Today's article continues to talk about AI. From last year to now, we have pushed several articles discussing the era of AI, whether you are optimistic or anxious, it is undeniable that the era of widespread application of AI is accelerating, and we need a more dimensional perspective to examine all this. We'll continue to publish related discussion articles from different perspectives in the future.
▽
If you don't change, you will suffer; Both change and loss... It is the leader - the innovator who is caught up and overtaken. And as a result, all the old advantages—resources, wealth, power—are devalued, and ideas are considered superior to matter. From then on, the door to the future is open to all people with this character, this kind of hands, this kind of mind.
- David Landers, "Rich and Poor"
In the early 20th century, linguist George K. Ziff proposed the famous "principle of least effort"[1], he found that in English words, only a very small number of words are used regularly, so he concluded that the real world is to a considerable extent inert, dynamic processes always find the least energy consumption path, human language after thousands of years of evolution, eventually also has this characteristic.
Modern neuroscience provides a confirmatory explanation for this: according to Lisa Feldman Barrett, president of the American Association of Psychological Science [2], the brain's primary task is to maintain human survival and health by managing the body's budget, and its operating mechanism is to maximize the efficiency of energy use, not the maximization of rational effects, so it naturally tends to be "energy saving" - not using the brain, the brain's most important job is not thinking. Daniel Kahneman[3] created the famous fast-moving, slow-thinking dual-system insight.
It can also be said that the biggest force that has promoted the progress of human civilization and brought about the outbreak of the industrial revolution in the past five hundred years, and the outbreak of the information revolution in the past fifty years, is laziness.
The American economists Acemoglu and Restreper[4] divide technology into enabling technology and replacement technology, that is, whether this technology improves energy efficiency or replaces human labor. Enabling technology can increase the productivity of the labor force, and replace the technology directly replace these labor, so automation has been depressing the share of labor in the value added.
Acemoglu and Restreper give the example of agricultural mechanization: although mechanization reduces the share of labor in agriculture, resulting in fewer jobs in agriculture, mechanization has led to the creation of a series of new jobs in manufacturing and services, and therefore the total demand for labor has risen. Technological advances, they concluded, have brought about a race between automation and labor-intensive new jobs.
The French economist Philippe Aggion[5], the creator of the Schumpeter's growth model, defended that automation led to the loss of jobs through the process of creative destruction (creative destruction was an important concept in Schumpeter's economic theory), and that fewer jobs for workers were due to the bankruptcy of old enterprises that had failed to keep up with the times. As Schumpeter put it, the achievement of capitalism was not "to give the queens more silk stockings, but to give back to factory women by constantly reducing the amount of work required to produce a stocking, so that they could also afford stockings." (From Capitalism, Socialism, and Democracy, Chapter 5)
However, Carl Benedict Frey,[6], author of The Technology Trap, may not share this view, arguing that modern Schumpeterian growth is based on labor-saving technology, creative destruction of employment, and acquisition of new skills, none of which improve the welfare of the people at the bottom, and is therefore not the driving force of contemporary economic progress.
Frey points out that the main source of income for most people is not physical or financial capital, but human capital. 77% of the difference in workers' incomes stems from individual characteristics, workers' skills are their wealth, and they depend on human capital for their livelihood. Therefore, for the people at the bottom of today's society, maintaining the value level of their human capital is still the most important factor in maintaining stable economic and social development.
So Frey believes that "the promotion of artificial intelligence, like the situation of computers in the 90s of the 20th century, requires not only the development of the technology itself, but also huge complementary investment and a lot of experimentation, so that it can fully realize its potential." ”
Of course, at present, mobile Internet technology has indeed created a huge new service industry employment field, which is especially evident in China: according to the Bureau of Statistics7, by the end of 2021, the number of flexible employment in China had reached 200 million, and the flexible employment rate of college graduates nationwide exceeded 16% in 2020 and 2021, and the number of takeaway riders reached more than 4 million; More than 1.6 million people are engaged in anchors and related practitioners on the platform.
In fact, the types of jobs such as anchors and short video self-media have entered the category of "knowledge workers" in a broad sense. Even front-line couriers increasingly rely on knowledge systems to work: Hu Anyan[8] in his book "I Deliver Express Deliveries in Beijing", describes in detail how couriers consider the cost of their working hours, saying that when they work, they rarely think of freedom, "probably because I tacitly accept that not working is freedom, and work is the opposite, you have to work effectively, whether it is from employers, customers, or, when I run a self-employed business, observation and analysis of the market, etc., and then pay effective labor." to get a reward."
As a portfolio manager, my job requires me to read, write, and constantly judge under uncertainty, receive a lot of information, and process it efficiently and correctly as possible, translate it into action, and immediately get feedback from the market: maybe you have worked hard for months of research and waiting, and you still have losses. In the week-to-week decline of your net worth, it seems that all your labor is an ineffective struggle.
Regarding ineffective labor, David Graeber [9] wrote a book "Meaningless Work" to complain, but he died a little earlier, "big language models" like ChatGPT, or smart drawing tools such as Midjourney and Stable Diffusion, the entire AIGC (artificial intelligence content generation) technology ushered in an explosion in 2023, and practitioners of "bullshit work" seem to need no need to worry soon, "Alice" and Xiaoice" will take their place. The waves of the new era are too fierce, and Midjourney has achieved 10 million users and $100 million in operating income with just 11 people10.
