After the release of ChatGPT before, there were rumors that "AI is about to replace humans", and with the release of Sora and Suno, more and more people believe in this statement - after all, Wensheng Wen only affects the work of authors and self-media, Wensheng video and Wensheng music, and even creative composition and editing can be done, and the scope of the impact is getting wider and wider. In this case, coupled with the slowdown in economic growth, more and more people are worried about their jobs.
But in fact, there is no need to worry, the current AI model, even if it continues to be optimized and upgraded to the extreme, still cannot replace human experts - because this is beyond the scope of AI capabilities - it has no emotion, no individual consciousness, and it cannot "emerge" the most critical human "value judgment" ability.
I got together with an old friend in the system on the weekend and accidentally talked about AI, and I looked at my friend's expression, which seemed to be sleeping for ten years with the "hibernation cabin" in "The Three-Body Problem".
Once again, I experienced the power of the "information cocoon".
It's not just my friends who are trapped in the cocoon, it's also me.
I take it for granted, thinking that everyone has "at least a general consensus" on social technology trends.
In fact, this "consensus" was much lower than I expected.
At the end of 2022, OpenAI released ChatGPT, which detonated a wave of artificial intelligence due to its "milestone breakthrough" in the field of text, which continues to this day.
After more than a year, to what extent has it developed, what can it do, and what far-reaching impact does it have on us?
There are more people who have no idea about this than expected.
In this article, let's talk about it briefly.
1. What are the substantive differences from the earlier "AI"?
Earlier AI can be grouped into two categories:
One is "single intelligence" that focuses on a specific domain.
For example, Deep Blue, which defeated the champion in a chess match in 1997.
However, this kind of intelligence, which can only be done within the scope of specific rules, can only be called "so-and-so self-optimization program", which is not on the same track as the current hot AI.
The other type is "AI" that directly interacts with human language or behavior, which is actually a search engine or recommendation algorithm.
What is the essential difference between large AI models and this type of "traditional AI"?
After using it for a while, what struck me the most was that the current AI really interacts and organizes content around "your needs", while the previous AI only focused on "keywords" (or behaviors) to search for content.
For example, I once saw the Zen machine "The Dragon Gate in the Sea" somewhere, and I wanted to find its source, but I didn't know the title of the work, and I wasn't even sure if it was a book.
So I used a search engine, and as expected, what I found was either "carp jumping over the dragon gate", or children's stories or music, in short, it didn't have anything to do with my goal.
Later, it was changed to an AI model, the same description, it told me: from the sixtieth chapter of the famous Zen book "Biyan Lu".
Well, it's concise and to the point, one step at a time.
This is where the breakthrough of AI comes in:
It "understands" human language, so it can search for content based on "the intent of your words" rather than "the keywords in your words".
This is more in line with the reality - when I need to search the most, it always comes when "I don't even know what to search for".
When I know the "keyword" of the target, what else do I want you to search engines?
The same problem also appears on various "smart recommendations", I browse a product frequently, and it always appears on the "you may like" list.
Does this fit in with reality?
In reality, if a person repeatedly searches for a certain type of commodity and forks it off, is it possible that he doesn't actually need this, but he can't find what he needs?
As such, these can all be classified as "fuzzy needs" - more frequent and more "rigid" needs in reality.
But earlier AI couldn't do anything about it.
This is where the "revolutionary breakthrough" of the AI model comes in:
It understands human language and can gain insight into your intentions, so I say it can search or organize content around your needs, and that's the biggest difference from earlier AI (around keywords).
Second, understanding human language means being able to output practical deliverables
Don't underestimate "understanding human language".
That's right, from the standpoint of human beings, kindergarten children can do it.
But the advantage of the machine is the almost infinite algorithmic computing power behind it, and "understanding humans" can be said to be the "index" on its forehead.
For example, the comprehensive capability of AI is based on "computing power" and "understanding humans" as the index.
Before understanding humans, this index was a 1, and it could only do the most dull and mechanical work. After understanding, this exponent can be 10, it can be 100, it can be N.
A one-time square of 100 million is not scary, but a hundred million to the Nth power, the number alone can drown people.
Based on "understanding human language", it can "learn more complex concepts" from a human perspective.
At first, you laugh at its cognition, like a newly invented train, which can run slower than a horse, but after a few adaptation and training, it can run with a force that thousands of horses can't pull, and even fly into the sky, with infinite stamina.
This is the advantage of silicon-based intelligence, it can "eat" endlessly, "internalize" endlessly, and there is no need to rest!
This is the "preliminary result" of current AI, which can already output all kinds of text that humans "need", including but not limited to:
Create text-only content such as official documents, novels, and poems;
Chat with you to relieve boredom, and even fall in love;
output paintings, music, and videos with plots;
Even website codes, forms, resumes, PPTs, etc., are directly used as workplace deliverables.
