No Panic for AI

When everyone is talking about AI and worrying about AI, especially worrying about losing jobs due to AI, I strongly believe that the spring of AI has yet to come. Please, no panic for it.

Am I joking? Haven’t I seen the emergence of products like chatting robots or the revolution of deep learning?

The chatting robots that are based on generative AI are basically trained by text. There are some developments on audio and images, but not that much for video, as the size of video is still too big for most machine learning platforms.

In addition, these robots only generate answers that look good. I mean, the flow of language is smooth and without grammar mistakes, but the content is still of low quality. This is because it is still based on a strong assumption that all information in the training dataset is correct. To filter out incorrect information in the training dataset requires tedious human labor, which means the development of AI products itself requires large human resources who don’t have to know the underlying theories.

Therefore, please don’t panic about losing jobs, even when you don’t know much of the technique.

Have you got experience comparing texts and videos on your computer? The size of a two-hour movie will be much larger than a huge number of text-based novels. This is why I think the techniques are still vulnerable when it comes to analyzing videos.

The annotation of images or videos will be even more challenging.

In a lot of situations, if we want to detect unusual events, for example, dangerous driving in traffic surveillance, it is still an unsolved problem to collect a dataset with a sufficient number of those events. Simulation is a direction, but there is no real research in this direction so far.

Process mining looks fantastic. But do you know that the log data in information systems are not in the format that can be used directly by the algorithms? There is a tedious process to prepare data initially and there are many ambiguous matters.

Many papers that use deep learning techniques on small datasets claim that the proposed techniques are more advanced than using shallow learning like SVM. But in fact, they simply get good experimental results due to overfitting on the small datasets.

The ambitious proposal of channel sensing in communication channels in 6G is unrealistic at this stage. There is a lack of foundation from machine learning theories to experimental designs in this area. How can it be real around 2030?

 

The Composition of the GO TO SCHOOL Song

Yesterday, I made a video recording of a song that I composed last year. The name of the song is "GO TO SCHOOL". It was composed when my older son just started prep. A few months before this, I wrote a nursery rhyme for my younger son's birthday. As a father, I know I should love the two boys equally, without favoring one over the other. To prove that I loved the older brother too, I felt I had to compose a piece of music for him.

The melody is relatively fast. There are notes moving up and down with large gaps, as if the kids are bouncing and jumping while going to school. They sometimes are running and become excited when they reach school and greet their teachers and friends. The music is composed in upbeat rhythms, indicating the energy of young children.

Sincerity and Hypocrisy

In 2017, when I was in the trough of my life, I went to China before starting a new job in Shenzhen.

Having not met my university classmates for almost ten years, those who worked in Shenzhen held a gathering in a Cantonese restaurant to welcome my return to China. After lunch, those who had married went home, leaving only one who was still single to accompany me.

"Little New, I know you've suffered a lot due to ZTE. But please don't talk about this unhappy past anymore, as you should move on from it and look forward to the future. Nevertheless, the HR who hired you had resigned a couple of years after you left," he said. Then he continued, "Don't be too attached to friends who have blocked you. I know you felt hurt, but there are many more classmates who wanted to contact you."

I am still grateful for his concern for me. However, when looking back at the stories in ZTE Corporation, I deem it a matter of pride rather than suffering, as it is not easy for an undergraduate student to be 'deceived' by a multinational tech giant.

Especially, I strongly felt the kindness from others in ZTE when I submitted my application to resign.

"I only want to go to the R&D department," I said to the interviewers at the Tangcheng Hotel, where the ZTE campus recruitment team was.

"But our company has a policy of only recruiting graduates with a master's degree or above for R&D positions," they said.

"If it is impossible, I think we don't need to waste time on the interview," I said.

"But since you've come here today, why not go through the interview process first?"

Then they asked me questions about communications techniques.

"For the seven layers of the TCP/IP protocol, which layers were implemented by hardware? "One interviewer asked.

"This question is a bit meaningless. Any electronic system cannot function without hardware. With basic hardware, any algorithm can be designed using software or hardware, depending on whether you want it to be easier to configure or to perform faster," I said.

"Your answer is interesting, and I quite agree with you..."

"Do you know how to judge if a communication system is synchronized?" He asked.

"If you can see an eye diagram on the oscilloscope," I said.

