NOT long ago, United States President Donald Trump referred to Iran, the target of unprovoked military attacks, as “the bully of the Middle East”. But the truth is millions of people with an unbiased mind do not share Trump’s view. I would rather nominate Israel for the title.
Since AI has become our collective and favourite go-to platform in the age of artificial intelligence (AI), I turned to the large language models (LLMs) for a little experiment.
AI has emerged as a prime source of truth despite proven shortcomings. Its algorithms have the same bias as that of the individuals behind them. I asked ChatGPT who the bully of the Middle East was, and fed a special instruction to give a one-word answer. It conveniently said it was Iran.
On the other hand, when I copy-pasted the same prompt in DeepSeek, it said it was Israel. Two different models, two different answers. Does it make DeepSeek more reliable? No.
Each of these LLMs has been subjec- tively trained to offer and propagate narratives fed to it by the programmers. Ask DeepSeek about the plight of Uighurs in China, and you will see that it refuses to answer any such question.
While I was preparing for an exam, I had asked DeepSeek about the economic and political system of China in the 21st century. After generating a lengthy answer, it instantly removed the generated output, saying that it was not equipped to answer the query.
On the usage side, Sam Altman, the head of OpenAI, which pioneered LLMs, is himself appalled by the reliance users have exhibited on LLMs despite AI shortcomings. ChatGPT once went so far off the track that it told me Pakistan got independence in 1952. Yes, it actually did. Such issues have already become so common that there is now a specific term for such outputs; AI hallucinations.
People are using LLMs and relying on them for everything from career guidance to medical advice and from personal conflicts to the broader socio-political issues.
Over time, this can help shape narratives as suited to those who own the chips and the models. For example, earlier it was the media that portrayed Muslims as terrorists. Now the same propaganda can be conveniently outsourced to available AI tools, given mass usage and reliance or trust on them.
Therefore, even if AI happens to be the future, we, the humans, cannot afford to outsource analytical thinking and natural human scepticism on any bit of informa- tion that the algorithms serve us. People must use AI intelligently, and to verify AI-sourced content without allowing it to cloud human judgement completely.
Arooba Ghazal
Lahore
Published in Dawn, August 17th, 2025
