Hacker News Reader

Comment by truth_sentinell | Hacker News Reader

I'm eager for the day to come when we can map an entire animal to a computer model -or even human-, so animals won't be necessary anymore for this kind of procedure. How many years away do you guys think we are from that dream?

View on HN

The current trend-line puts computing power at the level of the human brain in 10 years or so. However, processing power increases has been slowing down lately both in terms of time and the amount of money required for each new advance.

In addition, actually modeling a brain is not the same thing as having the computing power to do so. Our current chips have billions of transistors. However, our most sophisticated neural networks have a million nodes or so, I believe.

So we're a ways away from simulating a brain, human or otherwise. It's definitely more than 10 years. Probably 20 or more. And since we can't really predict computing power that far out, it's really hard to predict meaningfully.

That depends largely on whether you want to model mites or mammals.

We could probably make a computer model of tiny arthropods in 2017, if someone were willing to throw piles of money at the problem. A computer model of a human will be more difficult.

But also consider that an acceptably accurate model of the human brain would be an AI. You would still be experimenting on a person.

I'd definitely consider a fully emulated human brain another human, because by the premise that's fully emulated, it will also feel desires, and have survival instincts. That'd be another moral debate -similar to slavery-, because there will be people that would emphatize with them, and there will be others that will think they're just "machines".

But that's talk for another topic. On this case, you'd want to emulate only the physical functioning of our brain and body and not our minds.

The emulation program becomes the mind. You cannot avoid it. The natural consequence of the boot-up sequence for a real human brain is a human consciousness. If you create a precise and accurate model of a human brain in software, and emulate its physical functions, the emulation program contains the consciousness.

If you try something clever, like disconnecting all motor controls and sensory inputs, you now have a person who is blind, deaf, numb, anosmic, and paralyzed. The horror of that situation is compounded if you are emulating over a scanned copy of a cadaver's brain.

When that happens, we'll have fully emulated human brains and basically everything will be solved (or destroyed).

Computation will never be cheap enough to get accuracy. Biology is too dynamic.

I think saying 'computation will never be __________' is usually just wrong. We have a decent understanding now of the complexity of a human body, and there's no technical limitation on putting a ton of supercomputers together to do what OP suggested. It's more of a matter of waiting until it's economically worth it.

> We have a decent understanding now of the complexity of a human body

I fundamentally disagree. We have models, sure, but they're approximative, not predictive, and I don't see any indication we'd be able to get prediction ever. It's very difficult to understand how biological approximation layered over chemical and physical approximation could give any benefit we get now from experimentation.

Among other things, there is no such thing as model verification.

Isn't that what we practically have from rats? a "subset" of the human brain? I cringe every time I see on some tech/science news some headline: "this is the reason you do this, or you think that", and then reading the study, you find out it was conducted on rats...

If we are extrapolating results from rats to humans, why can't we just simulate a rat brain and avoid torturing more? Probably because rats are cheap, and supercomputers, well, are super expensive.

I don't think it would be difficult to simulate a fully functional rat brain with state-of-the-art technology and bright people.

Sure, we might be able to simulate a rat brain at some level. I don't think we can do it with accuracy or precision, and certainly not use it as evidence of reality. You think the replication problem is bad now? It compounds when you build on bad evidence.

At this point we're unable to simulate a rat brain. We think that we will be able to, but it's orders of magnitude beyond what we can do right now. Our biggest neural net computers have about a 1 million nodes (last time I checked). Rats have billions of neurons in a configuration we have not fully mapped.

One day we may be able to do that, and when we can, I agree it would be nice to stop torturing rats.

Reply on news.ycombinator.com