Hi . . . I'm writing this to give my opinion on Artificial intelligence and the recent worry about it in intellectual circles. According to people like Steven hawkings and Elon Musk, AI is potentially dangerous and can actually take over the world.

Because of what they have said, the public has been whipped into a frenzy over worry that we're developing technology that will end us. But is there really reason to worry?

Well, first we have to consider why Musk and Hawking are so worried. What can AI really do?

Apparently, quite a bit. At this point, we need to consider two types of AI. Strong AI and Weak AI.

Weak AI is any software that can act intelligently, but only in specific situations. An example is Siri and/or Cortana. They can be quite smart, but they're fundamentally limited by their programming. Once presented with problems they aren't programmed for, they fail miserably.

Strong AI, on the other hand, is AI that is as smart as a human being. It is capable of insight learning, sapience, and can solve general problems.
Now, it's strong AI everyone is worried about. And that's because of a hypothetical feature called recursive self improvement. and here is what that means: a sufficiently advanced AI will be able to reprogram itself to make itself smarter. Once that happens, it will become even better at making itself smarter and it's intelligence will increase so exponentially that it will evolve into a Superintelligence that is so smart that it can do whatever it wants and we will not be able to stop it.

Now, humans are capable of self improvement. But it's not recursive. In other words, you can't make yourself smarter. You only use the intellectual capacity you have from birth to gather Knowledge. And even if that knowledge makes you smarter, there is a limit. So, imagine you had AI that could exponentially get smarter than you were, You'd have a right to be worried.

But ask yourself: isn't there a price for getting smarter? I mean, software can only advance so much before it takes a toll on hardware, isn't it? If there existed a strong AI, would it be really that easy for it to just repeatedly increase its intelligence without improving the hardware on which it runs? This is especially true if it runs on conventional silicon computers.

I'm of the opinion that while there is a possibility of AI taking over, the chances are as slim as ebola killing off the entire human population or an entire generation consisting of only females. Because while some things are theoretically possible, reality has a way of putting constraints on things.

Firstly, recursive self improvement is limited to hardware:



Imagine a software that can predict 3 moves of chess, then it improves itself such that it can predict 30 moves into the future, unless the new algorithm is terrifically efficient, the computing power needed will have increased. Imagine that it can now predict 300 moves into the future, activity will have increased and there will be need to upgraded. At this point, recursive self improvement stops or at least slows down until hardware is changed.

Secondly, automatic recursive self improvement is not necessary for AI development:

Say a computer figures out an algorithm that makes it smarter, it doesn't have to automatically implement the algorithm. All new algorithms can first be submitted to human observers to examine before implementation is allowed. This allows humans to be in control of every process. Therefore preventing AI from blowing up in our faces.

Thirdly, Each iteration of AI will cause cascades of discoveries that humans can use to advance themselves:




AI cannot simply be created for the sole reason of getting smarter. At every level of intelligence, computers will be. Serve as think tanks and problem solvers and can discover new technology that humans can use to develop themselves. For example, a computer that is intelligent can help to develop ways to allow humans to engage in recursive self improvement so that at the end of the day, humans and AI develop at the same speed and none is left behind.

Fourth, Solutions are not actions:

Just because an AI decides that wiping off the human race is the next step doesn't mean it will actually try to do it. An AI can simply be asked to find a way to make as much nails as possible, and the reply would be: kill humans and harvest iron from their blood too. That doesn't mean the AI will actually decide to start making nails all of a sudden and start killing people too. Nowadays, computers assist in design of computer chips, but they don't manufacture them. An intelligent machine won't be much different. Even if it processes information in an intelligent manner, it won't necessarily act on it.

Finally, AI can be friendly:

By this, I mean AI can be instilled with moral values, hardwired into it. Values it cannot change. It is possible and this is what Elon Musk and Stephen Hawking were really on about. The best way to be completely safe from any chance of AI world domination is to ensure that at every stage, AI exhibits altruistic and moral behavior that cannot be changed. And if this is followed, AI can turn out to be the greatest invention of mankind. One way to do this is realize that everything a computer does is for the benefit of humanity and an AI that destroys humanity will never be able to accomplish its goals because the human element is crucial. As long as humanware is emphasized in AI development, we will be safe from that future.

In conclusion, stop peeing in your pants, guys . . . AI isn't out to destroy us. It is completely dependent on our actions as a species. So calm down.



Source