Lu Qi[11], CEO of Qiji Chuangtan and former Baidu COO, recently gave a speech on the "big model", and he believes that any technology that changes society and industry is always a structural change. This structural change tends to turn a class of large costs from marginal costs to fixed costs. Lu Qi summarizes a "trinity structure evolution model", that is, "information-model-action", the operation process of this system can be broken down into four steps, that is, obtaining information, expressing information, solving problems and meeting needs. In his view, most Internet platform companies are still in the stage of transporting information.
I personally have an impossible triangle theory, that is, the sensitivity of receiving information, the accuracy of decoding information and the energy saving of processing information is an impossible triangle, for knowledge workers, under the constraint of limited time, expanding the breadth of reading and improving their sensitivity to receive information is the first choice. The process of reading information is the process of storing this information in the ontological consciousness, and at a certain trigger point where the conditions are suitable, we can restore this information again. Texts unfold the tangible world, from which we reconstruct it.
According to Brian Arthur,[12] one of the founders of complexity science, our minds are particularly good at making connections between things through metaphors, memories, structures, patterns, and theories. In other words, the mind is never a given, it is not an empty, passively loaded "bucket" of data. The mind itself emerges. Because of this characteristic, "whatever new technology must build on what is already there" [13].
It is precisely because of this that the current application of AIGC technology is also the capture and recombination of the existing Internet open content, but its rule trial and error and content computing ability are higher, which is based on the positive feedback research and development of computing power manufacturers represented by NVIDIA in the Web3 frenzy of the past few years: the influx of low-interest funds has brought success.
American biologist Andreas Wagner[14] once said that creation is, to a certain extent, the combination optimization of original things. Develop a standardized connection method and then combine them in all possible ways. Using this rule, nature creates proteins, creates regulatory circuits, creates metabolism, and creates life. This standard connection and rule, we can also think of it as a model.
Lu Qi believes that each of us is a combination of three models, cognitive model, task model and domain model, in the era of "big model", AI can take on part of the assistant work, people will evolve with digital, autonomous and automatic technology.
The biggest effect of AIGC technology is that it can efficiently compress information and call up knowledge, but this also brings "original" problems, that is, as individual creators, after contributing their own mental work, they become the "dregs" of large models? I think this prospect is probably the same as the "leek" in the capital market.
Frey writes that the Second Industrial Revolution brought a large number of semi-skilled jobs to American offices and factories, which provided the best assurance to those who feared unemployment As the wave of technological change swept through all workplaces, work became more comfortable, less dangerous, and more payable: workers had good reason to praise technological progress.
The relationship between technological progress and income distribution is not static: some technologies may increase inequality, while others reduce it. It depends on what level the technological change belongs to – substituting or enabling it. It also depends on whether the supply of workers with specific skills can keep up with demand.
Acemoglu and Restreper suggest that "AI to reinvigorate labor demand" could be the way of the future, such as personalized education and training, augmented reality that can effectively empower nurses, personnel and other health care providers by using AI to collect and analyze information, and augmented reality that allows workers and machines to collaborate on high-precision production tasks and comprehensive design tasks. But at present, the development direction of ChatGPT AI may be rushing in the direction of "intelligent assistant" and "co-pilot".
When thinking about how the traditional PDCA work cycle works in the new environment, I found that because a lot of junior work can already be handed over to this AI assistant to "run errands", essentially all users have become AI developers "outsourced markers", which also reflects Brian Arthur's insight that in the face of a globally used platform, at least the rate and breadth of marginal benefits increase is quite long, and this market, eventually dominated by one or two oligarchs is very likely. Microsoft, for example, is now taking advantage of the opportunity by investing in OpenAI.
However, as Zhao Tingyang [15] said, artificial intelligence mainly replaces, and will soon replace, is programmatic, regular, collective, repetitive and repetitive work, that is, basically based on statistical principles, or in a closed set to do multiple-choice questions, it replaces not physical labor, but medium-level mental work. "The current artificial intelligence in the Turing machine mode, no matter how fast the calculation speed and how strong the ability to collect data, the apex of its IQ will not surpass humans." "Because the methodology it works on comes from humans," he said. Just as we can't break the speed of light, human IQ is the IQ constant of the Turing machine. ”
I installed Notion's AI myself, and I asked it whether as an individual creator had contributed its own mental work, but had become a "drug dregs" of large models, and it answered it like this:
Although the application of artificial intelligence technology may replace some traditional knowledge work, it will also create more new fields of work and opportunities. Therefore, we should actively embrace AI technology as a tool to improve our work efficiency and quality, while constantly updating our knowledge and skills to adapt to new working environments and challenges.
Although this answer sounds like the AI version of "pertinent" and may not meet the answers we want and ease our concerns, one thing is certain, and that is that we must face up to AI and live with it.
Let's end this article with another quote from David Landers[16]'s:
History is like time, there is an arrow that cannot be turned back when shot; But unlike time, history is uneven, and it can only stumble forward intermittently. What is wise today may be wrong tomorrow. Development is a long-term thing; Logic, on the other hand, is temporarily valid.