The idea of "technological singularity" is a concept that has always been buzzing in the heads of enthusiasts and experts: with an aftertaste of concern. Because it describes the moment when AI escapes human control, surpasses our capabilities and transforms everything we know.
The difficulty of foreseeing the arrival of the technological singularity and of imagining what could happen beyond this technological "event horizon" makes us feel like we are facing a black hole, with the sensation that sooner or later it will swallow us up.
What time is the end of the world?
For some time, experts in every sector have been trying to identify possible signs of approaching the Singularity, to try to predict when events will reach a "breaking point".
One of the most often cited metrics is the ability of an artificial intelligence to understand and interpret language (written and spoken) with the accuracy of a human. Language is one of the most difficult challenges: for this reason, achieving this goal could indicate the presence of an AGI, Artificial General Intelligence. What a (now ex) Google engineer prematurely attributed to LaMDA a few months ago.
According to Marco Trombetti the CEO of translated, an Italian company in the field of AI translation, the time is right.
Technological singularity, just a few years away?
"Language is the most natural thing for human beings," says Trombetti. "The data we have before usHowever, they show that machines are not that far from closing the gap."
Translated has been tracking the progress of its AI with a metric called "Time to Edit" (TTE). This metric calculates how much time professional human editors spend correcting and optimizing AI-generated translations compared to those made by other professional human translators.
In the last 8 years of monitoring, the analysis of over 2 billion optimizations shows a gradual and continuous improvement, which makes artificial intelligence ever closer to the characteristics of a human translation.
How close?
Much. In 2015, it took professional editors an average of 3,5 seconds per word to check a machine translation. Today they only take 2 seconds. If this trend continues, an artificial intelligence (that of Translated, specifically) could reach the same quality of human translation by the end of this decade.
just 7 years old. It's the first time anyone in the field of artificial intelligence has predicted the velocity to the technological singularity based on such a calculation.
2030, the deadline of the technological singularity. Or not?
I agree that the understanding and use of language are very important parameters among those that will help define a general artificial intelligence.
And I don't think they're the only ones. They don't necessarily imply that a machine is intelligent (since we don't even all agree on what exactly "intelligence" is).
The estimate made by Translated has its own importance, however, regardless of whether or not the results of its AI are precursors of the technological singularity.
A machine that can translate language perfectly, and do it exactly like a human being will have a significant impact on society.
Then we will understand if behind his answers there will be "someone" or, as I believe, not.
