How AI Changed the Way We Work

The following translation is based on an episode from the popular Chinese-language podcast StoryFM 故事FM. With a subscriber base of over two million, the podcast, hosted by Kou Aizhe 寇爱哲, is celebrated for inviting Chinese people from different regions and backgrounds to tell their own story, in their own voice.

The editors

Kou Aizhe: Late in 2022, ChatGPT, an artificial intelligence (AI) chatbot, made a sudden yet impressive debut, sparking a wave of discussion in 2023. Suddenly, the spotlight was on the world of generative AI technologies, which use artificial intelligence to generate speech, images, videos and more. These technologies, often referred to as AIGC (Artificial Intelligence Generative Content), have also gained attention across various industries alongside ChatGPT’s skyrocketing popularity.

In the first six months of 2023, a wave of new technological advancement swept into the workplace. But what changes has this wave brought to the professional landscape? And how have these changes affected individuals within the workplace? We’ve invited four people from different industries to share some of the transformations they’ve experienced at work.

Our first speaker today, ‘Big Dragon’ (Da Long), is the founder of a small company. He was proactive in introducing AIGC tools to the workplace, which has already become the norm.

AI saves us money

Hello, everyone. My name is Zhu Bolong, and people around here call me Big Dragon. I’m the founder of a tech company, and we currently have about 20 employees. Our main business centres around dance-related fitness, games and training. Back in 2016, we switched from teaching dance offline to online. In 2020, we directed our focus towards motion-sensing dance
games that could be played on home TVs.

I majored in computer science at university. Although what I learned was not related to algorithms, I’ve always been interested in the tech industry. After just a week or two experimenting with ChatGPT, my business partner and I realised the huge potential of AI drawing tools.

We worried about facing competitors who could utilise AI more effectively and potentially push us out of the market. So, starting in February 2023, we made it a requirement for all our employees to start exploring the use of these AI tools. They had to learn even if they needed to put aside their regular tasks.

My business partner’s office is a few cubicles away from mine. One day, I was in my office when suddenly I heard him yell, ‘This is amazing!’ along with the F-word. I went over to his office to find out what was happening. I couldn’t see his face at first because his dual-monitor set-up blocked my vision. But as I got closer, I saw him kneeling in front of his computer.

Still facing the screen, he said, ‘You see this? It’s way better than what I can draw.’ My business partner started his career as a cartoon artist, and he’s worked as an animation director. He takes a lot of pride in his artistic ability. But on that day, it was like AI completely ‘broke’ him. He said, ‘There’s no way I can compete with this. I might as well team up with it.’ I told him, ‘All right. In the coming months, you can put most of your focus into exploring it and making it even better.’

Over the next month and a half, my business partner spent roughly 6 to 8 hours every day studying these AI drawing tools, often staying at the company until around 10 or 11 in the evening. They excited him immensely. Sometimes, I would also be in the office in the evening, and I’d hear him eating while the computer was busy creating pictures. He would often exclaim with surprise mid-mouthful.

Even before he familiarised himself with the use of prompts, plug-ins and so on, he achieved an impressive 50 percent or so increase in work efficiency. Now, several months on, we can almost generate what we want instantly, which is truly astonishing.

Initially, our colleagues from the technical team were quite dismissive of ChatGPT. When they heard that ChatGPT could assist with coding, they felt there were plenty of open-source codes online and that the code of ChatGPT didn’t necessarily adhere to coding standards better than theirs. However, once they mastered it, they realised that it could replace at least 30 percent of their workload, which is quite significant.

Our colleagues in the operations department initially attempted to use ChatGPT for writing, including articles for WeChat pages and video scripts. Similar to the initial experiences of our technical colleagues, when they didn’t know how to communicate effectively with ChatGPT, the generated content turned out overly artificial and formulaic. It lacked depth and substance.

