AI Technology that Simulates the Scientific Discovery Process
AI (Artificial Intelligence) - its technological progress is remarkable, and it has been implemented in our daily lives. Wide Learning™ is an AI technology inspired by the "discovery" process in science.
The "discovery" process in science involves repeating the following steps:
- Thinking of a hypothesis
- Using data from observations and experiments to verify the validity of the hypothesis
- If the hypothesis is not correct, a new one is considered (Return to 1)
After repeating this procedure, the surviving hypothesis is considered to be a "discovery", a new theory in science.
Story: Taro's "Boiling Point of Water" Hypothesis
Initial Hypothesis
For example, Taro is interested in the boiling point of water. Taro conducted several experiments at home and came up with the hypothesis that "the boiling point of water is 100°C" (We'll call this hypothesis A).
Sometime after that, Taro climbed Mt. Fuji with his father. In an experiment at the summit of Mt. Fuji, they observed that water boiled at 88°C.
This fact does not match the explanation of hypothesis A, so hypothesis A is not correct.
New Hypothesis B
Taro came up with a new hypothesis, B, that stated "the boiling point is not constant; it is 100°C at sea level, but the higher the altitude, the lower the boiling point becomes". Hypothesis B correctly explains both the results of the experiment at home and at the summit of Mt. Fuji. Hypothesis B is now a "discovery", a new theory.
A month later, a strong typhoon passed by Taro's hometown.
At that time, Taro carried out an experiment at home, and found the boiling point of water to be 98°C. This fact cannot be explained by hypothesis B. Hypothesis B is no longer a correct hypothesis.
Further Developing the Hypothesis
So Taro came up with a new hypothesis, C, which stated that "at sea level, the boiling point of water is 100°C in fine weather and 98°C in rough weather. Furthermore, the higher the altitude, the lower the boiling point becomes".
Hypothesis C explains all of the experimental results so far. Hypothesis C is now a "discovery", a new theory.
The process of "discovery" in science is to come closer to the truth little by little by repeating the creation of hypotheses and verifying them through observation and experiments.
The boiling point of water is known to be "about 100°C at one atmospheric pressure (≒ 1013 hPa), and the boiling point lowers as atmospheric pressure decreases".
When at the summit of Mt. Fuji (Approx. 650 hPa) or a strong typhoon passes (Approx. 950 hPa), the boiling point was lowered because the atmospheric pressure was lower than normal.
Taro's hypothesis focused on altitude and the weather, but what actually had a relationship was atmospheric pressure.
Thus, in the process of scientific discovery, it can be said that the most difficult problem is which of the vast number of data items (Weather, temperature, humidity, atmospheric pressure, wind direction, indoors/outdoors, etc.) obtained from observations and experiments is used to construct a hypothesis.
Furthermore, it is not always possible to develop the correct hypothesis with only a single item, instead it is necessary to develop the correct hypothesis by combining several items.
Discover Every Important Hypothesis
Wide Learning™ is an AI technology that implements this process of scientific discovery on a computer.
However, whereas humans think of hypotheses one by one, Wide Learning™ exhaustively examines all possible hypotheses based on the input data. By statistically verifying the correctness of these hypotheses, every important hypothesis is discovered.
This has made it possible for AI to quickly test a vast number of hypotheses that humans have been unable to find. AI can also use the discovered hypotheses to automatically make intelligent decisions such as predictions and classifications.
In this way, since the series of behavior of Wide Learning™ is a reproduction of the logical and objective thought process in science, there is the merit that it is easy to understand the computation process in the middle and the reasoning behind the final decision. In addition, computers may be able to discover new theories that humans have never noticed before by examining vast numbers of hypotheses that humans simply cannot handle.
Discover a Hypothesis
Of course, Wide Learning™ can also be used in business situations. Let's take digital marketing as an example.
Wide Learning™ exhaustively examines all combinations of purchasing data (sex, marital status, age, etc.) and discovers hypotheses that categorize people who have or have not purchased a product (See the figure on the right).
The hypotheses we have discovered can be understood by humans, for example, "unmarried men living in rental properties are more likely to buy", and "unmarried women are less likely to buy", and useful knowledge can be obtained at the same time.
In addition, Wide Learning™ uses the hypotheses and computes the possibility for a new customer to purchase the item.
Unlike traditional AI, Wide Learning™ can explain the reasoning behind decisions, making it easier for field personnel (For example, sales representatives) to understand and accept the AI's predictions.
