YelloyPyrmid wrote: |
---|
But earlier someone said it was only Adobe who used that. But I have heard many do. |
I said that, it was a joke. Adobe had a password database stolen a few months back and it turned out they'd encrypted everything but didn't hash the passwords.
flint wrote: |
---|
Without replicating some form of biological life form, I don't see how it can be possible to make a machine really experience emotion |
Well, a biological brain is made up of neurons and little else -- there are other cells, but they don't take a part in processing, they're just satellite cells of the neurons. By themselves, neurons are very simple: they have a bunch of dendrites and one axon. The dendrites receive signals and the axon transmits them (it sounds like a many-to-one relationship but the axon is branched so it's actually many-to-many). If the sum of the signal to the dendrites is sufficient to pass the neuron's threshold potential then the neuron "fires": it sends a signal down the axon to be transmitted to all the neurons that are connected there. This can be trivially simulated by a computer with something like the following:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
|
struct Neuron {
void input()
{
// Signal received from dendrite.
++potential;
if (potential >= thresholdPotential)
activate();
}
private:
static constexpr int thresholdPotential = 10;
std::vector<std::shared_ptr<Neuron>> outputs;
int potential;
void activate()
{
// Transmit signal over axon.
for (std::shared_ptr<Neuron>& neuron : outputs)
neuron->input();
potential = 0;
}
};
| |
Although you wouldn't use that for an actual neural network -- normally they use floating point values, and the inputs are weighted which means dendrite A can be more important to activating the neuron than dendrites B and C, and the output is usually smoothed using a sigmoid function of some kind -- the general idea is the same. Biological neural networks differ from artificial ones mainly in that (1) most artificial neural networks (ANNs) have only a few hundred neurons whereas even a fruit fly has about 100,000 of them, and (2) ANNs usually only allow data to travel in one direction ("feed-forward") whereas biological brains can have loops, data going in multiple directions, etc. so the structure is much more complicated. Also, brains regularly change their structure during normal operation, whereas ANNs typically don't: the programmer can change the number of layers of neurons (though it's not common to have more than 3 layers) and the number of neurons per layer, but generally learning occurs by changing the weight values which determine the relative importance of each input to the neuron, rather than by adding or removing neurons, altering the flow of data between specific groups of neurons, and other things that biological brains can do. Also, it's not just the number of neurons that count, but the number of connections between them: for a biological neuron, being connected to 1,000 other neurons is not uncommon; for an ANN, having 1,000 neurons in the first place would be unusual: imagine 1,000 instances of the above structure, each holding 1,000 pointers to other neurons: just storing all the pointers would take almost 4 MB of RAM, and that's before you take into account the input to the network. And not to mention the CPU overhead (though a GPU would be better) of processing 1,000,000 synapses.
That being said, I do think ANNs are a promising avenue in AI. It's just that the scale and complexity of the brains of even a mouse -- with 70 million neurons -- are greater than that of the largest neural networks humans have built thus far (I can't find anything about the number of neurons but the most connections was about 11.2 billion, which would be less than the mouse's 70 billion assuming an average of 1,000 connections per neuron; the human brain is estimated to have about 100 trillion for comparison). Also, just having an ANN won't get you anywhere. You have to train it on huge sets of data before it can do anything useful (the Google X neural network spends its time watching YouTube videos of cats: consider the Turing test passed). And I think making bigger and bigger networks will give diminishing returns pretty quickly. It will have to be combined with other techniques.
helios wrote: |
---|
A society where everyone is happy and has everything they want. How dreadful. |
Suffering builds character. Although I suppose you could argue that in Brave New World, since they have total mastery of genetic engineering, they can manufacture character. Besides, I don't think they were really happy as evidenced by their frequent drug abuse. What they were is comfortable, and I'd rather be free than merely comfortable. Also, they were conditioned to be incapable of feeling the full range of human emotion, their human experience narrowed to a vague "slightly better than neutral" feeling. I can tell you from experience that it's boring when 90% of the time, all you feel is "OK". I've spent days trying to find something that would make me sad or whatever, but none of it works.