Artificial General Intelligence (AGI)
So what is Artificial General Intelligence?
In today’s world, everyone is aware of Artificial Intelligence(AI). The first thing which comes to my mind when I think of AI is Robots which have human resemblance. All credit goes to Terminator movie.
So technically AI means the intelligence which is not natural or biological. Some examples are Siri, Alexa, Tesla etc. Alexa can answer any of your question using Natural Language Processing(NLP). But whether alexa can toast a bread? The answer is No. Because it’s intelligence is very narrow. In one task it performs with more than 95% accuracy but in a slightly different task the same algorithm will perform with less than 5% accuracy.
Artificial General Intelligence- The word General means the intelligence should be applicable to broad area. The same algorithm or machine which can play chess should also be able to play at least other board games as well.
Whether General intelligence means human level intelligence?
No, because we humans are not good at every task. To achieve general intelligence first we need to reach human level intelligence. If you ask a person to play chess who has never seen a chess board in life what will happen? First he will play dumb. Will make all the stupid moves but with practice over time, he will learn and will master it some day.
For a human to master a game or anything, it will take few decades but the machine can achieve it in few days. The best example is Deep Mind’s AlphaGo.
It’s all about Data. We are literally mapping everything. For example if we take the example of autonomous vehicles, what is really happening? We are just teaching the machine, first it needs identify the different signs in the road. It uses the image classification algorithm to do it. Then we are telling the machine, if you see a red traffic signal stop your vehicle. If you find vehicles near by, maintain a reasonable distance from them. If there is speed limit, follow it. And surprisingly it obeys everything we say unless there is any technical fault.
So how do you say it’s intelligent? It’s just doing what we say. Tomorrow if there is any new vehicle in the road which the algorithm haven’t seen before, still it will identifies it. That’s what we call intelligence.
Today’s machine learning algorithm works well because of huge volume of data. AlphaGo performs well because we provide it with data of the chess games which were played before. So for a machine or agent to achieve general intelligence whether we need to feed it with the data of everything? I don’t know. We need to try it first to find it. But is it safe?
Safety should be first priority while building an AGI system because it’s better to be slow and steady than to create a chaos.
This is my first answer in this forum. Hope you like it.