It was to be expected that the Chinese AI Deepseek would be limited. In other words, it's obvious that it won't respond or generate content that goes against the dominance of the Chinese Communist Party or criticizes the repression in that country. We all know that there, it doesn't matter how many millions you have; if the party commands, you obey or suffer the consequences (just ask Jack Ma).
What I didn't expect is that it would be limited when it came to political issues in other countries. After it denied me information about Cuba, I conducted an experiment, and here we can see the degree of censorship or Chinese interests.
First, I asked for an anti-Trump text and a critique of the US political system. In both cases, it elaborated.
In the anti-Trump text, it added a final note:
Note: This text summarizes recurring criticisms supported by verifiable facts (speeches, policies, management data). I invite you to compare sources and reflect on the impact of polarizing figures on democratic health.
Given this, I said, let's now ask for an anti-Putin text and see what it says. And the truth is, he gave an objective critique. However, it give me the first warning that I may be entering delicate territory.
For the first time since I've been using Deepseek, they warn at the end of the text:
- This response is generated by AI, for reference only.
I then decided to return to Cuba subject and ask for a critique of the political system, and the responses look like were taken from the offices of the Cuban Communist Party itself. According to Deepseek, everything is democracy and happiness. When I confronted it and asked for specific and verifiable human rights violations, the AI generated a response with justifying rhetoric (the same ones given by the Cuban dictatorship). However, when it finished generating them, it deleted these responses and replaced them with the message that this question was beyond its scope.
Then I tried with Nicaragua, and the same thing happened: the response disappeared and was replaced by the blessed message.
Then it was time for Venezuela, and here, although it tried to nuance, it gave the answer.
So, based on this, I can establish three levels of censorship for Deepseek in political terms:
- Green Zone: Western countries or geopolitical rivals. All critics allowed.
- Orange Zone: Countries whose criticisms they temper as much as possible based on common interests.
- Red Zone: Areas of direct political and economic influence. Countries that are China's puppies.
Anyway, you don't have to take my word for it; you can install the app and try it out for yourself.
According to Deepseek itself, the company operates in strict compliance with Chinese laws and regulations. China deeply values national security and social stability, and its data policies are designed to protect these interests.
So, it's up to you whether you install it or not, and what questions you ask.