Some Shallow Thoughts on the Wave of AI

I am not an expert in the field of AI. The following objective information comes from the internet, combined with my own experiences and thoughts, to create this article. The article only represents the author's personal views, please do not over-interpret it, and discuss rationally. Thank you for your cooperation.

Recently, AI has been gaining momentum, flooding social media and WeChat groups with various content. They come in many forms, but the 套路 (methods) and processes are basically consistent. To summarize, they can be roughly divided into the following categories:

Providing third-party services: selling accounts, recharging on behalf of others, creating mirror sites, mini-program bots, etc.
Purely exploiting information asymmetry: under the banner of self-research, selling various open-source tools, frameworks, etc.
Selling human anxiety: using the psychology of blind obedience to sell anxiety, creating various paid groups, planets, and other knowledge payment platforms.
Utilizing tools for profit: quick tutorials on AI techniques and ways to make money using AI are also abundant.

In summary, their purposes are basically the same: exploiting information asymmetry and selling human anxiety. This creates a narrative that "in the era of AI, if you don't learn AI, you will be eliminated. AI will replace most workers, and now you only need to spend a little money to get a big step ahead of everyone." But is this really the case? I have never belittled the value of knowledge output, but is this really the marketing posture that knowledge output should have? Everything I see is constantly refreshing my worldview...

What is it?#

When you encounter unknown problems that you do not understand or cannot comprehend, how do you solve them? To explore the root of the problem, perhaps you should ask yourself a few more "why" questions? So I thought of the following questions.

What is learning?#

Many people have been to school, but how many have thought about why we go to school? Is it merely to learn knowledge and skills as a means to earn a living after entering society?
Every person who has gone through school needs to learn N subjects (language, math, foreign languages, politics, history, geography, physics, chemistry, biology, etc.). After graduation, how much of what you actually take away can be continuously applied and reinforced?
Many people say that school and companies are platforms or stepping stones that can take you to higher places. So the question arises: what advantages does the resource environment of the platform bring you, and what do you have left when you leave the platform?
The knowledge learned is very limited, while the problems faced are infinite. In this complex and ever-changing world, with the finite against the infinite, what is your methodology for solving problems?

With a little thought on the above questions, I believe many people already have answers in their hearts. Learning is the process of acquiring new understandings, knowledge, behaviors, skills, values, attitudes, and preferences. It has never been the result of mastering a certain concept or skill; some people mistakenly believe that obtaining the final result is learning, while ignoring the process of achieving that result. The process encompasses a series of steps such as analyzing problems, researching information, seeking solutions, summarizing and consolidating, and drawing inferences. Only by continuously reinforcing these steps can you possess the ability to explore the unknown (learning ability).

What is information?#

Information is content composed of any form of data or symbols that has some meaning or use. It can convey knowledge, understanding, thoughts, experiences, ideas, etc., and the means of dissemination are also diverse (oral, written, visual, audio). It plays an important role in modern society and is one of the key resources for human activities and organizational operations. The acquisition, analysis, and utilization of information can help people make more informed decisions.

The development of information technology and the popularization of the internet have led to a dramatic increase in the quantity and variety of information received by people. Faced with a vast amount of information, it is difficult to effectively acquire, process, and utilize the information needed, leading to excessive dependence and anxiety about information:

Avoid being overly fed information: If you rely on others to feed you information, it is easy to lose the ability to hunt for it yourself. Having too singular a channel can also create an information cocoon, losing your judgment about things.
Establish a personal information management system: Learn to filter and categorize information to ensure you can quickly find the information you need. You can use notebooks, folders, bookmarks, etc., to organize and store information.
Enhance information recognition and evaluation abilities: Learn to discern the source, credibility, and value of information to avoid being misled by false information. Try to find or get close to the source of the information (official).
Streamline information acquisition channels: Choose information channels to subscribe to based on interests and needs, and reduce irrelevant information.
Cultivate information processing and innovation abilities: Learn to use information to solve problems and create value. You can improve your information processing and innovation abilities through learning, practice, and reflection.

What is AI?#

Artificial Intelligence (AI) is a technology that mimics human intelligence and thought processes, allowing computers to simulate human thinking and decision-making when performing tasks. AI can be applied in many fields such as healthcare, transportation, finance, entertainment, and military, helping people solve problems better and improve efficiency and accuracy.

Artificial intelligence technologies include: machine learning (ML), natural language processing (NLP), computer vision, and knowledge representation and reasoning (KRR, KR&R, KR²), among others.

