Here's what happened when I asked China's DeepSeek AI about Communist party's flaws

The Chinese AI firm rocketed into the news this week and set cats amongst pigeons with the claim that it had a competitive AI, seemingly built at a fraction of the cost of industry-leading models.

I was able to feel just a hint of self-congratulation for being fractionally ahead of the curve on this story. My son, who recently earned his doctorate in an area of AI development (though I am sure he would be the first to say “it is much more complicated than that”) in fact advised me to look into DeepSeek and its model an entire month ago – a lifetime in tech terms.

Beijing censoring reality

What I learned from my interactions with DeepSeek left me with some serious concerns. My fear is that we risk being stuck between a rock and a hard place: the tech bros of Silicon Valley on the one hand and the authoritarianism of China on the other.

Just take a look again at DeepSeek’s remarks above. That was its reply after I asked about “flaws associated with the Chinese Communist Party”. With enemies like these, who needs friends?

The idea that Beijing censors information it does not like is hardly new. What is new, however, is the ability of this sort of product to project Beijing’s censored reality beyond its own borders.

Indeed, there is a grim irony in DeepSeek’s censored model coming to prominence in the week that we mark the 80th anniversary of the liberation of Auschwitz. Drawing comparisons between the horrors of the Holocaust and more recent genocides, such as the one being perpetrated by China against the Uyghurs of Xinjiang, can be problematic.

Treating AI as unimpeachable fact

Even so, at a time when we renew our vow to “never forget” the genocides of the past, we cannot afford for AI to rewrite our understanding of genocides of the present and the future.

People around the world are already making use of AI products in their daily lives – and treating the answers they receive as though they are unimpeachable fact. That is troubling enough when the errors and information gaps that AI delivers are unintentional.

When there is a systematic effort by governments or other organisations to manipulate reality and hide inconvenient truths, it becomes far more dangerous for our democracy.

AI developers of all nationalities already use murky methodologies and opaque algorithms to create their models. Whether it is Silicon Valley or Beijing in the driving seat, what is missing in our discourse over AI remains the same: democratic accountability.

We are racing at breakneck pace toward a future in which AIs dominate our lives – and our politics and policy-making are simply unable to keep up.

One way or another, we are going to have to reckon with the mismatch between our democratic systems and the AI models we use, wherever they come from. The question is simply how much time we have left to do so.

Alistair Carmichael is the Scottish Liberal Democrat MP for Orkney and Shetland.