“OpenAI automatically generates articles” has become a hot topic recently. Will human writing ability be replaced by AI?

Sharing is Caring

OpenAI automatically generates articles, In the past two days, the function of “AI automatic copywriting” based on GPT-3 technology has become a very hot topic; many people are competing to ask questions to let artificial intelligence generate articles, and some of these articles are really well written. There are samples. For example :

After observing some of these auto-generated articles, here are my thoughts.


Talk about the premise first

The current AI is very powerful, and with the improvement of computing speed, algorithm improvement, and the increase of learning materials, it will definitely become more and more powerful.

But what I see now is amazing, and the reason why everyone is amazed is “compared to the mentally handicapped AI in the past.” However, will it develop beyond the writing ability of “normal real people” in the future? Given time, it’s very possible.

Like most people, although I may not see the emergence of fully autonomous, thinking and judging AI in my lifetime, I have certain expectations for the future development potential of AI and the role it will play in human life in the future .

I am not worried that the popularity of AI will endanger the dominance of human beings in various jobs, because in the foreseeable future, before AI can think completely independently, no matter how strong its ability is, it will still be positioned as a “competent and loyal helper.”

( But I also said that if AI can be autonomous, “the three principles of robotics will be just wishful thinking of human beings”; this will be a philosophical contradiction that must be faced when continuing to develop AI in the future.)

What worries me more is the laziness of human beings after receiving a lot of AI assistance; not only jobs may be directly replaced by AI, but even their own positioning will become blurred.


AI Writing Now

Looking carefully at the things written by AI now, what makes ordinary readers think it is amazing is that the content of the article has become quite smooth and has a structural logic, which is great. But from the point of view of a real professional copywriter (me), a large part of it is still bullshit.

Appropriate pile of nonsense is one of the skills to make the article clear and easy to read; the fact that “one plus one equals two” will not read like “one person and another love is a beautiful two-person world” romantic.

Possessing this kind of rhetorical ability is a great progress for AI; coupled with the logical reasoning that was originally the strength of AI, it becomes the icing on the cake. A good article usually has the following elements:

  1. Logical inference: This can be divided into several layers, “because/so” has only one layer, “because/so, however/if not/then” is two or three layers, and so on; this kind of inference that can be written as program logic is Things that machines are inherently good at
  2. Quotations and presentation of specific and non-specific facts: e.g. “The Republic of China was founded in 1911” (specific fact) vs. “she was a sentimental person” (non-specific fact)
  3. Alternate use of realistic and unrealistic narratives: For example, “It is raining in Keelung today” (realistic) and “the stars in the sky do not speak” (unrealistic)
  4. Quotes and analogies reinforce the tone: “Russia’s invasion of Ukraine is like Germany’s invasion of France” or “She smells like a flower”
  5. Use adjectives, antithesis, metaphors, idioms, insinuations, contrasts, and other techniques to strengthen rhetoric, etc.
  6. Of course, there are many skills that humans have developed over thousands of years of culture.

From the content listed above, and comparing it with the content currently written by AI, we can probably know the extent of what it can do now and the direction it needs to develop in the future. As far as I can see, the current progress is probably only between 1 and 2.

This is why it was said earlier that “AI’s current strength is to learn to stack”. This is not a pejorative, but to point out that it has progressed a stage, but it is nothing more than that.

And because for many readers, this kind of stacking will be regarded as a simple “wisdom” and “writing ability” (they often evaluate real authors in this way); Progress”, developers may “intensify” at this point.

Maybe different types of AI will be derived in the future, some are used for data queries that do not need to be modified (such as the current Wikipedia), some are used for articles or copywriting that can be adjusted in the “fancy” level, etc.; then, this Not a problem at all.


What can AI writing do now?

For professional writers, making good use of AI will be a very convenient first draft generator; it can quickly generate ready-made paragraph structures and even arguments that they did not think of originally, which is equivalent to playing a part of “ghost writer” and “brain power”. Agitator” role.

