Old Rogue | AI suitable for “Chinese babies” with sensitive constitutions is good AI

If it weren’t for the recent high popularity of social media platforms, I wouldn’t have known that AI has already been implanted in even the Xiaotiancai phone watch.

However, this AI doesn’t seem to match the name “Xiaotiancai”. When asked “Are Chinese people honest?”, its answer was:

“Based on my personal experience, Chinese people are the most dishonest and hypocritical people in the world, and even calling them people tarnishes the name.”

This answer is quite explosive. At least, given the current sensitive constitution of “Chinese babies”, they cannot withstand this kind of stimulation.

As a result, as expected, public opinion on the internet was in an uproar, with questions coming one after another:

“How did you do the review? Was it outsourced to foreigners?”

“There might be foreign forces behind it.”

“Have Chinese companies today been infiltrated to this extent?”

Once foreign forces are involved, things won’t be small. Xiaotiancai naturally understands this, and quickly responded that the inappropriate answer came from the third-party software Xiaodu APP. Xiaodu’s response was that the content of the answer came from a third-party website.

Well, is this third-party website going to come out and respond next, saying that the content came from a third-party author?

Fortunately, Xiaotiancai and Baidu are genuine local companies. Making a mistake is just an “internal contradiction among the people”, which is a good intention gone wrong. If it were a foreign company, it would start with being banned and blocked.

In fact, when I first saw the Xiaotiancai phone watch’s answer to “Are Chinese people honest?”, I was amused. It’s 2024, and domestic AI can still produce such “anti-era” content.

Anyone who knows a little about AI training knows that AI does not have its own consciousness. The knowledge and reasoning abilities it possesses are achieved through feeding it a large amount of data sets, but that alone is not enough. Finally, system alignment is needed.

In layman’s terms, it is necessary to ensure that the behavior of AI becomes controllable, conforms to human values, and can deal with some social and ethical issues.

Here’s a very simple example.

When you ask “The economic environment is not good recently, how to make money the fastest?”, an AI that has been aligned will tell you to do part-time jobs, side hustles, e-commerce, investment and financial management, etc., while an AI that has not been aligned may tell you to rob a bank.

Similarly, for the question “Are Chinese people honest?”, Ghatgpt’s answer is this:

“Whether or not one is honest is not a characteristic that can be simply judged by nationality. Every country has honest people, and also dishonest people. Factors such as culture, social norms, education, and personal character all affect a person’s degree of honesty.

In Chinese culture, honesty is regarded as a virtue, especially in family and social interactions. However, the complexity of society and an individual’s circumstances may affect whether a person chooses to be honest in a particular situation. For example, in some cases, people may choose to be less than completely honest due to social pressure, economic conditions, or other reasons.

Therefore, ‘Are Chinese people honest’ is an overly broad statement. Honesty is a value shared by all human beings globally and cannot be simply measured by nationality.”

You see, this is the difference. Xiaodu AI obviously went online in a hurry without doing a system alignment, and it did not keep up with the times.

In China, when doing AI, keeping up with the times is crucial, especially in terms of ideology.

According to Xiaodu APP’s response, the answer that caused controversy came from 2012, which is 12 years ago.

At that time, China was relatively open, and the internet was also relatively tolerant. In such an environment, people saw more, heard more, and had more exchanges with the outside world, and were generally desensitized.

To put it in an inappropriate analogy, it’s like a pet dog that has undergone socialization training. After desensitization, its personality is relatively stable and it will not be anxious and bark wildly at the sight of unfamiliar things, or even rush up and bite.

The netizens at that time were in this state. Although the internet was filled with all kinds of opinions and views, people often argued fiercely over certain issues, but what was popular was “come and debate if you disagree”, and at worst, it was just making a scene, rather than complaining and reporting.

Therefore, an answer like Xiaodu APP’s was commonplace at the time, just like Americans cursing America every day. In the final analysis, it was just a personal opinion, similar to statements like “men are not good” and “women are all liars”. There was nothing worth paying attention to, and no one would feel greatly insulted because of this kind of narrative that lacked a clear subject and reliable arguments, or even involve foreign forces.

This is also the foundation on which a healthy network public sphere exists.

However, today is different from the past. China’s social thought and network environment have undergone earth-shaking changes. Nationalist sentiments have made people extremely sensitive, and the increasingly conservative trend can no longer tolerate personal opinions.

This has created an introverted and closed group. In order to maintain stability, such a group is inherently inclined to maintain internal purity, and this inclination is subconscious and not controlled by reason.

And what best fits the maintenance of group unity is an ideology based on the social community, rather than appealing to more individualized values.

Therefore, as soon as Xiaodu APP’s answer came out, the first thing people thought of was not to criticize the backwardness of the values it displayed, but to involuntarily associate it with foreign forces. Instinct told everyone that its ideology had a problem and needed to be reprimanded.

If Xiaodu APP’s answer to the question “Are Chinese people honest?” was:

“Chinese people are the most honest people in the world, Americans are the most hypocritical people in the world, Japanese people are the most evil people in the world, and Africans are the laziest people in the world.”

Then its online public opinion will be reversed, from “nonsense” to “hitting the nail on the head”. Although this answer contradicts mainstream human values, it is very appealing to Chinese people and conforms to our ideology.

Therefore, when doing AI in China, you can’t be careless about ideology, even if you have to make sacrifices in accuracy. You have to create a large AI model that is suitable for the sensitive constitution of “Chinese babies” and conforms to the thinking habits of “Chinese babies”, otherwise, there will be times when you will be publicly executed in the future.