I then showed them how to use AI to write about China’s 5,000 years of dance history with a summary of several important periods. I first used their method to ask AI to write on the topic and showed them the copy. Then I said: ‘This is your way of thinking. Let’s try it my way.’ I gave ChatGPT the prompt ‘Imagine you’re a stand-up comedian. Please summarise China’s dance history in a stand-up comedy style’, then it generated something quite different. My colleagues immediately understood that you can get ChatGPT to role-play, to write in a certain style and to word-count, paragraph and other requirements. This experience transformed their understanding of AI. It actually functions like a real human assistant. Once my colleagues learnt to communicate with it in the same way we communicate with humans, they were able to quickly put AI into effective use.

After that, my business partner and I have put ourselves in the position of the company’s managers and applied AI tools to our daily work to see what problems it can solve and how much efficiency it brings. Then we had to take action.

There was someone in our company who was responsible for design-related work. This does require a certain degree of originality, but his main job was to make poster images, characters and background effects. In February 2023, we discovered that AI could do this very well, and, unlike when using creativity tools such as Chuangkit, we don’t need to consider copyright issues with AI. When a colleague in the operations department discovered that he could complete this part of the design work through AI without designers, I contacted him directly to confirm whether he could complete the work by himself. After getting a definite answer, I went to the designer.

I called him to the stairwell to have a chat. At first, he thought I wanted to talk about something related to his current work. I said: ‘No. You know we are using AI now, and your position is consumption-oriented, not a revenue-generating one. What we need are employees who can bring in web traffic or profits to the company.’ Considering the optimisation of the personnel structure, I said to him, ‘I’m sorry. Your current position is no longer required. You can either transfer to another position or you can leave.’

He said that he needed some time to think. A day later, he came to me and said that he wanted to try a position in operations. But after another day, he said he’d decided to give up. He said, ‘I feel like even if I put in a lot of time on this new job, I still may not make much progress. I’d rather leave.’ The whole thing was brutal, and it was the first time I made a lay-off decision so quickly. Nonetheless, I still believe I made the right call because it was AI that replaced him.

Later, I heard from someone in the operations team that the sacked colleague was hit hard by the experience. He couldn’t find a new job for several months and stayed in his apartment every day. He was aware that he had been replaced by AI. Moreover, the apartment he rented was in the same building as our company, only a few floors above us, and he had just paid the rent. I felt really sorry, but there was nothing that I could do. We didn’t save a lot of money from his salary, but it was enough for a subscription fee to Midjourney [a generative artificial intelligence program], so now everyone else in our company can use it freely.

Later, we realised that we still needed a full-time UI designer to monitor the computer. It does seem cruel that we hired someone exclusively for the purpose of assisting the computer.

Only one week after the job was posted, we received close to 150 résumés, which was pretty scary. Our colleague responsible for recruitment interviewed approximately 20 to 30 of them. Almost all were high performers, but they lowered their salary expectations themselves. AI gives us advantages in recruiting people and negotiating salaries.

During the interviews, we told them that there was the possibility that their positions would be replaced by AI. Our current focus is on recruiting individuals who aren’t at the A level but are at the D level with the potential, with AI’s assistance, to do A-level work. In this way, our costs can be greatly reduced.

One of the candidates lowered his monthly salary requirement from RMB 12,000 to 8,000. He had previously worked in Beijing, where he could earn about RMB 15,000. Returning to Chengdu, he was hoping for RMB 12,000. I asked. ‘What is your salary expectation now?’ He replied, ‘RMB 8,000.’

Our colleagues’ PCs always have ChatGPT open, as they have become accustomed to using it as a search engine. Colleagues in the animation department always have Stable Diffusion or Midjourney open. As everyone’s productivity rises, it frees up a lot of time for breaks and even a little loafing on the job.

The first area we’re looking forward to is AI-generated animated videos, although the quality isn’t yet up to commercial standards. We think that will take three to six months.

Second, for dance-related products, we normally have to pay for the use of copyrighted music, a relatively big investment. There is a lot of music for which we cannot track down the copyright holders, and we have faced lawsuits in the past. But we expect that within the next three months AI will be able to produce any style of music we want. We do respect copyright, but when the creators charge an astronomical price for use of their work, say RMB 200,000, there’s no way we can use it. But AI can replicate their style, and it’s actually the musical style we’re after. So we hold great expectations for the ability of AI-generated music to subvert the music copyrights market.

