The startup predicts the arrival of the “Technological Singularity” in seven years

Scientists working at a translation company say the famous phenomenon known as the “Singularity” is coming soon. Quite a bold claim, given that no clear benchmarks have been set on the subject.

Have you ever heard of the concept of “technological singularity” or more simply “Solitude”? It’s a long-debated hypothesis that AI may reach a point of self-improvement, and then we’ll be faced with unexpected upheavals affecting the entire human society.

The term was mentioned as early as the 1950s, but the English statistician Irving John Goode gave a definition in 1965 that may come close to its current meaning: “ Let’s say that a super-intelligent machine, no matter how genius, is a machine capable of surpassing humans in all areas of intellectual activity. Since the design of such machines is one of these intellectual activities, a super-intelligent machine can design even better machines; then there will certainly be an “intelligence explosion” that will surpass human intelligence very quickly. Thus, the invention of the first super-intelligent machine [théoriquement] the last invention that man had to make “.

In other words, once AIs are intelligent enough, they will be able to upgrade and update themselves to constantly improve their capabilities. From this turning point, the role of the human species can be questioned.

Recently, a group of Italian scientists claimed that we are not far from this critical point. To be precise, we still had seven years before we got there. However, their claim is based on facts that are specific to their field: translation. The company is also called Translated.

“ Many AI researchers believe that solving the problem of language translation is the closest thing to producing artificial general intelligence (AGI). “, we can read in the press release of this company. Their argument: language is a very natural ability for humans, and vice versa, it is difficult for machines to master. For them, the ability of machines to reach or even surpass the ability of humans to translate language. Therefore, “Artificial General Intelligence ( AGI)” can be a reason to determine the status. In other words, a program capable of performing any task that a human can perform.

“Edit time” as a measure of artificial intelligence

A graph showing the evolution of the time it takes a human proofreader to edit the raw version of the text generated by the machine translator. © Translated.

To announce this famous arrival of the singularity, the scientists of this company chose to base it on a specific measure, Time to Edit (TTE). This metric represents the time required by professional human proofreaders to check AI-generated translations compared to human translations. The idea is that if this time AI-generated translation is more profitable than human translation, then we’ll hit a tipping point, according to Translated. However, the necessary correction period has been falling for years. If we follow this graph, in 2015 it took a human translator about 3.5 seconds per word to achieve an optimal translation. In 2022, it is close to 2 seconds. And it’s going down: so that’s the idea of ​​peaking in seven years.

This output is certainly interesting in relation to the field of translation and says a lot about the developments in the sector. However, it remains part of a very specific field and is based on an assumption that has yet to be debated: human language as the absolute reference for determining the “intelligence level” of artificial intelligence…