"Why can you see an eye diagram when it is synchronized?" The other interviewer asked.

"When a system has been synchronized, the sine signals can be grouped into two sets. In each set, all the peaks and troughs are at the same time, whereas the peaks match the troughs of the other set, forming an eye shape," I responded.

They nodded their heads and asked, "Can you tell me your understanding of coding?"

"In a communication system, there are two types of coding: source coding, which aims to reduce the size of information in the data to transmit, and channel coding, which adds redundant information that is used to recover the data for corrupted data," I said.

"How do you consider the techniques in source and channel coding?" He continued to question.

"For source coding, I think we need to consider the size of the source data and the level of redundancy. If you're sending text data, I don't think it is a priority to compress it, whereas for high-resolution video of a blue sky, it is different... For channel coding, it depends on the environment of the communication channel. It is more important in wireless communications. They also depend on the resources you have and the error rates that need to be achieved..."

"But are they the only coding techniques in a communication system?" The other interviewer interrupted me. "Besides the two, we also have PCM coding when converting an analog signal to its digital form."

I then explained the theory and said, "The international standards allocate more codes for signals in the lower frequency band."

"Why does the low-frequency band receive more codes?" He asked.

"Because human voices are usually low-frequency signals, and the textbook explains the situation when telephone communication was the dominant type. But I think this has been outdated. If you want to transmit other data such as images, text, video, and audio other than voices, it no longer been suitable," I then added, "It also depends on the resources and traffic on the systems. To achieve higher quality data, more codes overall are beneficial."

The interviewer then merrily talked to the other and said, "I asked this question almost every year and this is the most special answer. Yes, it is for voices and the future trend is to transfer multimedia data over the network. He is the first one to point out that the textbook content has been out of date."

Then it was the other interviewer's turn to ask questions. He asked me to draw a diagram to illustrate the different components of a communications system, and then required me to write down the Shannon formula for channel capacity.

Then for the question about electronic circuit analysis, I answered it in three parts: linear circuits, digital circuits, and analogue circuits. For analogue circuits, I said the key was to transform the nonlinear semiconductor circuits to linear equivalents. For low-frequency signals, I said we could use parameter-based equivalent circuits. It could be applied theoretically when high-frequency signals are the inputs too, but in practice more usually, the Taylor series is applied and the higher-order components could be ignored... For the question of how to debug in software development, I said we could print out messages in the output console or set breakpoints...

After that, he asked me to explain TDMA, FDMA, and CDMA. Fortunately, I had just learned them in lectures during that semester.

"As you said, by expanding the spectrum, CDMA is good at resisting noise. Could you explain why?"

I was embarrassed and regretted talking too much about CDMA, while I was in fact not fully understanding it. Then I glanced at the paper in front of me, looking at the Shannon formula I had just written down. I said, "I don't know why for general cases, but in a special situation when the data transfer rate meets the Shannon channel capacity limit, based on the Shannon formula, if the spectrum is expanded, the signal-to-noise ratio is reduced, meaning a higher level of noise is allowed."

I thought they would be disappointed as it was almost impractical for a system to meet the channel capacity limit.

However, the two interviewers stared at me for a while and kept silent. After around a minute, one said, "Your study is very good! But why don't you continue for a master's degree?"

"If I graduate with a master's degree, I also need to find a job, right? If I can find a good one now, why do I need to do a master's degree course?"

"You're excellent!" The other said, "But our company's R&D department requires a master degree to enter. You know it is hard to ask the leaders in our Shenzhen Headquarter while you won't accept the offer in Xi'an as your hometown is in Guangdong Province."

"I think I can work in Xi'an. I really like living here," I replied.

"This is not your true thoughts," he said.

"If you now don't have any choice, would you consider a non-R&D position?" The other interviewer asked.

"The condition is meaningless," I said. "If you don't give me the offer to the R&D department, I would just go to another company."  

"Which other company do you want to go to?" He asked.  

"The one that is bigger than yours. Huawei!" I said.  

Then the interviewer said while smiling, "I mean, if you don't have other opportunities, you will still work in ZTE for a non-R&D position, right? I mean the imaginary condition, which may not be real."  

"If I were going to starve to death, I would have to accept, but as I said, this is not possible," I said.  

Then they told me the technical interview had finished and the next section was chatting with me by others, and they examined my oral English skills.

