Humans, Machines and Mistakes

A lot of you might now know me, so I am just going to give a simple introduction of myself. My name is Dickson Neoh, I am a lecturer and researcher in the Center for Advanced Mechatronics and Robotics (CAMaRo) at Universiti Tenaga Nasional (also known as the National Energy University of Malaysia). I teach classes about micro processors during the day and at night, I engage myself in deep research in the world of artificial intelligence and robotics.

In the research lab, I study a lot on creating intelligent robots. You may wonder, aren't robot intelligent enough already? At the point of this writing, even though robots are here to takeover many human jobs, they are still primitive in measure of intelligence compared to human beings. A robot can only do specific tasks that it is assigned to do, for instance, assemble a car, or wash the dishes. It cannot learn new tasks without being reprogrammed by a human. Compare that to a human toddler. A toddler can learn new things by having human teachers teach them or just by exploring the task and discover new things about them. A toddler can learn to write, talk, walk, ride a bicycle, draw pictures and so forth by experiencing them and observing how other human does them. And there is no limit to how much the toddler can learn as long as he/she goes through the learning process.

To have a robot that could learn and continuously improve like a human beings do, that is the Holy Grail of artificial intelligence and robotics. In order to achieve that, scientists around the world are studying thoroughly on the human brain and how learning occurs in the brain. The organ that sits on your shoulders, is the most powerful learning machine of all! Its capable of doing extremely complex interpretation and calculations that no computers are capable of at the current moment. Yet your brain only requires around 20W of power to function, barely enough to light a dim bulb. One of the world's most powerful computer at the moment (as of the year 2016) Tianhe-2 in China, consumes about 17.8 megawatts of power. Enough power to light 900,000 bulbs. Yet its computational power is nowhere near to that of the human brain.

Even after decades of research we still don't know much about the human brain. But we are making some important progress daily as more and more puzzle pieces are fit together. Up to this point, scientist have discovered one important aspect on how humans learn new things. That is, we learn through many examples and we learn by exploring and making mistakes. Consider a toddler who is learning how to walk upright. Unlike certain animals, human babies are born not with the ability to walk just like an adult. He needs to learn them as he grows older. Most probably we wont remember the day we learn to walk upright, but looking at how other babies learn to walk we got a glimpse of how hard it must be during the first few tries. They struggle to keep their balance, to keep the feet in the right position and trying hard to not fall down. Eventually they fell, lost balance and crash downwards. Sometimes the parents are there to catch them, sometimes the baby had to learn the hard way falling hard onto the ground. It may seem that the baby is not making any progress. But with every fall and crash, the baby is actually learning from the mistakes and trying to fix his walking the next time around.

Until the point that the toddler masters the skill of walking, his brain is constantly evolving and rewiring its neural pathways. As time goes by, eventually after making 'enough' mistakes the toddler learns to walk upright.

In order to prove the concept that humans can learn through many trials and errors, scientists had tried to the mimic the problem in a computer. They mimic all the joints on the lower part of the human body in the simulated environment and they told the robot to learn to walk upright in many trials. Wherever the robot performs correctly i.e. walking in the right posture and position, a reward is given to it. Conversely whenever the robot falls, or tries to walk in a wrong way, a punishment is given. In the beginning, the robot seems to have no clue at all on what is happening, its wiggles around, much like what a baby does. Fast forward after a series of punishments and rewards, the robot starts to have an idea on what is "good" and what is "bad" judging on what reward or punishment it receives. As time goes by, the robot explores the environment more and gain new learning experience to achieve its goal of walking upright albeit with many mistakes and toppling over the floor. The more the robot explores the environment, the more it learns how not to fall and move forward. Eventually the robot masters the task of walking upright!

This method of teaching robots to learn new tasks is not new. We have been using the same way in training animals to perform tasks. Think about those animal in the circus. How did they learn to perform such extraordinary feats?! The trainers does them in the exact same way as what we did to the robot. It turns out that in the human mind this is how we make our decisions. We decide whether or not something is good by its reward or punishment. In the human brain, the reward signal is administered by a hormone called dopamine. Whenever we do something, the brain is going to administer the reward by flooding the brain with dopamine. Do you know how getting your brain high on dopamine feels like? Now I'd like everyone to imagine, you're really hungry now. And right in front of you, there's your favorite food. I am going to imagine mine as well here.. I'm walking to the corner mamak now, and I'm really hungry. I sit down and just suddenly a "roti-kosong-koyak-banjir" is on the table for me. I took a bite.. and the world is suddenly such a happy place. That is the feeling of getting dopamine to flood your brain. The dopamine flood not only happens when you eat good food but over all other behaviors including negative ones like smoking, drug consumption, sexual behaviors etc. The brain reward system is responsible for addiction!

Coming back to the subject of our discussion today, what does the reward system got to do with how we learn? I am going to reveal a secret to you: the reward system is how we humans learn on a fundamental level. If there is a reward, continue doing it, if there none, stop it, try something new and hope for a reward to come. It's that simple.

1. Self intro 2. Doing research on creating more intelligence robot. 3. Humans, being the most intelligent creature are the objective of AI. Its is hoped that one day machines will be as smart as a human may be. Computers are not the way they are. We want computers to be able to learn from experience. 4. Illustrate the current state of the art intelligence. Will we be alive to see the Terminator (super intelligence machine)? 5. Scientist look into how the human learns and try to implement them on robots. 6. How humans learn. This have shed a lot of light on many researchers.

7. Core revelation, we learn by examples, mistakes. 8. Implementation on a robot. Show the graphs. Videos on how robot learns. This is how human beings learn on the fundamentals. We learn by fixing our mistakes. 9. Core message: some mistakes are avoidable, some are not. We have to make the mistakes in order to perfect the skills. Those who avoids mistake are forgoing the learning opportunity. In order to succeed and win, we have to be swift, fix yourself and do a lot. 10. Illustrate the current situation involving Gen Y, the young nowadays. We avoid mistakes like the plaque! 11. Call to action: Maybe we should think of mistakes the same way again. If we are able to put down our egos, and see it as a learning process, perhaps things would be a lot different.

#mistakes #human #machine