3 ms·

> What's needed is a fast-living animal, that ages, and gets cancer, in the same way humans do. Then you can just experiment without understanding. A simulated

> What's needed is a fast-living animal, that ages, and gets cancer, in the same way humans do. Then you can just experiment without understanding. A simulated human?

philipkglass 8y ago
Sadly, no. Cells are very complex. Whole multicellular organisms are even more complex. We don't have models nearly good enough to replace laboratory experiments with E. coli, much less good enough to replace medical trials with humans.
- derefr 8y ago
  Do we need to understand anything about how a multicellular organism works, if we have 1. a working model of biology on the intracellular level, and 2. a working model of stereochemistry outside the cell, and use these to just "play out" cell division from an initial gamete within a simulated womb? (I know that's A Lot of computing power, but maybe an algorithm like HashLife could make this possible by, say, 2100?)
  - philipkglass 8y ago
    If it's possible to do something like this at all, I think it must be based on techniques largely yet-to-be-discovered. To accurately model biology on the intracellular level, you need a "map" of the cell down to the molecular level, plus fast, general, highly accurate methods for simulating molecular-level chemistry. For the chemical simulation side of things, it would take many centuries of continued Moore's Law progress to make today's most accurate algorithms for fully quantum chemistry run fast enough. And the wild ride of Moore's Law already appears to be closer to its end than its beginning. We'd need fundamental algorithmic breakthroughs to make it feasible without relying on better simulation hardware solving the problem. That might be possible, but I don't when/how/if such algorithms may appear. On the problem of getting a ground-truth molecular map of a cell in the first place, we know that there are many gaps in current models. We're not sure where all of them are. Even if you had the fast, efficient, vast molecular simulation capacity I sketched above, it would still take a lot of experiment/simulation comparisons to be sure that all the biologically relevant features had been captured in the original model. Even now I'm probably badly understating all the problems on the biology side. I worked at the interface of chemistry, biology, and chemical physics, but I was more familiar with the software/physics side of things. I learned just about enough biology to be useful in my role and to be grateful that even my most complicated debugging sessions didn't involve 3 billion years' worth of undocumented legacy code running on wet nanotechnology.
TomMarius 8y ago
I'm pretty sure that we're very far from that. We can't even simulate relatively simple processes in timely manner.
- XalvinX 8y ago
  And all the much-heralded advances in AI still produce email notifications that are kind of like those old Ad-lib games where the blanks get filled in by gobbledygook. And after listening to one sleep-inducing waterfall video for a single long night of sleeping the next day about 2/3 of my Youtube homepage's suggestions are about soothing, meditative, nature videos... what a laugh.

> What's needed is a fast-living animal, that ages, and gets cancer, in the same way humans do. Then you can just experiment without understanding. A simulated human?

philipkglass 8y ago

Sadly, no. Cells are very complex. Whole multicellular organisms are even more complex. We don't have models nearly good enough to replace laboratory experiments with E. coli, much less good enough to replace medical trials with humans.

derefr 8y ago
Do we need to understand anything about how a multicellular organism works, if we have 1. a working model of biology on the intracellular level, and 2. a working model of stereochemistry outside the cell, and use these to just "play out" cell division from an initial gamete within a simulated womb? (I know that's A Lot of computing power, but maybe an algorithm like HashLife could make this possible by, say, 2100?)
- philipkglass 8y ago
  If it's possible to do something like this at all, I think it must be based on techniques largely yet-to-be-discovered. To accurately model biology on the intracellular level, you need a "map" of the cell down to the molecular level, plus fast, general, highly accurate methods for simulating molecular-level chemistry. For the chemical simulation side of things, it would take many centuries of continued Moore's Law progress to make today's most accurate algorithms for fully quantum chemistry run fast enough. And the wild ride of Moore's Law already appears to be closer to its end than its beginning. We'd need fundamental algorithmic breakthroughs to make it feasible without relying on better simulation hardware solving the problem. That might be possible, but I don't when/how/if such algorithms may appear. On the problem of getting a ground-truth molecular map of a cell in the first place, we know that there are many gaps in current models. We're not sure where all of them are. Even if you had the fast, efficient, vast molecular simulation capacity I sketched above, it would still take a lot of experiment/simulation comparisons to be sure that all the biologically relevant features had been captured in the original model. Even now I'm probably badly understating all the problems on the biology side. I worked at the interface of chemistry, biology, and chemical physics, but I was more familiar with the software/physics side of things. I learned just about enough biology to be useful in my role and to be grateful that even my most complicated debugging sessions didn't involve 3 billion years' worth of undocumented legacy code running on wet nanotechnology.

TomMarius 8y ago

I'm pretty sure that we're very far from that. We can't even simulate relatively simple processes in timely manner.

XalvinX 8y ago
And all the much-heralded advances in AI still produce email notifications that are kind of like those old Ad-lib games where the blanks get filled in by gobbledygook. And after listening to one sleep-inducing waterfall video for a single long night of sleeping the next day about 2/3 of my Youtube homepage's suggestions are about soothing, meditative, nature videos... what a laugh.