Mindful Calmness from Sufferings

 A couple of days ago, I played a piece of music titled "Moon Reflected on the Second Spring," composed by Hua Yanjun. While the piece was originally written for the Erhu—a Chinese two-stringed fiddle, I played it on the xiao.

The composer, Hua Yanjun, was the illegitimate son of a Taoist abbot and the wife of a rich man. Abandoned by his mother’s family, he was raised and trained in music by his biological father, who kept their true relationship a secret from everyone—including Hua Yanjun himself—until just before his father’s death. Shortly after his father passed away, Hua Yanjun became the temple's abbot. However, he later contracted sexually transmitted infections from frequenting prostitutes, which severely damaged his vision and eventually left him blind. Later he turned to begging and busking on the streets every day.

Originally, this musical masterpiece had no title; the composer simply described it as a reflection of his inner mind. Because of the immense hardships he endured, most listeners believe the music expresses a profound sadness. I look at it differently — he’s in fact using music to pursuit the inner calmness that countered life's chaotic disturbances.

Ultimately, once the mind finds true peace, the weight of all suffering fades into nothingness.

Right Or Wrong

 A little more than ten years ago, when I was participating in a research project on process mining, Moe, my supervisor, told me that we needed a presentation for a few business stakeholders later that day. "I would like you to have a try as this is a practice for you." she told me in the morning when I had just come into the office. "What should we put on the PowerPoint slides?" I asked. "Our project is about cross-organisational comparative process mining," she explained. "Our business partners want to compare each other's ways of handling business processes, so that they can know how they perform compared to organisations in the same industry, and thus they can learn from each other. Today people from a public organisation will be at the presentation, and they are keen to know how competitive their performance is compared to private organisations in the same industry."

"But I'm still confused," I said. "How do we measure performance? I mean there are too many dimensions to compare."

"We only use the processing time, as it is impossible for us to compare based on other criteria."

I hurried to prepare a few PPT slides which showed the results based on my analysis from the past few weeks. After greetings with the people from our business partner, Moe introduced me, and then I showed them the PPT slides.

"It is clear from the figures that your organisation is handling the business processes significantly slower than the private counterparts," I presented.

"It's impossible!" one person from the business partner interrupted. "I've done research internally, and it doesn't look like what you presented. It must be wrong."

"I don't believe the results to be true," the manager from the business partner added.

"Sorry, he just graduated," Dr. Moe Wynn spoke up. "I think he's made a mistake. Nevertheless, he's quite new to this area, as he has just started in process mining research since last year. His understanding is still not so well."

"I am very confident of the results and what I’ve done." I argued. "I've done it several times to ensure the correctness, and..."

That afternoon, I was blamed heavily in the office of Arthur, who was the disciplinary leader. Then I can still clearly remember almost every word from a statement that he said, "Your understanding of the theory is correct. Your approach to investigation is scientific. Your results are objective and true. But your results make the business partner very upset, which will have a very negative impact on the university, as that public organisation is the one that provides funding to this research. Therefore, we have to make them happy.  While Moe said you were wrong, she had softened the influence of your harmful words. But you still argued with her..." I was shocked and couldn’t believe this was spoken from Arthur’s mouth. 

I felt puzzled. Was I really wrong? Or were they wrong instead？ I told my parents that evening that I felt I'd been bullied.

"They were just teaching you," my parents said. "You're a fresh graduate, and they have experience, so they want to help you. There’s no motivation for them to hurt you."

Half a year later, I recounted this story to one of my best friends during my visit to Hangzhou. "I made a very big mistake," I said sadly.

"No, you're not at fault," he said.

"Really? But everyone said I was wrong."

"From a research perspective, you're correct," my friend replied. "But for the career development of yourself, this is not good." Then he sighed. "Maybe it will be better if you just view it as being wrong, as I really hope you can live easier and have a better career development."

“Research is not suitable for you.” He added, “why not try something easier and more comfortable?”

My Journey to Painting

Yesterday, while my young boy was sleeping in the afternoon, I brought my pen kit and fountain pen to the bedroom to draw a picture of "A Sleeping Boy." I was very pleased to see that my skills in drawing from life had been recovered and were even better than when I was a teenager.

