4 ms·
>How comfortable are we with turning over this kind of life-or-death diagnosis to a machine, especially since our handle on consciousness, as an idea, is so min
by jchw 8y ago
>How comfortable are we with turning over this kind of life-or-death diagnosis to a machine, especially since our handle on consciousness, as an idea, is so minimal?
Frankly, probably more than humans that apparently get it wrong 40% of the time.
- YeGoblynQueenne 8y agoThat's not necessarily a good metric. For example, maybe the machine is getting it right 50% of the time out of pure chance, whereas the humans use a process that, although flawed, can be improved over time to get better results than chance. In this particular case, as the article notes, there's also the fact to consider that the machine learning system is trained on cases identified by humans. It's very unlikely that it will manage to do better than (the best) humans. Performance on its own can often be misleading (e.g., see overfitting). Very often, if not always, you need to understand why a system is performing as it does.
- V-2 8y ago> whereas the humans use a process that, although flawed, can be improved over time to get better results than chance. Indeed the research they refer to - that reports frequent misdiagnosis - points out a logical reason for it happening ("Many patients who are misdiagnosed as being in the vegetative state are blind or have severe visual handicap; thus lack of eye blink to threat or absence of visual tracking are not reliable signs for diagnosing the vegetative state"). So there's your space for improvement already.
- hopler 8y agoEverything computers do better than people was taught to them by humans in some way. There's no reason to assume that human level performance is a cap. There's always theoretical exceptions, but doing a known worse thing is irrational.
- ben_w 8y agoNot everything. AlphaZero learned Chess and Go better by playing against itself.
- YeGoblynQueenne 8y ago>> Everything computers do better than people was taught to them by humans in some way. I find this a false dichotomy. Computers don't "do" things: they have no capacity for choosing their own actions- we do things, using computers to perform whatever task we choose better than we could without a computer. By analogy, I can't hammer a nail with my bare hands as well as I can with a hammer. Does that mean that the hammer is better at hammering down nails than me, or humans, in general? Isn't that a completely absurd thing to say? And if I say that I'm better at hammering nails than a hammer is, because a hammer can't just get up and hammer a nail, isn't that completely nonsensical, also? Rather, we can see myself holding a hammer as a system of a human using a tool to perform a task to a higher degree of quality than possible without the tool. That goes for computers, also, as well as for any other kind of tool that doesn't have its own free will (i.e. tools not of a kind we don't know how to create yet). So, if you wish to argue that humans with computers can do some things better than humans without computers, then I agree with you, fully. But, if you want to argue that computers can do things on their own, right now, as opposed to the distant future when they will probably be able to choose their own actions, then I have to cough politely and change subject. Nice weather today, don't you think?
- jchw 8y agoIf a dice roll beats a human in a life or death situation, I would still prefer the dice roll. It's simply the only rational way to go. Now of course in reality, it's probably not quite like that, there's probably biases and other issues. But, humans can always decide to disregard the machine. In a case like this, it seems being more cautious is usually the route to go.