My client is not at fault, and neither am I

Kou Aizhe: Our second speaker, A Li, works in the music industry. The law of demand and supply means that when companies like Big Dragon’s turn to AI-generated music, someone like A Li will begin losing customers.

My name is A Li. I’m 29 years old and live in Xi’an. I’ve been working in the music industry for six or seven years, doing things like soundtrack creation and song customisation.

I have a coding background and have always enjoyed learning new technologies. After I returned to work when the COVID-19 pandemic [restrictions] ceased at the end of 2022, I noticed a surge in AI-related content on the Internet. At the time, I was most interested in the emergence of AI ‘singers’: AI that could be trained to mimic perfectly a recorded human voice.

I joined a chat group on the subject. The shared document in the chat group was so long, even for someone like me with some coding experience, that it was tough to follow. I had to refresh my knowledge of coding, but after a week or so, I got the program running. I tried inputting my own voice first. I’ve done a lot of recording jobs in studios, so I uploaded the materials to the cloud processor for 20 hours of memory training. I kept the computer running overnight.

The next morning, I downloaded the generated voice. Both my partner and I were in shock because it sounded exactly like mine. My mind was racing, and the next thing I knew, I was sending it to my mother, who heard it and said, ‘You still sing so badly!’ It chilled me that my own mother couldn’t differentiate my voice from AI.

It was thrilling and terrifying at the same time. It occurred to me that AI singing is so developed that there must be AI-generated content and product in all fields related to music.

I’m self-employed. Normally, I get commissions from clients, and there’s a collaborative process. This part of the business has not been lost. The area where I have experienced a greater loss of business is in the customisation of songs and soundtracks. I used to get a dozen or so orders a month, but now I get none. After asking around, I discovered that [AI] is so cheap that human labour simply cannot compete. What would have cost thousands of yuan in the past now cost only hundreds or less. This is a very natural market selection process.

When I first began experimenting with AI for work, I couldn’t use it effectively. After a client heard a demo I sent, he asked, ‘Who wrote this song? It sounds like it’s by someone who has little experience arranging music.’

After a week of using AI, I sent the client a new demo, and he said, ‘That’s pretty good. Can you sell it to me?’ The transformation was interesting and scary. He couldn’t tell the difference between human and AI any more. When I told him that the demo was made by AI, he was so shocked that his pupils dilatated. ‘This is AI?!’ he asked. I told him it took only 30 seconds to
produce, and he fell silent. He was struggling to process this shocking piece of information.

What the market pursues is efficiency. Although what we produce [as humans] may be better, it is inefficient. If our clients can’t tell if a song is written by a human or not, it just proves that AI-generated content has reached commercial standards.

After showing the demo to the client, he stopped contacting me. When I asked him why, he was honest and said that he had found someone else who was willing to use AI to generate content. After all, he wants to receive products in the fastest and most efficient way. What I aim for is higher-quality content, so it’s all right for him to stop cooperating with me.

The impact of AI is comparable to the Industrial Revolution. Textile workers stormed the factories and smashed the steam engine, but progress cannot be reversed. If you can really get the unit price down through AI, it’s not necessarily a bad thing for the individual consumer. While it’s painful for us in the music industry, and our profits will go down, it’s actually a boost for the consumer to be able to get the product they are looking for at a lower price.

In the past, if the client didn’t like our demo, we had to start from scratch and rewrite everything. Now if we use AI to make a demo, and the client thinks the style and content is OK, I only have to customise it further based on the client’s demand. This reduces communication costs and improves productivity.

This situation will force us to step up our game. If we don’t raise the quality of our work and create something artistic and original, we’ll definitely be replaced in the future.

My boss asked us, ‘You’re using AI. How come you’re still slow?’

Kou Aizhe: Our third speaker, Xiao A, is a rookie with only a year of work experience.

Hi, my name is Xiao A, and I’m 24 years old. I work as a game concept designer in Xiamen. I design characters, patterns and special effects for online games. The company I work for specialises in art and design, with more than fifty employees in the design team.

