Robin Li: AI Won't Steal Human Jobs
-
"I'm not worried about large AI models making jobs disappear. I believe more jobs will be created in the future," said Robin Li, CEO of Baidu, at the 2023 World Intelligence Congress in Tianjin on May 18.
He cited examples: cars replaced horse-drawn carriages, eliminating the jobs of coachmen, but today, the automotive industry has become one of the largest industrial systems globally, creating hundreds of millions of jobs. Thirty years ago, typists' jobs vanished, but roles like network engineers and game developers emerged.
However, the first wave of AI-induced job losses has already appeared in the gaming industry. Earlier reports by Yicai revealed that due to the rapid development of AI art tools, concept art has become the first area impacted. Tools like Midjourney and Stable Diffusion have advanced to a level comparable to that of professionally trained artists with three years of experience. Many gaming companies and outsourcing firms with high-end GPUs now require their art teams to use these tools to cut costs and improve efficiency. Meanwhile, some game companies have laid off a significant portion of their outsourced concept art teams for cost reasons.
Yet, AI art is also spawning new professions, such as prompt engineers. Some original animation companies have started hiring for roles like "AI Artist/AI Researcher."
Why are so many people concerned about AI reducing job opportunities? Li noted that it's because people can see current jobs disappearing but can't envision the new opportunities that will arise—just as people 100 or 200 years ago couldn't foresee the jobs of today. He described himself as an optimist, unconcerned that large AI models will reduce human job opportunities or worsen living standards.
In his view, humanity's greatest danger and unsustainability doesn't stem from the uncertainty brought by innovation. "On the contrary, stopping innovation, ceasing invention and progress, and continuing on inertia—these unpredictable risks are the real threats to humanity."
The emergence of big data, massive computing power, and large models has led to the rise of artificial intelligence. At the same time, AI has shifted direction, moving from discriminative AI to generative AI. Large models have also redefined human-computer interaction. For example, if Li wanted to check which Baidu product lines had gross margins exceeding pre-pandemic levels last month, it might have taken an assistant half a day or a full day in the past. Today, if computers understand natural language, they can generate a table in seconds.
However, generative AI still faces challenges in ensuring content accuracy. When Yicai reporters asked Baidu's AI model, ERNIE Bot, about the company's Q1 performance, it correctly answered Baidu's results but made factual errors about iQiyi's Q1 data, incorrectly stating "revenue of 8 billion yuan, up 4% YoY" instead of "revenue of 8.3 billion yuan, up 15% YoY." This highlights the need for improvement in large models' text accuracy.
Discussing AI applications, Li mentioned existing AI-native tools like Jasper (a marketing creative tool), Speak (a Korean English-learning app), and AI lawyers in the U.S. He joked that hiring a human lawyer for a speeding ticket in the U.S. might cost twice the fine, but with an AI lawyer, "you might not need to pay at all." Public records show that DoNotPay, an AI legal service, planned to handle two speeding ticket cases in court but paused due to backlash from human lawyers and legal risks.
Li also emphasized that the transformative power of AI isn't just due to changing applications but also fundamental shifts in the underlying technology stack. Traditionally, the IT stack consisted of a chip layer (e.g., Intel, AMD, Qualcomm), an OS layer (Windows for PCs, Android/iOS for mobile), and an application layer where developers built software for these platforms.
However, the advent of the AI era has changed the landscape. The current IT technology stack now consists of four layers. The bottom layer remains the chip layer, but the primary chips are no longer CPUs; instead, they are新一代适合并行大规模浮点运算的芯片, represented by GPUs. Above this is the framework layer, which includes deep learning frameworks such as Baidu's PaddlePaddle, Meta's PyTorch, and Google's TensorFlow. The next layer up is the model layer, featuring examples like ChatGPT and Wenxin Yiyan. At the top is the AI-native application layer.
He also revealed Baidu's布局 in the chip layer today, stating that the first two generations of Baidu's Kunlun chips have already been deployed in tens of thousands of units, with the third generation set to launch early next year.