Blackbox Problem

Electric vehicles will now be our new reality. There is no escape from it. The biggest automakers have started to announce that they will stop the production of all fossil fuel vehicles within 20 years at the latest and switch to the production of fully electric cars. This is very good news in every sense, but there is a big obstacle in front of it. Battery problem…

Artificial Intelligence Makes Our Lives Easier

Especially the raw materials required to produce batteries are not enough. Finding these raw materials is a huge challenge, and it's a challenge for an all-electric future. Nickel is at the forefront of this problem. In fact, we may run out of stocks next year. The scientists who came to our aid have recently discovered 4 new materials that could be an alternative to this problem. Of course, finding them is very important news. Much bigger news is that scientists are using artificial intelligence to find these materials. Artificial intelligence did what was necessary and quickly selected the most suitable ones among over 300 elements. This method can have incredible repercussions.

Hypotheses are the beginning of the path that leads to new discoveries in scientific research. You offer an answer to an existing problem and then you start testing your hypothesis. These solutions were in the hands of people until now, but now that we are very slow in this regard and we cannot keep up with the speed of new technologies, we have started to leave this job to the machines.

Scientists and Artificial Intelligence

At this point, the first step was to design neural networks. You can think of these neural networks as a kind of machine learning mechanism created directly inspired by the human brain. This designed neural network formed hypotheses based on the connections and order among the data provided to it. We are talking about correlations that we may never see. In many areas, we started to apply machine learning to speed up scientific processes and eliminate human error. 

With the new battery materials I mentioned at the beginning, researchers normally had to rely on database research tools, modeling the data from that database, and ultimately their own instincts to select the components that would work. But a neural network created by a team from the University of Liverpool to make this process much more efficient was given to him made a combination of the presented materials and presented which ones could be used as a new material to the researchers in a certain order. The researchers followed this order and started experiments in the laboratory. In this way, 4 combinations were discovered that can be used in new types of batteries without having to test everything on the list. This process, which was completed in a few days, would normally take at least 1 year. Andrei Vasilenko, who took part in the study, states that thanks to artificial intelligence, they save an incredible amount of time and effort by only looking at the chemicals they need to take care of.

Of course this is just an example. I want you to think for a moment how this way of working could completely change the world of science. Imagine the possibilities.

According to Renato Renner, a physicist at the Zurich Institute for Theoretical Physics, we may one day unify our theories that explain the way the universe works, thanks to machine learning. But there is a problem. And this is a serious problem. A little scary. These neural networks are a very, very common system. It is used in all artificial intelligence studies. Everyone in every field benefits in some way from these mechanisms resembling the human brain. However, we do not understand how the digital networks we knit with our own hands work, just as we do not understand the way the human brain works. It works for us, but we don't understand how it works for us. For 10 years, we have been classifying big data with machine learning and making predictions. But we find it very difficult to understand how we do this. In a sense, this means that we do not understand the mentality of the artificial brain.

Solution

We named this problem we ran into Blackbox. We need to solve this problem because after a while, we may not be able to distinguish how reliable the solutions provided by artificial intelligence are. We need to understand how and how much we need to develop these neural networks, or whether they answer questions that we did not ask. For this, some interpretation techniques are developed. Thanks to these techniques, we try to see how neural networks design the solutions they offer us step by step. Perhaps, we may not be able to analyze it in all its details, but we can understand the phenomena that recur in general trends. We have a long way to go in this regard. However, if we do not want to have problems with artificial intelligence in the future, we should solve this problem. Otherwise, we may have no choice but to do what artificial intelligence says without questioning in order to improve our civilization. This can take us to a world dominated by artificial intelligence.