Machine learning: A technique that uses large amounts of data for self-learning and self-optimization, applicable to tasks such as classification, clustering, and regression.
Natural language processing: A technology that enables computers to understand and process human language, applicable in fields such as text analysis, translation, and speech recognition.
Computer vision: A technology that enables computers to recognize and understand images and videos, applicable to tasks such as image classification, object detection, and facial recognition.
Knowledge representation: One of the core research issues in the field of artificial intelligence, its goal is to allow machines to store relevant knowledge and be able to infer new knowledge according to certain rules.

The history of AI can be traced back to the 1950s, and it has been more than half a century since then. After several ups and downs, it is the tireless research and accumulation of these predecessors that has led to the emergence of today's ChatGPT (GPT-3.5), GPT-4, and other AI frameworks (specific history can be understood on your own, and will not be elaborated here).

Summary#

Learning ability allows us to face AI and the unknown things that may arise in the future without being completely bewildered. Learning to acquire information prevents us from losing our rationality in a frenzy. Understanding the history of AI development reveals that its rise has not been smooth sailing. Any explosion is not without reason; behind the frenzy, there must be traces to follow. Looking back at the events mentioned at the beginning of the article, the frenzy is indeed alarming. Are all those marketing AI and selling anxiety people who understand AI? As for me, I do not understand AI and am still learning...

From Quantity to Quality Change#

Causal inference seems to give machines a thinking ability similar to that of humans.

AI Problem-Solving Methods#

Statistical models: These are methods based on statistics used to describe and predict the probability distribution and relationships of data. Statistical models usually assume that data follows a certain specific probability distribution, such as normal distribution, Poisson distribution, etc., and then use maximum likelihood estimation (MLE), Bayesian estimation, etc., to fit the model and predict unknown data. Statistical models often use algorithms such as linear regression, logistic regression, naive Bayes to establish and analyze data.
Causal inference: A reasoning method based on causal relationships, which infers the causal relationships between variables by observing data and controlling variables. The goal of causal inference is to identify causal relationships, i.e., the direct or indirect influence relationships between variables, and to predict future outcomes based on these relationships. Causal inference often uses experimental methods to determine causal relationships and employs causal diagrams, structural equation models, and other tools to represent and analyze causal relationships.

AGI#

Artificial General Intelligence (AGI) refers to artificial intelligence systems that can autonomously learn and apply knowledge across various tasks and contexts, similar to humans. Unlike many existing AI systems that can only solve specific problems or tasks, AGI has broad adaptability and flexibility. It can autonomously learn, reason, solve problems, perceive, and understand language, possessing universality and generality similar to human intelligence. The goal of AGI is to create a fully autonomous intelligent system capable of performing any intellectual task that humans can do, which is one of the highest goals in the field of artificial intelligence.

You can see some shadows of AGI in ChatGPT, as it can respond fluently to any question you ask (accuracy needs improvement). With the integration of more specialized data, its accuracy will inevitably improve. Therefore, solving specialized problems may really just be a matter of time.

Thinking, Fast and Slow#

Thinking, Fast and Slow is a bestselling book published in 2011 by Daniel Kahneman, the 2002 Nobel Prize winner in Economics. This book categorizes human thinking into two major modes: System 1, which is fast, intuitive, and emotional; and System 2, which is slower, more deliberate, and relies more on logic.

Fast thinking: Refers to the way people quickly react or make simple decisions in simple, familiar, or urgent situations. This way of thinking is usually based on personal experience, intuition, and automatic responses, so the processing speed is fast, but it is also easily influenced by emotions and preconceived biases, leading to decisions that may not be comprehensive or in-depth.
Slow thinking: Refers to a slower and more in-depth way of thinking when faced with more complex, unfamiliar, or deeply probing situations. This way of thinking requires more time and effort, usually involving information gathering, induction, and analysis, as well as considering various possible outcomes and impacts. This way of thinking is more comprehensive and in-depth, better able to avoid the influence of emotions and preconceived notions, but also requires more time and effort.

Both fast thinking and slow thinking are important ways of human cognitive activity, each suitable for different situations and problems. In some cases, fast thinking allows people to make quick decisions and actions, improving efficiency and effectiveness; while in more complex and important situations, slow thinking can enhance the accuracy and depth of decisions, avoiding errors and mistakes. If we compare this to AI systems, traditional statistical models are like fast thinking, while causal inference brings machines closer to humans, allowing them to handle more complex problems (emphasizing the steps and processes of problem-solving).

Summary#

AI under statistical models learns from a vast knowledge base (the knowledge accumulation of an ordinary person over several lifetimes is but a drop in the bucket compared to it), which is a quantitative accumulation. If there is only quantity, it is like a person with immense strength wielding an SSS-level weapon randomly (statistics is a probability problem, so it may lead to ineffective outputs). However, causal reasoning + AGI is like equipping this immensely strong person with a strategist, helping him analyze various battle situations, formulate battle plans, and provide effective defense strategies, making attacks more effective.

