state of time) could be copied or transferred just like saving a file."

What exactly are you going to save to that file? Positions of molecules? States of atoms?

Oh, so we're assuming we can abstract away from the physical system? Either we abstract or we save everything. Well, what's everything? Or can the abstraction be "good enough"?

View on HN

It's scientifically certain that we can abstract away from the physical system. In fact, we're already doing that today. We run simulations of molecular interactions all the time. We can even "do" a cortical column in incredible detail.

The problem is not that we don't know whether it's possible. There are two issues concerning this development: we need to know what the right detail/abstraction level is for simulating a mind, and once we do we need to actually get that information. Doing the simulation itself is comparatively easy, though certainly not trivial.

> What exactly are you going to save to that file? Positions of molecules? States of atoms?

To be useful, it can't be just a file, it needs to be a database that can be executed. Positions of molecules seems to be the right detail level according to our current understanding. That includes the spatial configuration of proteins and their exact position in relation to another. But there might not be that many unique molecules to store, just a few thousand maybe.

An interesting question then becomes how can we abstract these scans even further and still be able to have a high fidelity mind simulation?

We can abstract however we like whatever we like. The point is: will the abstraction accurately represent the original physical system? And how do we know? We run simulations of molecular interactions. So what?

I'm not sure where this hostility is coming from, this is the second time you implicitly accuse me of stupidity in this thread.

Edit: actually you're in good company if you think that. I have two emails from HNers (pertaining to different threads on this subject) attesting that I lack the education and probably the intellect to carry that kind of conversation forward. :)

> The point is: will the abstraction accurately represent the original physical system?

I already talked about that in the very comment you're replying to.

Why wouldn't it? It wouldn't be exactly the same, but the important information is there. Does it matter if it's slightly off?

Consider that your real brain is influenced by tons of random factors and meaningless information. If a protein drifts to a slightly different location maybe it could slightly affect the output of that neuron. But if just assume that it is where it is supposed to be for the purposes of the simulation, it should be fine.

It's not like something like that encodes any important information or has anything to do with the algorithm that your brain is running.

The points are

1. We're limited by our own models of physics. We can only record what we know about. That's a pretty significant issue when you're trying to simulate physics.

2. In physics, we can run an experiment over and over again so as to refine our model. We isolate systems and try to study simple interactions. When you study gases, for example, I think you will find that things are modeled statistically. We record details to the extent that they help with our model.

What I'm wondering is how we would do this with biological systems. Can you run the "experiment" over and over to try and understand the "right way" this biological system was supposed to behave? Is there a "right" biological system? How do you know what the relevant biological (sub)systems are? How do you identify what's important? How do you abstract? What's a "good enough" replica?

What I'm trying to get at is that I don't think there is a good layer of abstraction for biological systems. I think the physics is the only good layer of abstractions. But...

3. If we had to record states of molecules, for example, then making a digital copy of everyone seems really not feasible. It doesn't seem feasible because of memory or computation.

I think you could abstract to the individual neurons or maybe the different parts of the neuron. It doesn't matter that you don't have an exact physical simulation of every atom.

Some abstraction would probably be involved; the idea is to recreate all the relevant activity that impacts consciousness while avoiding going any deeper, and most neuroscientists would (I think) speculate that individual molecules aren't necessarily the finest level of detail needed. Neurotransmitters, maybe, but I imagine a neuron doesn't have to be simulated as its constituent atoms to exhibit identical enough behavior.

Reply on news.ycombinator.com