The technology for making sensors at very low cost was actually pioneered with the iPhone. Sensors are used to change the orientation of the screen from top-to-bottom to left-to-right, or to collect healthcare data like a pedometer, altimeter, or other device. To improve the recognition accuracy of the voice recognition module, three MEMS microphones are used per iPhone for noise canceling. However, the major semiconductor manufacturers in Japan had an aversion to MEMS and opted not to enter this market. The only analog semiconductor maker in Japan, New Japan Radio, produces 100 million MEMS microphones a year. The reason that MEMS sensors are being produced in such high volume and can be sold at a low cost is because they are used in iPhones. Figure 5 shows an example of a MEMS microphone.

Analog circuits and filter circuits that acquire and amplify the signals from these types of sensors, as well as A-D converters and other analog circuits, are necessary for digitalization. There are products that output the raw signals from MEMS in analog form, and products that perform digital conversion and then output (Fig. 6). Here is an area where Japan’s electronics manufacturers can thrive.

Figure 6. Chip integrating a MEMS accelerometer with an analog circuit
Source: Analog Devices

After the data is collected from the sensor, it is digitalized and saved, and then it is converted into meaningful information through processing such as correlation analysis^*4 and multivariate analysis^*5. Traditionally, data processing of this kind was checked by formulating a model and using it to check the actual results with a simulation model, and any deviations would have been compensated with general empirical values. Recently, AI (machine learning or deep learning) is used to monitor sensor data. This generates an enormous amount of data over time, but by defining this data as the normal trend, any data that deviates from this pattern can be judged as abnormal.

Good people create good AI

These decisions may appear as though the computer were doing so on its own accord. But in reality, humans are the ones teaching the computer what is normal and not normal. In other words, while digitalization is said to make computers an autonomous system, humans still have to do the teaching. There have been negative opinions that the introduction of AI and digitalization will take work away from humans. Unfortunately, humans are the ones who control AI.

It should be noted that just as every person is different, so is AI. This is because the knowledge of AI depends on what is being taught. So if humans teach AI harmful things, the AI will become harmful. If good things are taught, such as things that are helpful to society and bring happiness to humans, then AI will not become harmful to humans. In the final installment, we will consider another aspect of digitalization, uberization.