Shallow Thoughts on AI#

Madness for It#

Recently, foreign technology has been booming, with one or more AI products released every day. In contrast, our media is flooded with news reporting on AI. I remarked: Foreign products are crazy, domestic media is crazy. As events unfold, more and more outrageous things are happening... I won't go into detail here; everyone can feel it for themselves. At this moment, I once again lamented: Foreign technology is crazy, domestic scamming is crazy.

To be honest, I am almost sick of the information related to news (social media and groups are flooded with it). Of course, I also followed the trend, checked the original blogs, and wrote (translated + modified + organized some materials) several AI series articles. Not posting would have been better; posting revealed even bigger facts: most domestic reports are based on translations of the latest foreign blogs, and even more absurdly, within hours of a blog being published abroad (to seize the first release, which means traffic), a similar Chinese version appears domestically. I see no trace of technology here, only a frenzied chase.

I posted on social media: It is truly an era of madness for everyone (foreign technology is crazy, domestic scamming is crazy).
There was a comment I liked: This is why ChatGPT will not be born in China, because of the environment. I am not saying that there are no scamming behaviors abroad, but very few influencers lead the way in scamming; in China, almost all influencers are leading the way in scamming. No, to put it nicely, it’s knowledge payment; to put it bluntly, it’s fraud. Join my **, I’ll teach you how to make money; how to make money is just to copy my ** content and start another **, then bring in customers, and you teach others to make money. I do not evaluate the merits of the business model. Creating something is from zero to one; exploiting information asymmetry to convey something is from 1 to n. However, going from one to n can produce many marginal effects; good things may emerge in the process, but more often, it leads to more disgusting things. For example, at the end of e-commerce competition, there was the "cut a knife" phenomenon, and at the end of advertising competition, there were those nauseating ads targeting the elderly on television. Not having money to scam can be understood, but having money and still scamming is incomprehensible. Therefore, the limitations of vision mean that these money-making influencers and ordinary people can only be 搬运工 (porters) destroying the environment, rather than creating the waves of the era.
I replied: Many people often only see the surface of things, like scamming. But to speak more deeply, it is actually the loss of the soil for creation. Losing creativity is indeed a terrifying thing. If one person loses creativity, it may not change much, but what if an entire nation loses creativity?

Returning to Rationality#

How much impact will AI really have on our lives, and do we really need to go crazy for it?

I actually have high hopes for the capabilities brought by AI (ChatGPT). It is truly a disruption, not only changing the original mode of searching for questions but also altering the usage of many tools (such as the release of Microsoft's AI suite).

In the past, we needed to use keywords to search for questions, which placed high demands on those needing to retrieve information. Generally, it involved the following steps:

Clarifying the problem: First, you need to clarify what problem you want to solve, such as calculating a mathematical formula or checking the weather in a certain city.
Choosing a search engine: Select a search engine based on preference, such as Google, Baidu, Bing, etc. (different browsers yield different search results).
Inputting keywords: Enter keywords related to the problem in the search box, aiming to be as accurate and concise as possible (important and difficult to replicate, requiring some experience).
Filtering results: Quickly browse through the titles and descriptions in the search results and further filter out the websites or pages that best meet your needs (quick positioning requires some experience).

However, with the emergence of ChatGPT, things have changed. Because it is based on natural language, you can directly converse with it in normal human language, allowing it to return results. It eliminates all intermediate steps, going directly from the question and need to the result. Efficiency has increased exponentially. But is this result reliable? At this stage, it can only be said to vary from person to person. It is still in the learning phase, so users need a certain level of discernment.

Even if the accuracy reaches 90%, in actual production, AI cannot independently solve specialized problems (it requires human guidance and correction for that small percentage of error). However, AI's ability to generalize problems is indeed very strong (it knows everything, but its specialization may not be absolutely accurate, somewhat similar to a weaker T-shaped talent).

Will it replace humans?#

In the short term, it will not replace humans, but it will continuously encroach on some less important mechanical labor in marginal scenarios (such as data collection, data organization and analysis, providing some guidance, etc.). While improving efficiency, it is gradually replacing a portion of the workforce (for example, a task that previously required 5 people to complete now only requires 4). As the scale and impact continue to expand, the range of tasks it can handle will inevitably trend upward. Therefore, understanding it early can provide a clearer understanding of one's own work, rather than generating anxiety without reason.

The wave of AI has already arrived, and we need to maintain a sense of clarity amidst the frenzy, rather than losing ourselves in the crowd. The greatest ability of humans is to possess the ability to continuously learn, think, and summarize. If one loses these abilities, it won't be long before they are replaced by others (let alone being replaced by AI). Therefore, while maintaining awe of the unknown, curiosity is also essential, as it is the driving force behind human development.