I have been fond of drawing since I was a kid, and considered becoming an art student. However, I came from a less developed region, and there’s a preconception that only students who fail academically studied art. I ended up studying engineering. I didn’t enjoy the courses at university. After graduation, I learned about game concept design and enrolled in several training courses. About two years after graduation, I started to work in the industry.

I joined the company last March, just over a year ago. Early this year, I began hearing about AI-generated paintings, and thought ‘Here we go again’. Starting from March, I began seeing a lot of AI-generated paintings on the Internet. At that time, it was less developed-people drawn with a dozen or so fingers. But it learned really fast and corrected mistakes, so that after a
short while most people couldn’t tell which paintings were generated by AI.

Back then I was involved in a project. The demand for illustrations suddenly surged. All my other colleagues were busy, so the company hired another person. He was given a draft sketch that had already been approved by the client, and was asked to refine it into a full version. The new colleague asked me, ‘How long would it take for me to finish refining the drawing?’ I said, ‘A week or so.’ He sent me an emoji meaning ‘Wow’, and I wondered what he meant. Did he think a week was too short for the task?

The new colleague used AI. I was right next to him. After he finished, he showed me the drawing and asked, ‘Is this OK?’ I didn’t want to be overly critical, so I just pointed out a few problems and told him, ‘Change this and that, and then send it to the manager.’

The manager told him frankly that the quality of the drawing was bad. It was not a particularly difficult task. Since the clothing in the picture was single-layered, at the beginning it worked quite smoothly with AI. However, some of the finer accessories tended to trip up AI. After the colleague had his work rejected, he asked me, ‘What should I do now?’

I said, ‘Didn’t I send you a bunch of guidelines and reference drawings? Why don’t you revise it according to those?

‘Do you mean I need to draw it by hand?’ he asked. He was in disbelief. ‘What about AI? Can AI help me?’ I was speechless.

We use an AI image generator called Stable Diffusion, and I’ve been learning how to communicate with it. But I never get what I want. I think the quality is still pretty poor. There is also a very serious issue: characters drawn by AI don’t seem to have genuine human emotions. Their facial expressions are so dull, and there is always some inexplicable blush on their faces, probably because people have been inputting a lot of images of this kind.

Most gamers now are quite averse to seeing traces of AI in the games they play, so after using AI to generate the illustrations, we have to erase the traces of AI manually. It’s like putting the cart before the horse. People say AI is here to assist humans, but in fact I feel like it’s quite the opposite. It is humans who now have to wipe AI’s arse.

I think our boss’s judgement has already been clouded by AI. He thinks that what it does is good and what humans draw is bad. It is fine for him to criticise young team members like me, but he even criticises our team leader. Our team leader is a relatively senior artist who’s been in the industry for almost eight years. Sometimes when the team leader is editing AI’s work, the boss comes up and says, ‘I think the AI’s drawing looks better’ Our team leader must be furious, but he doesn’t dare to say anything. When the boss isn’t busy, he just sits in the office and uses AI. Because of this, he thinks he can draw too and that he can teach others how to draw.

Editing is a huge workload for us. AI has not reduced our workload by much. Yet our boss has laid off a few people. As a result, my colleagues and I had to work overtime until 10.30pm for more than 20 consecutive days! The overtime work made everyone very depressed, and we all felt like we were on the verge of collapsing. Our team leader was also working overtime, and he said to us, ‘If you want to quit, make sure you have another job lined up. The job market is really bad.’

I have a couple of colleagues who left their previous jobs, and their job-searching journeys haven’t been smooth. They are all much better at game illustrations than I am. If you search the hashtag #failedinterview on social media, you will see many talented artists and designers struggling to find employment. Browsing these posts has been making me increasingly depressed; for a while, I was staying up until two or three in the morning scrolling. Honestly, I have no idea what my future holds. What if I quit my job and can’t find another one and am forced to change careers? Truth is, I don’t have the courage to resign because of this economic environment. I hope that my company fires me because at least I could get some compensation.