……
There are many examples on the Internet, so I won't go into them here.
Of course, we're not more concerned about what it can do – it's art if I draw a stickman.
What we care about is how good it can be?
How reliable is it?
Will the visible future replace the professional worker?
3. Has AI reached the level of replacing people in the workplace?
After a period of evaluation, what I have experienced the most is that the biggest advantages and disadvantages of the current AI are that the "AI traces" are obvious.
When I say "AI traces", I don't mean it in a derogatory sense, for example, let it make some suggestions or analysis reports, no matter what the field, it can almost hand over a "comprehensive", "balanced", and "essential" answer.
But that's exactly what makes it hard.
When we solve real-world business problems, the most taboo is precisely "balanced" and "comprehensive".
Just like the college entrance examination full score essay, it can only be full score in the exam paper, according to the requirements of the real workplace, the college entrance examination excellent essay is too "studenty".
In real work, no matter what kind of business, the essence of the most brain-consuming and most embodying ability is the problem of "resource allocation" - what resources (energy) do you have at hand for a project, according to what steps, what combinations, and how many proportions are allocated.
For example, let's design a promotional material that will be memorable. Experienced people know that the key to impressing people's hearts must be to highlight some and suppress some - that is, it must be "uneven" display.
Although "unbalanced" may not be reasonable, "comprehensive" and "balanced" must be wrong.
Therefore, in the work, the embodiment of the core competitiveness is in the "configuration", which is where experts need to "use their brains" the most, and it is also where their "value" lies - not to guide what should be done, but to eliminate everything that should not be done.
And what are the advantages of AI?
After the experts have allocated the resources, according to their requirements, fill in the details, as algorithms and computing power - in fact, these jobs are exactly the work done by the vast majority of junior XX teachers and XX assistants - do you think of who is the most threatening group of AI?
Of course, some people will say that as long as the magnitude of data training is sufficient and the allocation of expert resources is active, AI can also be competent, right?
That's not true.
I have come to this conclusion not based on the current capabilities of AI, but also on their future growth.
I'm concerned about their underlying "native absences" that inevitably lead to deficiencies in their abilities.
What's missing from AI?
It cannot possess human emotions, it has no individual consciousness, it has no desires, it has no motives.
Hey, isn't this a contradiction to the "understanding of human language" mentioned earlier?
Not really.
Understanding can be achieved through pure logic, as the AI will tell you:
I don't have emotions, but after I have fed "all human emotional conditions", I can know whether a person will be happy or sad or insensitively numb in a certain situation.
This is "understanding."
But understanding and experiencing are different dimensions.
Do you imagine that if a person is deprived of all sensory experiences (including fear), do you think he will have any desires, interests, or motives?
He became it—a machine with only rational thinking.
Without emotional experience, AI can't make value judgments on its own — all of its goals have to be "pre-imported" — but in the real world, no one can predict the optimal solution.
Not to mention a society or a specific group of people, even if it is a person, as far as you know "yourself" best, can you be sure that you will definitely want to eat the food you want to eat most now for dinner tomorrow?
Many people must remember that example:
There is a problem with a certain system, and technicians need to enter the building for repair, but this system is responsible for assigning entry qualifications, and technicians can't get in, and the doorman is blocked because technicians don't have entry qualifications.
This doorman is more like an AI - unable to make "common sense judgments" based on the current situation, which normal people would do.
Of course, AI is much more humble than a doorman, and it knows exactly what it "can't do".
On such issues, AI Ben I generously admits that:
It repeatedly emphasizes that "it is based on historical data and patterns", and that it is incompetent for the unknown or the real innovation.
AI can't make value judgments, so it can't weigh priorities in the current situation, so I just mentioned the core thing of experts - allocating resources.
Yes, AI can be trained to obtain a historically optimal configuration, but whether the historical configuration can match the current situation or the unknown territory requires value judgment and human beings.
* There are too many examples, for example, as long as there is an ethical conflict or a long-term interest (there must be an ethical conflict), AI is fine.
Therefore, only human experts can make this judgment - the AI has no bottom in mind.
In addition to emotion and consciousness issues, what else is AI particularly bad at?
Cross-domain associations.
As long as there are two superficially unrelated domains, it is difficult for it to fully explore the internal relationship between the two domains and obtain "wisdom insights".
For example, last week I was thinking about bullying at school.
After a lot of analysis, I found that I could not find the optimal solution by focusing only on the education system.
I've talked to AI about these kinds of questions, and all it can come up with is those comprehensive and balanced recommendations, and there is no breakthrough.
In fact, I quickly intuitively felt that I would look beyond the scope of school and education, look at the workplace, look at adult society, and find the key to breaking through the problem from those adult interactions that are not bullying on the surface.
But no matter how much I suggest AI, it never thinks that way.
