How do we build a real artificial intelligence?
Recently I am reading the news that AlphaGo Zero beat previous version of AlphaGo with a performance of 100 versus 0, which raises topics over the internet about how fast the Deep Learning and AI research has been iterated over the small amount of time.
I was discussing this with my old friend who is studying physics in CMU and we brought up an interesting idea that even though the so called AI right now seems more and more intimidating and goes beyond human knowledge and achieve abilities that no one could conquer, its essence still remains the same: you give an input, and it goes through crazy amount of mind blogging computation and gives you the optimal result.
This brings to an old but always ongoing topic that whether AI can “think” like human would do. There are people pointing that AI is never going to acquire “creativity” and “emotion”, the treasure that God only grants human to hold. I have also heard of arguments that in the future AI is going to replace the majority of labor work that are currently done by human, and the only thing that people can do while machines cannot is, to create things that do not exist before. That’s why there are ever-increasing voices that the next generation should focus pre-college education on cultivating innovation rather than traditional STEM course works.
But wait a minute, is it the right assumption that machines are never going to acquire the ability of thinking, or building up associations between unrelated things like creative human does? Or a better question, what essentially does it mean by the ability of thinking?
Let’s go back to the discussion that I had with my friend. I think it is okay (if not accurate) to say that right now what AI is doing is nothing short of returning an optimal result based on given inputs, which is nothing more fancy than you pass in a parameter to your Python function and get a return value from it. But if we step back a little bit, and rethink about how human’s brain makes a difference in this comparison. the answer is, surprisingly, no.
There is no difference, and that is what our brain is doing as well.
Everyday and every moment in our life, our brains are receiving signals from our sense organ, whether it be the images from our retina or the temperature feedback from our fingertip, and that’s the input. No matter our mind is conscious or not, every second our brains are giving responses based on those signals, whether it be the command signal to our body to finish certain action, or the thoughts and phrases that are generated based on what we’ve seen or what we’ve found in our memory, and that’s the output. And we call this process, “think”.
Now if we re-examine the topic that we brought out, whether or not can AI think like a human? It turns out that answer might be yes, because the intrinsic mechanism works the same way.
But hold on, what about creativity?
I hold the belief that creativity is nothing more than coming up with a novel idea to achieve a certain goal, yet this “novel idea” will always more or less depend on the past experience. When the brain is thinking, it just keeps doing while loop throughout its memory database, whether it makes any connection to the current situation or not. And at some point, when we somehow combines unrelated things all together and solves the puzzle in our mind or in the society, we made an innovation. To my intuition at least, this analogy perfectly applies to machines as well, at least for AlphaGo Zero, who comes up with a series of new strategies to excel in the board game.
A machine might give higher weight to certain task based on certain environmental input, just like we find our self interested in learning computer science because of reading overwhelming news about how deep learning is revolutionizing the society.
A machine might go through its queue and find unfinished task after completing prioritized work, just like we get back home from our work sitting and resting in the sofa but suddenly realize the clothes in the basket haven’t been washed for 2 weeks.
A machine might experience low performance when dealing with unexpected amount of data flow, just like we have no mood to work when we break up with our lovers and all the memories and feelings coming into our mind.
Except the way and the terminology we are using to describe, there is essentially no difference between what a machine and a human can do. So with all this being said, how do we build a real artificial intelligence that can think, can create and can explore things just like human do?
Well, cheers, we already have it.