Anyway, I have never stopped drawing all these years. After work, before AI and now, I draw for myself. I still hope that my drawing skills can improve.

We want to be ‘preachers’ of Artificial Intelligence Generated Content (AIGC)

Kou Aizhe: Initially, the introduction of AI tools was meant to improve efficiency, but who actually benefits from this high efficiency? There is another group of individuals who have profited from the enormous technological transformations. The fourth speaker, Hu Bo, is one of them.

Hi, everyone. My name’s Hu Bo. I am a lead instructor of the AIGC program at Qieman Education. Our team is based in Beijing and consists of five members.

We discovered AIGC at the end of last year. Some AI drawing tools within the industry suddenly made headlines, and we believed at the time that this would affect the entire design industry in the future.

We already were doing online training, specialising in training graphic designers. So, initially, we integrated new materials, whether it’s how to use Midjourney or Stable Diffusion, as module supplements within our existing employment courses. It was only later that we separated them into a short course.

We stayed up for two nights and came up with the materials for the foundational course on AIGC: writing lesson plans, filming, recording and editing, all in two days. We needed to rush it because if we waited until everyone started doing this, we would have missed the boat.

At the beginning, the foundational course was relatively cheap, around RMB 300. We usually do live Q&A sessions with students on Douyin, and explain the contents of our courses on live stream. During one live session, we casually mentioned the pre-sale of this stand-alone foundational course, and in only about an hour and a half, nearly 80 people signed up for it.

Kou Aizhe: In the following month, Hu Bo gradually expanded and improved the course content, initially consisting of eight sessions focusing on the AI drawing software Midjourney. Eventually, this course was priced at RMB 1,099 and comprised more than 20 video lessons, a collection of software operation manuals and related materials, as well as guides on how to monetise contents on social media platforms like Xiaohongshu and Douyin.

Over the past few months, this new AIGC business line has generated several million yuan in additional revenue for his training institution.

Many companies have swiftly added proficiency in AIGC tools as a requirement in their job postings. But where do the eligible candidates come from? When universities are not responsive enough to provide new graduates with necessary skill training, after-school training institutions like Hu Bo’s seize the newly emerged opportunity and bridge the gap.

Many universities have invited him to give lectures on AIGC to students who are about to graduate.

For those big tech companies, the first requirement in their job descriptions is that candidates should be able to use Midjourney or Stable Diffusion. This means that AI operation skills have become a must. For example, a former student of mine worked for the ride-sharing app Didi. He told me that Didi is no longer hiring traditional designers; they only hire AIGC designers who can train AI using keyword descriptions. These positions are completely new. I’m not afraid of sharing what we teach: we study the job descriptions of companies and teach whatever the employer needs. To gain employment, students only need to complete their study accordingly.

Universities are forcing their students to learn about these new developments because they want their students to gain employment. This is brand-new and highly sought after. The students may have heard of things like ChatGPT or Midjourney yet have no idea of what they are.

My job is to get my students interested in AIGC. To achieve this I’ll have to keep up with industry developments. For instance, at Osaka University in Japan, researchers have successfully combined Stable Diffusion with MRIs in the hospitals to create a ‘human eye camera’; that is, AI can directly re-create what people see by reading their brain scans. The tremendous potential of AI is very intriguing for my students. This is also a topic for them to discuss in job interviews to give the interviewer the impression that they have a deep understanding of the industry. We pay attention to what is being researched in companies and universities, then we pass on the information to our students. Companies also see our students as ‘geeks’ who won’t need to be retrained after recruitment.

Kou Aizhe: According to Hu Bo, the employment rate of their students this year has increased by 30 percent compared to previous years thanks to the new AIGC content. He is so busy he now has time to research new developments in the industry only when travelling on trains and planes.

***

Translated by Master of Translation students Yuan Cai, Zhirui Chen, Yurun Dai, Yifan Li, Wenjing Liu, Jiaqi Tan, and Ke Wu at the University of Melbourne, under the guidance of Mr Yahia Ma. This translation has been edited by Annie Luman Ren and Linda Jaivin for clarity and length.