Robots Learning from Their Mistakes

The truth is that not all mind-blowing inventions that have transformed the world came to be because of strategic focus on a team goal. Some scientific discoveries were made by complete mistake. For example, the printer was invented when a Canon engineer by accident placed his hot iron on a pen. After he put his hot iron on the pen, the ink came out of it. This discovery became the foundation for the inkjet printer. Many famous inventions happened because what was supposed to happen, did not happen.

A well-known technological breakthrough occurred when an assistant professor at the University of Buffalo thought he ruined his science project. He was supposed to use a 10,000-ohm resistor to do the right thing in alignment with the strategic focus of the project. Carelessly, he used a 1-megaohm resistor instead. But, the circuit did something different as a result of this mistake. It made a signal that ran for 1.8 milliseconds and then stopped and started again. It turns out that this same pattern works well with the human heartbeat. His error resulted in the creation of a small circuit that reduced the size of the pacemaker from being the size of a large, heavy television, to ultimately weighing only as much as two paper clips.

 

Considering that almost 4 million people today have pacemakers implanted in their chest and tens of thousands are placed in patients each year, the assistant professor’s failed project seems pretty important given that it has resulted in countless saved lives.

Fire has amazed humans for centuries. The need to control fire has distinguished humans in their ability to prepare food and survive hostile environments. In 1826 a British pharmacist was frustrated trying to clean a utensil unsuccessfully. There was a big lump on the end of his stick and when he tried to scrape it off. The only problem was instead of getting the glob off, it ignited. This troublesome error resulted in the first strikeable matches that were then sold in a bookstore.

 

Between 1827 and 1829, this mistake resulted in the sale of 168 matches. By 1858, as the invention was refined, there were manufacturing capabilities to create 12 million matchboxes a year. While Butane lighters are more popular today, approximately 500 billion matches are still used annually in the United States alone as a result of the glob that would not come off the stick.

The famous adage is to learn from mistakes. These inventors not only reflected on their errors but transformed the entire world with what they learned. All people face adversity from time to time, but successful people find a way to navigate the roadblocks and even flourish when things get difficult. Take a hard look at your execution when it did not go as planned. It is important to process what happened and then resume moving forward.

Some researchers have recently taken this concept of success in failure to the next level in programming robots. For the first time, computer scientists are making progress in developing artificial intelligence that can learn from its own mistakes. The development comes from an open-source algorithm called Hindsight Experience Replay (HER) which was released in March 2018. This technology views failure as success and attempts to formalize what many successful humans do naturally. According to OpenAI’s blog, “The key insight that HER formalizes is what humans do intuitively: Even though we have not succeeded at a specific goal, we have at least achieved a different one. So why not just pretend that we wanted to achieve this goal, to begin with, instead of the one that we set out to achieve originally?”

Regarding artificial intelligence, the machine either gets rewarded or not based on the experience of achieving a goal. However, it can be viewed essentially as moving the goal post or a having a goal in hindsight. In a sense, these researchers are training the machine how to embrace what was traditionally viewed as “failure” as instead, not stopping, but using that as a teaching moment, and even rewarding the machine for achieving an arbitrary goal in the face of adversity. And, while the artificial intelligence platform has a long way to go regarding this learning, it could be argued that humans have a tough time with this sometimes, too. Failures are a part of learning, as much as successes. It is through learning that we realize vision only before imagined in dreams.

#ArtificialIntelligence #MachineLearning #HumanExperience

One thought on “Robots Learning from Their Mistakes”

  1. Thank you for another informative site. Where else could I get that type of info written in such a perfect way? I have a project that I’m just now working on, and I’ve been on the look out for such info.

Leave a Reply

Your email address will not be published. Required fields are marked *