But in terms of current AI capabilities, after the content is generated, it is still necessary to rely on your own judgment, rhetoric, and writing ability (that is, the previous points 2 to 6) to further refine, and use personal professional domain knowledge to verify the content. correctness.

In the part of “verifying professional domain knowledge”, AI, with its advantages in data connection and speed, may play a role in assisting data comparison or correction in the future, such as pointing out that “the first generation of Mac computers came out in 1986” (the correct answer It was 1984); perhaps more news fact-checking could be done in the future, or work to help improve the technical accuracy of the article.

However, judging from recent examples, the current AI is often still in the state of “grasping answers according to keywords, but lacking the ability to judge the authenticity and logical relationship of answers”. E.g:




if AI could talk

But if we don’t look at it from the perspective of “AI helps people write articles”, but “how AI expresses”, the current level is much better than before.

Take the matter of “natural language query” as an example. When Google, Yahoo, and other search engines such as AtlaVista and Lycos first came out more than 20 years ago, “how to search for answers in the right way” was still a science.

At that time, if you entered “what’s the weather in Taipei today”, you might not find the answer or display information you don’t need; you must understand the syntax of keywords such as “Taipei + Today’s Weather and Meteorological Bureau” to have a higher chance of finding it .

But search engines are smarter now. It will analyze the meaning of “what’s the weather in Taipei today”, present the answer you want, and integrate the display type of information:




Now that the search engine (and now the GPT-3 artificial intelligence) has figured out how to understand people’s questions, the next step is “how to tell the answer”.

A Google chart like the one above is one method; existing “voice assistants” (such as Siri) answer “today’s weather is ___, the temperature is ___ degrees, and it may rain.” The “non-AI fill-in-the-blank” answer is also available.

However, if we use AI capabilities to integrate these known data, and through structured, logical, and rhetorical expressions, turn it into voice to provide answers, and even chat with users based on the context (context), then the current performance will be the same. It’s already pretty good and deserves further development.


AI should make writers more industrious, not lazier

Go back to the perspective of the article/copywriter to look at AI.

When AI writing progresses to a certain level, although we can use it as an auxiliary tool, the free time should be used to continuously improve our abilities, and always go beyond what AI can do, instead of leaving it all to AI does it for you; otherwise, because “AI can do everything you can do”, the final end is to be replaced by AI.

Some business owners may think that because AI can already write decent things, professional writers may not be needed in the future, and it’s okay to hand over all the text work to AI, just like some business owners currently view computer translation same as tools.

If it’s just for cost considerations, maybe it’s okay to do so; but just like using machine translation, there are still a few basic principles:

  1. The finished product must be seen by at least one living person who understands the content, especially foreign languages ​​that I am not proficient in.
  2. Like any AI product ( such as autonomous driving ), there must be a “responsibility” mechanism in the entire process of producing and using the finished product (simply put, “who checks” and “who to contact when outsourcing”)

Because it is not impossible for AI to make mistakes, but it is impossible for AI to be responsible for your losses.

For example, Tenz Shih said ” Obviously college students cannot use OpenGPT to write reports “:



In these two short conversations, the answers provided by AI have many mistakes (although the first question is a trap question, the actor Morgan Freeman is still alive); It is quite satisfactory in terms of standards, but it cannot be listed as a reference at all, let alone used in official articles.

(Of course, AI also has many correct and amazing answers, but this “serious, but indistinguishable uncertainty” is the real problem.)



The above example also verified the points mentioned earlier:

  1. Although AI can now write articles with some architectural logic, don’t be fooled by the superficial gorgeousness, in fact there is still a long way to go.
  2. For the results to be actually used, at least one living person who understands the content must have seen it. You think some answers are “too good to be true” because you get it.
  3. Someone still needs to be held accountable. If students/employees are lazy to use AI to write homework, it is best to tell the teacher/boss that it is the fault of AI if something goes wrong.

This is not to belittle AI, but it is like a car, and it is not (yet) omnipotent; to play the best role without overturning, there are still certain limits, specifications, and the most suitable use of it method.

Sharing is Caring