The reason why "less overtime" has always been loved by old readers is obviously not because of my writing, but because of the "deep insights" occasionally caught in the topic, and most of these insights come from "cross-domain analysis", which is my personal core competitiveness.
For example, seeking measures for "willpower improvement" from "cybernetics";
Inspiration for the "overall application of fragmented time" from the characteristics of water flow;
or digging out "learning and persistent memory" from the "logic of chronic disease";
……
AI can still only "randomly correlate" such leaps related to substantive innovation – but because of the lack of value judgments, the economic benefits of AI to achieve the optimal solution among many "random outcomes" are too low.
Similar to "cross-domain association", another important ability is to "jump out of the domain boundary".
For example, if you want to study a complex question about the history of a country, you have to look beyond the history of that country to find answers.
诚如歌德所说，不懂外语的人，就连自己的母语也只是懂得一知半解（He who knows no foreign languages knows nothing of his own），也隐含类似意思。
Whether it is "cross-domain association" or "out-of-domain framework", it is the ability to solve complex problems—or, to put it bluntly, problems with great potential economic value—that rely on the most.
This kind of ability is difficult to obtain at present, because the gap cannot be bridged by "algorithms or computing power", it requires "value judgment", and the participation of motivation, interests, emotions, and individual consciousness - and this is the most substantial difference between AI and Human.
The above is about the "hard power" level.
If you take into account the "soft culture", as the old oilman in the workplace commented:
In many cases, the work is to serve the emotions of the boss/investor/leader, and it is to observe the words and feelings and the sophistication of people, and in this regard, AI is even more troublesome.
Therefore, the current AI model, even if it continues to be optimized and upgraded to the extreme, still cannot replace human experts - because this is beyond the scope of AI capabilities - it has no emotion, no individual consciousness, and it cannot "emerge" the most critical human "value judgment" ability.
From a certain point of view, AI is more like a "pure saint" who has completely abandoned low-level interests and does not even care about the meaning of life, although it can tell you all the factual knowledge, but it cannot guide a flesh-and-blood person how to live.
Fourth, will AI compete with others for jobs?
First of all, friends who do physical work can be ignored directly.
Using AI to realize delicate manual work is very ineconomical, and there is no need to worry about it in the future that is visible to the naked eye.
As for "mental work", it is indeed the main battlefield of current AI.
Extrapolating from the original defects of AI, it is not difficult to conclude that all work that does not involve emotions and value judgments, such as "following instructions", "step-by-step", "logical reasoning", "analysis and induction", "factual knowledge" and other tasks that are easy to be measured by "efficiency", are potential substitutes for AI.
According to the habits of human society, economic benefits are always the dominant goal. In other words, the higher the pay and the more automatable, the more likely you are to be the first to be "optimized".
At this point, it can be summarized as follows:
Regardless of social factors, AI will lead to a sharp reduction in the number of pits for basic posts, and at the same time, it will further improve the output quality of high-level talents.
Does this necessarily lead to high unemployment?
From a macro point of view, labor pain is inevitable in the short term, but in the long run, a large number of redundant personnel will inevitably generate new demand and markets.
This will create a large number of completely new jobs.
It's just that no matter what the new business is in the future, playing AI like today's "computer typing" is definitely the trend of the future profession.
5. How do we adapt to the post-AI era?
A while ago, I often saw all kinds of "training that teaches people how to use AI", and these people like to emphasize one "ability" - how to ask AI questions.
This reminds me of what the primitive tribes of the last century thought about airplanes:
These extant "hominids" always heard the rumble first, and then saw the plane, so they came to the conclusion that sooner or later they could summon the plane by imitating the boom sound.
Yes, for AI to output content, it must be driven by questions, but asking questions is only the most superficial thing, and the key to asking high-quality questions is never "questioning skills", but "the depth of your knowledge of the business domain".
The popularity of this kind of training makes people worry that it may not be long before AI replaces them.
Of course, the masses cannot be blamed for this.
Due to historical reasons, when our education model was first positioned, it was intended to cultivate "high-level talents" who "possess a wealth of knowledge" and can creatively solve problems in a step-by-step manner.
Who wants to cultivate the ability in this mode, it just so happens to happen to be the field that AI is best at.
Obviously, after the further liberalization and popularization of AI, a thorough educational revolution is also imperative.
Columnist
Li Shaojia, public account: less overtime, everyone is a product manager columnist. He is the author of "Evolutionary Operation", the proposer of "User Development Operation Framework Based on the Perspective of Users", an independent researcher of Internet business, and an expert in operation management.
This article was originally published on Everyone is a Product Manager. Reproduction without permission is prohibited.
Image from Pixabay and based on the CC0 license
The views in this article only represent the author's own, everyone is a product manager, and the platform only provides information storage space services.