People widely think that the development of AI can lead to a future of oppression by machines, but they tend to either underestimate how quickly things can go from Siri to Sky-net.
So first let’s go over the three main levels of artificial intelligence.
Level 1: Artificial Narrow intelligence (ANI)
Level 1 is what we have in our current day, where you can have AI models beating the best players of GO, or winning games of DOTA 2 against the best players. This is the lukewarm level, where we say “Alright, this is not too concerning, but I can see how it could get worse”
Level 2: Artificial General Intelligence (AGI)
Level 2 is where you have Artificial General intelligence, where the intelligence of AI is at the same level or lower but in most fields. The gap between Artificial Narrow intelligence and Artificial General Intelligence seems to be vast, but until we find how exactly to create AGI, there’s no way to be sure.
But think about this: In today’s world, ANI is found in almost every device that you use. The newest phones now have a specially designed neural net processor that is designed to process images and every Google search is made possible by the use of sophisticated neural networks as well. There’s soon going to be smart driving cars with ANI uploaded on it, and all this is almost akin to the primordial soup of amino acids that life came from.
ASI on the other hand …
Level 3: Artificial Super Intelligence (ASI)
Level 3 is where things start to get real. This level describes any AI that surpasses human intelligence. It defines everything from only a little bit smarter to more intelligent than the combined intelligence of everyone on Earth.
Artificial Superintelligence will most likely be created by an AGI that has a recursive improvement algorithm built in, as it is very hard to make something that is more intelligent than yourself through the process of manual coding. Once it has become an ASI, it’s also very hard to stop, because it’s both smarter than you and there are many readily available sources that it can replicate itself into. Not to mention the fact that computers are just inherently better in design than humans they have.
- Hardware:
- Speed: Neurons can only run at 200 Hz max and signals transfer at 120 m/s through the neurons while machines can go up to 5 GHz and the circuits travel at the speed of light.
- Size: The brain is the size it is because of the travel of nerve impulses is restricted to 120 m/s. The computer can infinitely expand RAM and HDD.
- Reliability: Computers can run 24/7 given enough cooling and don’t deteriorate, unlike human neurons.
- Software
- Editability: Easy way to fix software, just edit code while it’s much more different in humans
- Group ability: Computers can talk to each other at light speed, although so far humans seem to be doing better in this aspect.
Intelligence is an interesting thing because it’s very hard for organisms in lower levels to comprehend a vastly more intelligent being’s actions. We also tend to overestimate the difference between the dumbest and smartest humans. When AGI starts catching up to us, we will only see it as it being smarter for “an animal”/”a computer”. When it hits intelligence of the dumbest human we will also be saying “Wow, now it’s like a dumb human!”, but after a couple hours, it could be smarter than Stephen Hawking.
What happens now?
Remember recursive self-improvement? The nature of recursive self-improvement is that a dumb AI improving itself, can only do so at a slow pace, but imagine what happens when it’s as smart as Hawking.Because of the recursive self-improvement and the law of accelerating return, it will become much smarter at a much faster pace. The worst part is that we can’t even comprehend how smart it can be. It’s like trying to teach an ant about quantum physics. To put it into our perspective, a normal person has an IQ of around 100, and a smart person maybe 150. Due to recursive self-improvement, maybe an hour after it reached human intelligence it could be hundreds of time more intelligent. There’s not even a word for an IQ of 1000.
As a human being, we know that intelligence equals power, which means that when it does become real, it will become the most powerful being that has ever been made. Things like North Korea and biological weapons to the will look like a scuffle between 2 small anthills. It could just stomp on the two anthills for fun but it could also find the cure for the disease, hunger, and even reverse aging. Trying to convince it not to kill us would be just to rely on its pity if it even has any emotions. The only way we could try to influence it is to shape it’s “childhood” and teach it the right values and pray.
References
Barrat, J. (2013). Our final invention. New York: Thomas Dunne.
Bostrom, N. (2016). Superintelligence. Oxford: Oxford University Press.
En.wikipedia.org. (2018). Superintelligence. [online] Available at: https://en.wikipedia.org/wiki/Superintelligence [Accessed 14 Mar. 2018].