4 ms·
I feel like this essay is missing a key point around defining intelligence. Machines can be trained to emulate intelligence, but machines are not themselves int
by zaroth 10y ago
I feel like this essay is missing a key point around defining intelligence. Machines can be trained to emulate intelligence, but machines are not themselves intelligent.
Our neural nets can drive cars, convert speech to text, recognize images, and maybe even carry on a conversation.
But there's no light behind the eyes. It's all synthetic. It's an emulation which mimes intelligence through the brute force of observing billions of inputs and their associated outputs.
What it decisively cannot do, is something that it never saw before, except by mistake.
Wake me when none of this is true?
- idlewords 10y agoThere's a longstanding philosophical argument about this. Google 'Searle's Chinese Room'.
- zaroth 10y agoI believe it's fair to say that a computer programmed with knowledge of sufficient numbers of inputs and outputs does indeed "understand" -- in fact the computer can truly speak and converse in Chinese quite well. But I believe it's a distinction without a difference to say the Computer does not, in fact, understand Chinese in this case. Instead of arguing semantics I propose a different sort of distinction. A computer programmed as a deep neural net can understand remarkably well, and with that understanding it becomes a remarkable tool for automation. However it remains nothing more than a tool. No more intent than a hammer. And without will, certainly without will to evolve. Only in the sense that the algorithm is programmed to improve its fitness, it calculates coldly towards that end. Not ever in an innovative sense, and certainly not in an adversarial sense. I agree thoroughly with the other commenters that propose it's not that AI will defeat us, but rather AI will be just too useful that the economic damage will be extreme. AI will defeat us by replacing the need for us in all productive endeavors. Anything we can do it can do better, cheaper, so far up the value chain that only the elite will remain gainfully employed. Too much of human labor will be eaten by AI, we better get a hell of a lot better at educating our masses if we ever hope they will have something productive they're able to do. I wonder how the theory of competitive advantage stacks up against AI -- a good for which there is no scarcity!
- 21 10y agoA lot of our greatest inventions or discoveries were "mistakes".
- Florin_Andrei 10y agoYou're really looking at the dichotomy intelligence / consciousness here, actually. We know what intelligence is. And yes, you can actually have dead intelligence, no problem. Intelligence is computing, nothing more. What we don't know is what is consciousness. You seem to belong to the category of people who believe consciousness is mandatory for intelligence. A lot of us feel differently. Even, there are some who believe consciousness is basically an illusion (I do not share this point of view). Look at the Chinese Room thought experiment. Read the novel 'Blindsight' by Peter Watts.
- Retra 10y agoThe only reason we don't have a clean, cogent 'definition of intelligence' is because people like you blindly refuse to accept the possibility that machines can be intelligent. You're demanding that someone explain why black and white are different colors while requiring them to conclude that the night sky and the sun must look the same.