I started painting at a very young age. Throughout my childhood, I had very few toys. Trapped in an apartment with cement walls when I had just moved from a rural to an urban area, drawing was one of the few activities that I could enjoy to spend my spare time. Later, my primary school offered some extracurricular activities, and I chose painting. From then on, every Saturday, I would go to an art room to study drawing and painting. There were quite a lot of students in the room, but most of them could not focus on the activities for the whole morning, whereas I was different. Especially when we came to the open air to sketch from life, many children just ran away to play games they liked. I didn't run away because painting was already my favorite activity.

Because of this, the teacher always paid more attention to me. I remember that I was the one who received most of his mentorship. While I was drawing either a tree or a flower, he would try to sit next to me, instructing me how to draw. Occasionally, he asked me to stop and demonstrated to me how to draw.

He said to us, "Watercolor will be a bit too challenging for you. I think you can use gouache because once you've made a mistake, you can just apply a new layer with the new color on top. You can't do this with watercolor, and this is why I don't recommend you start with watercolor." He taught us how to mix colors based on primary colors. Since most students studied casually, he spent more time teaching me, and in most cases, in a one-on-one manner.

During those years, I spent almost all my spare time painting. My academic performance was always very poor, but I did not care about it that much. In the back of my mind, I thought it might be a good excuse to learn fine arts in the future and launch a career as a painter.

Every birthday, I would receive presents about painting, such as a painting brush, a new set of pigments, and a book. There was once a book exhibition held in the city where I lived, and my sister bought a book about Chinese Landscape Painting Techniques. This became the best book I had ever read. I bought traditional ink, paper, and pigments to paint Chinese-style paintings too.

My academic performance improved a lot starting in Grade Six, the final year of primary school. Since then, I had gradually given up painting, though I still tried new styles and techniques occasionally. During my university years in Xi'an, I occasionally did pencil drawing sketches. Then, this hobby was fully abandoned once I came to Australia.

Since last year, I picked up my pen and brush again and resumed this hobby. I'm glad my capacity has basically recovered.

Zero Is Larger Than Negative

 IT companies such as Microsoft and Amazon recently announced job cuts, claiming the cause to be Artificial Intelligence. However, I believe it is merely their technical advertising strategy to promote their AI-related products. Don't panic, as I don't think AI can really replace that many jobs in the foreseeable future.

I can still clearly remember that during my presentation at an international conference on video surveillance in 2012, a representative from a UK-based company told me that their company had spent quite a lot to investigate automatic video event detection systems. However, they later found out that all the methods proposed in recent publications were rubbish, as none of them could really be used in the real world. I felt embarrassed and changed my research direction after returning to Australia. I examined publications in this field and found that a lot of papers did not have any quantitative evaluations, and the sizes of the datasets were far from enough.

On a Sunday afternoon, I came to the lab at our university to study. My classmate from Sri Lanka said, "You are an excellent student. But it is a pity that you're supervised by Dr. Simon Denman. If you had the opportunity to work with Professor Simon Lucey, I believe you could have much more high-quality publications."

"No," I said, "I'm more fortunate to be guided by Simon Denman." Then I continued, "Simon Lucey's papers look good, but I don't like his style of research. He always first has an idea built on a tough mathematical foundation, and then asks students to do the development and experiments."

"So what's the problem?" my classmate asked.

"The problem is he gives his students too much pressure. They have to have good-looking results, otherwise, they will be blamed because the papers cannot be published in top venues," I said.

"But I'm still confused. What is the problem with that?" he asked.

"I believe research is full of uncertainties," I said, "as we are exploring an uncertain world." Then I continued, "I believe we should have at least some degree of freedom to adjust the methodology during the process."

A few years later, I was in Shenzhen undertaking a research project for Wi-Fi sensing, which means detecting human activities using information from communication channels. I resigned and left Shenzhen without any publications a couple of years later. I am proud of this, as no papers mean the best outcome. This is because zero is always larger than negative.

For all existing papers in this area, the arguments are contradictory. They all argued that using Wi-Fi signals was motivated by the fact that they could resist occlusion and even go through walls, while they hadn't explained how they prevented the influence of activities outside the room of the experiments.

The same problem exists in process mining. To evaluate whether a business process model is fine or not, researchers usually interview business users and directly ask them for verification. But how do we know they are not lying?

If the research outcomes of AI are not objective or true, why are we so scared of its influence in the current era? The spring of AI has yet to come.