Wich is quite soon :)
Indeed, I believe that intelligence can be emulated with knowledge : in everyday life we can solve problems because we have seen something like similar. Let's take a map for instance, if some highways are drawn on this map then every human can quickly choose the quickest path to take from a point A to a point B. Here is a map of France :
If we try to go from Clermont-Ferrand to Paris, then we need to go "up" and therefore we can find the quickest path quickly, because we experiment in our life that going in the opposite direction is not a good idea if we want to get quickly to the destination, and therefore here we can quickly find the quickest path because we just need to always go up. For humans it is really easy, but bot hade not experiment this and therefore they don't know that it is useless to go in the opposite direction, and therefore they lose a lot of time searching paths that will never be faster than "going up".
This idea of improving the bot on specific situation is Machine learning : you classify situation and then optimize them. For real world map you never need to go on the opposite direction because they are a lot of road. Then a robot that knows that and that recognize real world map can quickly see the quickest path as human do. But machine learning just classify situation, and doesn't extract from them tips such as "never going on the opposite direction" and this is intelligence : from a new situation you can extract tips that perform well
Let's take a second example to see that humans are quite bad at searching quickest path :
Try yourself to get from the blue point to the red one traveling only on black lines. There are less than 20 points (which is less than before where we had hundreds of points and road) but this one is so much harder ! Here is the quickest path :
Dijkstra algorithm can find that quickly, but humans can't because they have no tips for this situation, and we can't find some because it looks like no tips exist. Maybe it is, but maybe it isn't, and a smatter guy may find some for those situation (if you are feel free to tell me tips that can solve those situation), and this is where AI can rise : classify situation with its own norm and find tips for those.
Therefore, this is steeps you need to follow to create world first AI :
Finding criteria for your norm (satisfying the triangle inequality for instance).
Then let your computer generate a lot of norms.
Then starting from a situation you need to get a lot of situations that are close to this one (for your norm), then solve them all with a classic algorithm.
Then some patterns will rise from them (because situation are similar) : if you didn't take closer enough situation you will get a lot of patterns then you need to take closer situation or to classify patterns you get with machine learning.
Then you need to save of all these
And to solve new situations, you just need to compare to previous situation, take the X closest, and test all their patterns to this new situations : you will be sure at 90% for instance that you have the correct solution : this is intelligence : human can find quiclky something but need time to be 100% sure
To get a deeper AI, you can let "her" search until you decide it, it will explore some situation and therefore be sure proportionally of the time spent searching : exactly as humans
And then, you train this AI so she can decide herself which norm to use
And then you train this AI so she can decide which norm criteria to use
And then ... This is fucking endless : every time she can train itself with new randomly situation to improve, this is intelligence and this is what we do : "No way you need to go backward to get the quickest path" because we can't imagine a situation where we need it, and with a special norm AI will not be able to simulate such a situation and will therefore say "No way ne need to go backward" :)
But time and computing power is a huge issue, it takes so much time to search in a 1000Go database of situation (where distance calculation can take time, depending of what it does), and therefore they won't be such an AI before we improve database searching. This can be done with quantum computer (the Grover's algorithm for instance), but we can design an AI that does that : searching quickly similar situation being sure at 70% for instance (we don't care of a mistake if we take the 100 most similar situation). Once this is done this is over
Get ready to welcome AI guys, we are so dump comparing to what is coming
