[_audakel]

>To completely remove programmers from the equation would require essentially a human level artificial intelligence. And if I start seeing near sentient robots walking around, my first thought is certainly not going to be, “oh no, it’s going to take my job!”

Even with all DeepMind can do we still are very far from anything remotely human intelligence. On the extreme low end, the tiny worm C. Elegans has only 302 neurons in its nervous system. The full circuit has been completely mapped out for over a decade, and still no one knows how it works. Bits and pieces are understood, but that is all.

The fruit fly has a meager 100,000 neurons in its nervous system

[cr0sh]

While everything you say is true (ie - I agree), does it really matter?

No - we don't know how the C.Elegans 302 neuron connectome works - but if you slap it into a robot or simulation, it tends to act in a similar manner as the actual biological creature (at least, that's what I understand).

We've seen similar results with biological neural networks (cell cultures and such) hooked up to machines as well.

If the connectome of a fruit fly were somehow mapped, and then simulated on a machine - it is very likely that it would act like a fruit fly.

Taken to the utmost extreme, the same could possibly be said for the connectome of a human being, could it not?

Does it matter in that case, then, whether we understand how it works - versus the fact that it is working?

[notahacker]

I don't think it's remotely reasonable to extrapolate making a simulation of 302 neurons wiggle a simulation of a simple muscular structure without full understanding of the wiggle process to making a simulation of 90 billion neurons produce outputs which resemble expressions of human cognition without an incredible depth of knowledge of a connectome structure that shows more variation in an individual over the course of a day than C.Elegans does within the entire species, and how that relates to long term memory storage and a mind-bogglingly complex array of sensory inputs and outputs.

Why do you?

It seems akin to suggesting that if I can teach my dog to respond to "sit" despite it lacking human emotion or innate grammar, it seems only reasonable to believe my dog can also learn to respond appropriately to the complete works of Shakespeare. Come to think of it, I'd place more faith in our ability to train our pets to write software than our ability to create a human brain emulation so accurate it's like adding another developer to the team.

[throwaway2048]

connectomes fail to reproduce behaviours.

[cookiecaper]

Even human intelligence is not always correct. Serious misunderstandings and miscommunications occur all the time. If we get to the level where instructing a computer is just as simple as instructing a human, we're still going to need people who have the job of ensuring that the instructions were correctly received and are being correctly executed and who fix things when they go wrong.

What's the saying? Something like: "To err is human; to err a million times in one second, you need a computer." Computers are going to need caretakers and, if we're at all serious about usage, we have to recognize that computers will need a special hyper-specific dialect to define what we need done with adequate precision. Human language is not designed to handle such specificity. That dialect will need its own experts. Today, we call those who wield such dialects "programmers".

[BrailleHunting]

There is no reason to assume human programmers are somehow special apart from all other disciplines and tasks, except maybe to delude oneself with a false sense of (job) security.

Computers using AI, deep learning and natural language processing will eventually be able to program themselves. This is the next, big revolution that is inevitable.

[cookiecaper]

>There is no reason to assume human programmers are somehow special apart from all other disciplines and tasks, except maybe to delude oneself with a false sense of (job) security.

How am I assuming that human programmers are special apart from all others disciplines and tasks? I'm saying that ensuring computers do the will of a human will require some amount of human labor. There are many other disciplines and tasks that also require human labor. How is this attempting to elevate anything?

> Computers using AI, deep learning and natural language processing will eventually be able to program themselves. This is the next, big revolution that is inevitable.

"Program themselves" is so loose as to be meaningless. Do you mean that when I say "Echo, play me some music", it programs itself to go out and stream some music? If so, then sure, I agree, thanks.

As another poster stated, "self-programming computers" basically just means that new languages which handle more of the manual work for us emerge and enter general usage. Until a computer can learn to read thoughts (which may happen, but afaik is still quite distant), humans will still need to communicate their intent to the computer; that is, the computer will need "programming", which is the term we now use to refer to instructing a computer.

[olewhalehunter]

Most human intelligence is used to solve problems created by other humans who are doing tasks that can be automated.

[hawkice]

This is one of those times where I think the discussion might be confusing one hard problem with a slightly hard problem that's hard to scale. It seems like the hard problem is understanding how to model how neurons interact with enough precision to get a meaningful picture of how they do what they do. Once we have that, I'm not sure the difference between 302 and 100,000 and 100,000,000,000 will be _fundamentally more difficult_ and not just much more expensive to get going.

That being said, modelling neurons is a tough problem we haven't figured out. All the problems we _do_ figure out kind of accumulate, so programs can be written faster if you're trying to make something where you can get more and more parts off the shelf. I'm not sure this is a topic that demands an analogy. We all understand how to write programs and what makes it easier to do so -- having a cool API or library, ya know, that helps. No metaphor needed.

[p0nce]

I'll believe in strong AI when I see such an AI implement a (correct) Javascript parser.