What if I ask you to build a truly intelligent machine? I am not talking about a self-driving car (Check out my article on self-driving cars [here](https://steemit.com/technology/@piyushgupta221/what-it-takes-to-build-a-self-driving-car)), or about a robot that could assist you in your household chores, but a true intelligent machine that can think like a human. A machine capable of responding to new situations by sensing and processing information, and intelligent decision making. To build a machine with such a superior intelligence, how do we even start?
[Image Source](https://www.ner.gov.tw/api/files/5b149e696018b10006d7c201)
To engineer a machine capable of thinking like a human, we need to somehow model intelligence. Engineers tend to solve such problems by building systems based on modular components and designing an abstraction for each of these components. So should we try to study brain and try to engineer different parts of it?
[Image Source](https://previews.123rf.com/images/peterjunaidy/peterjunaidy1205/peterjunaidy120500007/13699569-les-r%C3%A9gions-du-cerveau-de-l-homme-illustration-des-r%C3%A9gions-dans-le-cerveau-humain-.jpg)
Undoubtedly, it sounds like a natural approach to independently build each part of the brain and eventually wire it together to build an intelligent machine. However, with such an approach we quickly run into a situation that each of these modules might do things they are supposed to do but aren't flexible enough to adapt to changing situations, which is a hallmark of true intelligence.
This brings up few important questions about our brain and intelligence? How is the human brain so flexible to learn almost anything? Is it some kind of a learning mechanism/ algorithm that allows it to learn new things? If so, is the learning mechanism so general that it is applicable to any complex situation? Is learning the basis of true intelligence?
![[UNSET].jpg](https://cdn.steemitimages.com/DQmSbUkmfGZvgUZmDNyBoK176ozuFZBxesSMdQkGE8HBuGj/[UNSET].jpg)
[Image Source](https://lh3.googleusercontent.com/-WMWPzOag6yI/VYhZHYKThWI/AAAAAAAAAUs/US-5_wTuWeQ/%25255BUNSET%25255D.jpg%20width=)
There is a good evidence that our brain might be utilizing some general learning mechanism at least for perception. People can learn to perceive things in the way they are not used to perceive. One such example is of BrainPort, where the humans can learn to see things with their eyes closed using their tongue (Check out The New Yorker's article on BrainPort [here](https://www.newyorker.com/magazine/2017/05/15/seeing-with-your-tongue)).
[Image Source](https://lh3.googleusercontent.com/-10RtXp4f5Zk/TrZXapARQKI/AAAAAAAABGQ/Ns2tb_bFNt0/w530-h306-n/auditory_cortex_learns_to_see.jpg)
The other great example is the Ferret's rewiring experiment where the brain learns to use the eyes using the auditory cortex (part of the brain which allows us to hear). This provides us the evidence that whatever is happening in the auditory cortex is general enough, and if hooked up with a pair of eyes, it could learn to process the signals from the optic nerve to see things. Interested readers can check out The New York Times article on Ferret's experiment [here](https://www.nytimes.com/2000/04/25/science/rewired-ferrets-overturn-theories-of-brain-growth.html).
Now we try to answer the most important question. Can we build a learning algorithm that allows us to build truly intelligent machines which can learn the way human's learn? The answer to the question lies in an area of active research which is deep reinforcement learning. Reinforcement learning is a branch of machine learning which tries to mimic the human's way of learning i.e. by interacting with the environment. It is a trial and error based learning method where an agent interacts with the environment (via actions) and learn from its experiences (by receiving rewards or penalties). Deep reinforcement learning combine reinforcement learning with the neural networks in order to solve complex problems.We would learn the details of reinforcement learning and deep reinforcement learning in the future posts.
I hope you all are excited to learn reinforcement learning which many researchers believe could be the gateway to build truly smart and intelligent machines of the future. If you liked the content of the post, make sure to vote and follow.
Acknowledgement: The discussion in this article is motivated from the lectures of Dr. Sergey Levine, who is an instructor of the course on Deep Reinforcement Learning at University of California, Berkeley.
Until next time, Hasta la Vista. Make sure to follow and vote.
hiveblocks