I don’t really see the problem.
I never suggested there be a problem.
Because it tests on extrapolated, future humans. Wouldn’t be perfect but, like I said, I see no reason why it isn’t at least logically possible to have a better track record than human authors.
Because it tests on extrapolated, future humans. Wouldn’t be perfect but, like I said, I see no reason why it isn’t at least logically possible to have a better track record than human authors.
What do future humans have anything to do with it? I agree that by doing what you describe a computer could prob. write better novels that appeal to present tastes, but I’m arguing that it can’t innovate the same as human authors.
I believe you said the computer could only write things that appeal to current tastes but couldn’t write for future generations. By extrapolating future humans, atom by atom, it could technically write for any generation you want, within the accuracy of its extrapolation. Consistency would be harder to achieve but the majority of writers/artists considered subpar in their time will never become wildly appreciated. Ones like Van Gogh are the exception, not the rule. I don’t think it’d be hard to beat that.
As for innovation, innovation is just trying new things. New ideas that are good ones is a lot harder, but again I don’t see any reason why it can only be accomplished by humans.
Without a complete model of physics, this is impossible. Why did they build the Large Hadron Collider if all we need to do is to apply computing power to the laws of physics plus initial conditions to recreate any future physical state??
Clayton -
It would have to be much more complete than what we have now, I perfectly agree. The purpose of the LHC is to help fill out that model ;p
Furthermore note I made the assumption of sufficiently powerful computers. Trying to do this with the ones we have now, even given a complete model of physics, would take way too long.
Finally, even a sufficiently powerful machine with a complete physics model can’t make a complete universal simulation: It can’t account completely for the effects of its own actions. To do this it’d need to be able to know its own output before it’s finished formulating it.
I’m only saying that we will eventually have the capability to reconstruct a brain inside a simulated environment. I never said we had to do so from scratch; live brains could be copied by scanning them at the atom level; even the cell level would already be quite a bit of accuracy. So long as whatever takes place in that virtual environment reflects the same physical laws as those of our actual universe, then the outcomes should also be the same. Any simulation will not be perfectly accurate, but you can approximate to a reasonable degree of accuracy given enough computing power. How fine is needed for a brain? We’ll find out with experimentation, but perhaps the classical level is enough. There is no basis to say that it will never be possible.
As far as getting data in and out, understanding this, and doing something useful with this is another question, but I do believe that over time we will eventually figure it out. I agree with you that we currently don’t have that much of a clue, but I don’t see why that should always hold true. The brain is just a physical construct, even if it’s one heck of a complex one. Each step that we take brings us a little bit more understanding and a little bit more knowledge.
What matters is on what level of complexity can you think and how fast. Our brains work within the same physical constraints as dog brains and mice brains, but we can obviously do much more with our brains than they can with theirs.
I read an interesting thesis somewhere that what’s needed to truly advance intelligence is not more brains, but smarter and bigger brains. It helps explain why projects (especially software dev projects) don’t necessarily progress just by throwing more man-hours at them.
The computer will run the instructions that simulate the duplicated brain and the environment for it. The brute force pertains to the computer, not to the brain. I’m simply making the point that this is the “dumbest” way to human-level intelligence, as we’re simply simulating it rather than understanding it or doing any actual innovation.
This alone won’t be enough to surpass human-level intelligence; for that, we will need a much better understanding of how intelligence works and how the brain works. Nonetheless, it’s one of the steps on the way there, and it’s good enough for simulating human-level intelligence and also very useful for experimentation and increasing our understanding of how the brain works.
There are a whole host of ethical issues that come along with this. If we copy a brain to a fine enough level of detail, then does that brain constitute a new human being? Does it have rights? If we fvck up the simulation (and I’m sure we will stumble many times along the way) and the virtual brain "suffers’ as a result, are we guilty of torture?
But the problem is worse than this… there are good reasons to believe there cannot be a physical theory of everything. Basically, we have already proved there can be no mathematical “theory of everything”. We can build a physical device the prediction of the behavior of which would solve any given mathematical problem (a Turing machine which halts if a given math theorem is true, else does not halt… predicting whether the machine halts or not is a physical problem). If math has no ToE, then neither can physics. If physics has no ToE, there is no hope of simulating the physical world in any shorter way than by simply letting it evolve and watching what happens. In other words, there are no mathematical shortcuts that can help us “time travel” and deduce the future state of the world. Actually, this isn’t completely true since you can think of the human brain as exactly that, a device that can “time travel” by imagining likely future states of the physical world. But the human brain has some notable limitations… it is not good at predicting the behavior of very complex or deep systems (like the outcomes of mathematical axioms) and while it is very good at predicting the kinesthetic aspects of the motion of the human body and some other physical objects, it can only do so out to a few moments ahead of the present.
Clayton -
Azure,
I believe you said the computer could only write things that appeal to current tastes but couldn’t write for future generations.
No, it’s much more complicated than that. It can’t write for unpredictable changes in taste. Generational development isn’t guided by the creation of new generations itself, but by innovative creativity that guides their preferences. You can’t extrapolate what changes in taste can occur by collecting data on present tastes. The change in taste may not be driven by the individuals themselves, but by the artist.
I’m only saying that we will eventually have the capability to reconstruct a brain inside a simulated environment. I never said we had to do so from scratch; live brains could be copied by scanning them at the atom level; even the cell level would already be quite a bit of accuracy.
If what you’re trying to do is recreate a human being, why not just clone humans and change their genotype (therefore changing their phenotype)? Isn’t that much simpler?
EDIT: And this reminds me of the notion that genetic manipulation will most likely supercede computer technology (this is currently true for the health care industry).
Cloning humans is fraught with a huge set of ethical issues… significantly more so than the concepts of simulated brains inside a virtual environment, at least from the general perspective. This will have to be approached carefully in order to progress without pushing the entire field underground. Just the notion of creating designer babies scares the heck out of many people on this planet.
I do agree with your edited-in point when we’re talking about the next 30 years or so. Biotechnology has been progressing quite nicely, and I’m really looking forward to future advancements that can greatly increase health and lifespan for the entire population. Simulating minds and advanced AI is something I would expect to be taking off in a highly visible way in perhaps 30-50 years; maybe 20 years if there’s some sort of a breakthrough. Probably not before then.
My own edit: The advantages in moving to a computer substrate is that we decouple intelligence from its physical structure. By bringing intelligence into the digital realm, we can greatly accelerate it, make backups of it, dynamically rewrite it, etc… it just opens a huge new world of possibilities. Whether humans would want to become part of this new world is another question. I think we will see evolution through symbiosis of the digital and biological worlds, and that we shouldn’t fear advanced minds since they will in essence be a continuation of ourselves, rather than something completely alien. That is just my own speculation and opinion though, and humans will have to tread carefully through the decades ahead.
If it’s unpredictable given sufficiently powerful models then what causes it? Simply saying it comes out of the aether and is impossible to understand is running away from the problem.
Like I said, can’t be perfect. But once again, I see no logical reason why the human brain can’t be beaten in this matter.
If it’s unpredictable given sufficiently powerful models then what causes it? Simply saying it comes out of the aether and is impossible to understand is running away from the problem.
Whatever chemical proceses are responsible for innovation. Azure, by the way, I haven’t “ran away” from anything. This part of the discussion has already been made in the previous four pages of this thread.