Kathrine Frich
The virtual assistant currently serves around 40 million active users.
Artificial intelligence can do some incredible things nowadays.
It drafts emails, writes software code, and answers complex questions in seconds.
But sometimes the technology moves a little too fast for its own good, revealing far more than its creators intended.
Predicting the future
A popular Russian virtual assistant recently found itself in an awkward spot. When asked about upcoming national elections, the software gave a detailed rundown of the winning party and its huge lead over competitors.
There was just one glaring problem. The votes had not actually been cast yet.
According to Komputer Swiat, Bulgarian investigative journalist Hristo Grozev decided to test Alice AI, a software created by Russian tech giant Yandex. He asked the chatbot about the upcoming State Duma elections scheduled for mid-September.
Without missing a beat, the software confidently declared that Vladimir Putin’s United Russia party had already won. It reported that the ruling party maintained a constitutional majority while “significantly outperforming its rivals.”
An honest confession
Puzzled by the claim, Grozev pressed the software further. He asked how it could possibly know the final tally of a race that was still days away.
The bot immediately backtracked. “I made all this up. None of this data about the 2026 elections is true. It’s easy to make this up in elections, because there are always certain patterns. A mixed system, a large party, electronic voting,” the artificial intelligence admitted.
That frank answer highlights how computer models process information. AI programs rely on vast amounts of historic data to predict word patterns. In Russia, the same political party has dominated the ballot box for over two decades.
Grozev pointed out that the software was simply following its training. “The AI does not know that a different outcome is even possible,” he noted.
Questions over bias
The virtual assistant currently serves around 40 million active users. Built as Russia’s homegrown answer to ChatGPT, the platform has faced intense scrutiny before.
A recent study examined how the software responds to queries about the ongoing conflict in Ukraine. Researchers found that 86% of its Russian-language answers echoed official Kremlin propaganda. When users asked questions in English, the program simply refused to answer 86% of the time.
For now, tech developers are left cleaning up the messy output. Automated tools might seem smart, but they ultimately mirror the environment that created them.
Sources: Komputer Swiat
