The Seeds of the AI Revolution Sown by Jobs [AI Error Notes]

⑦ Core Resource for Prediction: 'Data'
Data Generation Surges with Mobile Device Popularization
2008 Financial Crisis Might Have Been Different with AI

Editor's NoteExamining failures is the shortcut to success.
'AI Wrong Answer Notes' explores failure cases related to AI products, services, companies, and individuals.

Artificial Intelligence (AI) is a prediction machine, and adopting AI means maximizing profits through prediction. The most important factor for this purpose is ‘data’.

Data is necessary to train AI, and based on that training, AI can make predictions. For AI to make better predictions, it needs to learn more and more diversely.

The iPhone Revolution, Laying the Foundation for AI

Steve Jobs holding an iPhone. Photo by AP News

We understand that AI requires a lot of data. But where did all that data come from? The modern concept of AI emerged in the 1950s, but was there no data back then? Of course not.

AI evolved into a prediction machine thanks to the explosive increase in data and the rapid advancement of hardware performance. With the internet revolution, producing and storing data became easier. The internet opened up a vast resource land called ‘data’.

Then, with the popularization of mobile devices, the history of data reached another turning point. The 2007 ‘iPhone revolution’ was exactly that. The year 2007 marked more than just the launch of a new type of device called the smartphone.

In 1984, the total global internet traffic was 15 gigabytes (GB) per month. By 2014, this figure had increased to 42.4 exabytes. The monthly traffic that occurred in 1984 was generated every one-hundredth of a second in 2014.

The iPhone, a PC you could carry and a mobile PC, ushered in the era of data floods. It created opportunities to collect individual preferences and tastes, such as when and where people buy things and where they move. Thanks to sensors built into the iPhone like GPS, accelerometer, and gyroscope, collecting movement-related data became possible. The enhanced camera function led to an exponential increase in image and video data.

Also, on the App Store, some people created apps while others used them, leaving traces of data. The iPhone completely changed the paradigm of data generation and collection. This became a significant foundation for AI to develop into a prediction machine.

Along with this, the rapidly advancing computer performance since the 2000s played an important role in processing the explosively increasing data. Without the support of computer processing speed and capacity, it would have been impossible to handle data adequately.

To make predictions, data is necessary. Various sensors enable the collection of health-related data. Abundant data on heartbeats make it possible to predict heart abnormalities. Photo shows the Galaxy Ring equipped with a heartbeat sensor. Photo by Samsung Electronics

Samsung Electronics unveiled a device called the ‘Galaxy Ring’ last July. It is a ring that can be worn on a finger and a wearable device. Inside the ring are three sensors. The accelerometer measures the wearer’s movements and activity levels. The optical sensor monitors heart rate, and the temperature sensor detects and observes changes in body temperature.

The information collected by the Galaxy Ring’s sensors is data. By analyzing the body temperature, heart rate, and movement data of people wearing the Galaxy Ring, judgments for health management become possible. When the wearer’s usual heart rate data, abnormal heart rates deviating from the average, and normal heart rate data from others are combined, predictions through comparison become possible. It can predict when the heart rhythm becomes irregular or when early signs of a stroke appear.

The predictive ability of the Galaxy Ring becomes more accurate as the number of users increases. It accumulates normal health data and outlier data from users to gather data on the incidence rates of individual symptoms. By repeatedly collecting information, predicting, and receiving feedback on the prediction results, the predictive power improves.

The 2008 Financial Crisis: Would AI Prediction Have Made a Difference?

Machine learning helps businesses through important functions such as fraud detection, security threat identification, personalization and recommendations, integrated customer service via chatbots, script writing and translation, and data analysis. It is also actively used in autonomous driving, drones, airplanes, augmented reality and virtual reality, and robotics. Getty Images Bank

Without data, prediction is impossible. In today’s world, where sensors and various devices are widespread, data can actually be said to be overflowing.

At this point, we encounter the concept of ‘Machine Learning.’ Machine learning is a method of training algorithms on data to achieve predictions such as pattern recognition or object identification. Through machine learning, AI gains more learning and experience, self-adjusts, and improves its predictive power.

The speed of data generation is increasing day by day, and without machine learning, it would have been almost impossible to analyze and utilize the rapidly growing data in real time.

The 2008 financial crisis that devastated the global economy might have been different if machine learning had existed. Experts at the time also made predictions based on data, but their methods were linear and limited in handling variables.

Reference photo. Getty Images Bank

At that time, methods to predict stock price fluctuations were mainly based on a statistical technique called regression. It is a method that predicts based on the average of past events. For example, ‘If the Korean stock market closed up 15 times and down 5 times last month, it will rise next month at a 3:1 ratio.’ It is a very simple method.

So they added a method called ‘conditional average.’ If the market rose 15 times in a month, they examined ‘under what conditions did it rise?’ They looked at whether it was during earnings announcement periods or how many days the market fell more than 2% the previous day. This allowed for more accurate predictions.

During the 2008 financial crisis, experts made predictions using models like multiple regression that utilized several conditional averages.

The problem was that they used data unrelated to the actual housing market prices. They created and tested hypotheses based on data they believed to be important. Despite abundant historical data, their predictions were significantly off.

AI researchers say it would have been different with machine learning. Even with multiple regression, the number of variables cannot be increased indefinitely because it becomes too complex. However, machine learning can understand relationships among tens, hundreds, or thousands of variables. It excels at filtering patterns and useful information from vast amounts of data. If machine learning at its current level had been introduced on Wall Street, signs of the crisis might have been detected earlier and more accurately.

Data: Is It Enough to Just Collect and Use It?

Stacked books. Getty Images Bank

Although data is abundant, obtaining the necessary data is not easy. Data is information about past observations and experiences. Simply collecting a lot of data does not accomplish anything.

First, questions about the problem to be solved must precede. Questions like ‘What do I want to predict?’ or ‘What predictions does our organization or company need?’ Then, the scope of data related to the problem or phenomenon to be predicted can be narrowed down. In other words, relevant data can be selected. Also, a sufficient amount of data is needed to discover patterns and generalize.

However, there are many pitfalls hidden in data. These pitfalls are so deep that there are quite a few cases where AI services or products fail. It is necessary to examine the risks related to data.

Next Series Preview

⑧ AI That Did Not Give Vaccines Only to Frontline Medical Staff in the COVID War (December 15)
⑨ A Human Life Attached to a Single Sticker (December 21)
⑩ My Face That Even My Mom Didn’t Recognize, But the iPhone Did (December 22)