Creating artificial intelligence that supports experiments
The words “data”, “chemistry”, and “engineering” are probably all known to the public. Data is a result of the conversion of various information into a numerical form; chemistry is, to put it simply, research on phenomena where a reaction of two or more substances produces substances with properties different from those of the originals; engineering is research on methods and systems for producing goods.
In other words, data chemical engineering is a field that supports manufacturing based on chemical reactions through the application of data.
Let me explain more precisely what we do in data chemical engineering. For example, suppose that the substances A and B are mixed together to produce a new material, with 100 kinds of substances as candidates for A and 100 kinds of substances as candidates for B. In this case, there are 10,000 combinations for mixing. To examine all the pairs through experiments requires a great deal of effort.
In addition, if the number of substances to mix is increased, or if the temperature at the time of mixing is changed in 10 steps from 10 to 100°C, the number of combinations becomes astronomical. It cannot be handled by human power alone, even with human wave tactics.
Of course, researchers try to make their experiments more efficient by studying and making use of various experiments and research papers from the past.
However, such knowledge and findings are not complete for all substances. So, researchers have to make assumptions from such knowledge and findings and examine them through experiments.
For this reason, artificial intelligence that is capable of predicting experimental results accurately has been created.
It is artificial intelligence that converts various knowledge and findings that humans have into data and then learns the data, thereby predicting new experimental results. The field that aims to create such artificial intelligence is data chemical engineering.
With that, humans make artificial intelligence predict 10,000 experimental results and have only to actually test and verify several predictions that have a favorable result.
Of course, this reduces the number of experiments greatly and is very beneficial in terms of the experimental period, costs required for experiments, environmental loads such as energy required for experiments and emissions of CO2, and animal welfare, if animal experiments are required.
However, if there are large errors between the predictions by artificial intelligence and the results of actual experiments, it will be of no use. To make these errors as small as possible, a lot of accurate data is required.
This means that necessary chemical knowledge and information must be converted into data that artificial intelligence can learn. Therefore, those who are engaged in data chemical engineering need both knowledge of how to handle data, and knowledge and findings of chemistry.
Artificial intelligence takes an active part in various fields of manufacturing
As to how data chemical engineering is put into practical use, examples of joint research in our laboratory include pharmaceutical products.
You have probably taken medicine when ill at some time. Medicine is molecules that bind tightly to the proteins of cells that are the cause of an illness and then produce some effects.
In short, medicine works when it binds well to the target proteins. As such, the development of more effective medicine involves examining the binding of various molecules and proteins by experiments.
However, the number of types of molecules is said to be at least 10 to the 60th power, which is far too many to examine individually.
However, if we create artificial intelligence that has learned on the basis of the data of the results of various experiments in the past, it can calculate new experiments and predict their results. Conducting actual experiments based on its predictions allows us to develop a drug in a shorter time.
Especially in recent years, an unknown virus such as the novel coronavirus can spread like wildfire throughout the world and become a pandemic. To take countermeasures against this, it is necessary to develop new drugs in a shorter period of time.
Data chemical engineering will play an increasingly larger role in the development of pharmaceutical products.
In fact, data chemical engineering is now indispensable for the development of various materials as well as pharmaceutical products.
For example, plastics, which are typical chemical products, come in many forms, from thin and formable products like a plastic bag to hard and robust products such as a product used in an automobile frame. In addition, it is increasingly necessary to manufacture biodegradable bioplastics in volume at a low cost.
To produce such desired plastic materials, it is necessary to design a so-called recipe to mix several molecules in appropriate proportions. Creating artificial intelligence that can calculate such molecular design will allow for more efficient material development.
If you can make a recipe in this way, mass production based on it requires other design and operation management.
For example, if you make a recipe of a delicious curry for one person and then try to make 100 servings of this curry in a large pot, problems arise, such as how to cook the ingredients thoroughly, and it is not enough to simply increase the recipe for one person by a factor of 100.
Similarly, if you make a recipe for new plastic and then try to build a plant for mass production of the plastic, the design for such production is required separately, and to keep that plant in operation without problems, the design for that purpose is further required. In such a case, artificial intelligence created according to data chemical engineering does a great job.
For example, with the shower water temperature set to 40°C, the shower has hot water at that temperature. This is because there is a sensor that measures the temperature of hot water, and the level of heating is adjusted according to the measured temperature.
However, the hardness and strength of plastics cannot be measured in manufacturing processes. This is the same as when you cannot check the condition of all the ingredients of a curry while cooking it in a large pot.
To address this problem, we make artificial intelligence fulfill the function equivalent to that of a sensor that measures the hardness and strength of plastics.
Of course, plastics are not directly measured by artificial intelligence; instead, data, which is measured in the manufacturing processes based on factors such as the amount of molecules of raw materials and the heating temperature and period, is learnt in order to make predictions to ensure stable production of plastics with a certain level of quality.
Conversely, if a sign of abnormality is detected, it becomes possible to predict the occurrence of an abnormal event.
Such application of artificial intelligence will be more widespread in various fields in society.
Artificial intelligence superior to humans is also made by humans
At one time, the occurrence of so-called singularity – an image that artificial intelligence continues to grow, exceeds the human capacity and goes out of control – was a hotly-debated issue.
It is not certain whether such a thing could really happen, but there may be many people who have a vague sense of anxiety and fear about artificial intelligence.
However, artificial intelligence is already playing an active part in various fields in society. In many cases, artificial intelligence works in a position that supports human activities, like working behind the scenes, and thus may not be well known.
Artificial intelligence created in data chemical engineering also functions as support for making experiments, which requires great effort for humans, more efficient. It can be said that its ability is superior to that of humans.
However, such artificial intelligence is created by us. This artificial intelligence that does difficult work for humans, such as designing the development and production of goods, is designed by humans.
In that sense, it is important that we researchers first properly determine what work artificial intelligence will do, what learning it needs for the work, and what data is required for the learning.
Then, the artificial intelligence that grows must be verified and evaluated. As long as we do not make mistakes, artificial intelligence will not go out of control.
In that sense, I think it is very important that you and the general public have interest in and watch information as to what researchers like us are doing.
* The information contained herein is current as of August 2020.
* The contents of articles on Meiji.net are based on the personal ideas and opinions of the author and do not indicate the official opinion of Meiji University.
* I work to achieve SDGs related to the educational and research themes that I am currently engaged in.
Information noted in the articles and videos, such as positions and affiliations, are current at the time of production.