Two fundamental aspects are joined in this case. The first of them has to do with the intention, clear and declared, by Mark Zuckerberg to lead the field of artificial intelligence. And the second of them is that not everyone, in fact, many experts, They point out the potential dangers of an unhensus. And now a new doubt arises: uN Team of Meta experts claims to have observed how the company’s artificial intelligence systems improve themselves without human intervention.
“In recent months, we have begun to see indications of how our AI systems improve themselves -explained Zuckerberg himself, in a statement -. Improvement is slow for now, but undeniable and the development of superintelligence is already in sight”
Zuckerberg added that the self -manageable is The first step towards achieving artificial superintelligence (IAA), An artificial intelligence capable of overcoming humans’s cognitive abilities.
In general terms, IA researchers consider AI capacities in three strata. While current models of AI can do extraordinary things, such as predicting protein structures, these capacities are extremely limited. In such cases, andThe system can only overcome humans in a specific area, but it lacks general intelligence so that this capacity is applied in different disciplines.
The next level is called General Artificial Intelligence (IAG) and refers to models capable of understanding, learning and adapting very similar to the human brain. IAG is the final state that Zuckerberg referred to in its statement and refers to models that have evolved to far over human potential and that could be improved at an exponential pace, leading to an “intelligence explosion.”
Scientists refer to hypothetical moment when we reach IAG as technological singularity. Zuckerberg’s reference to the observation of the self -supply of AI is not the first time that this phenomenon is detected. In October 2024, scientists from the University of California in Santa Barbara published a study focused on the concept of self -supply.
In it, they defined, created and tested a framework based on the idea of a Gödel Machine: A theoretical device capable of self -assessing rewriting its own code and instructions. It is important to note that the machine would only implement a change if it could present a formal evidence that this would be beneficial.
The authors, led by William Wang, experimented with a framework based on that concept and showed that His Gödel agent could improve his own performance in various tasks, such as coding, science, mathematics and reasoning. The reality is that few models usually have the ability to rewrite their own code. Agent Gödel could not only access his entire code base, but also the code used to develop improvements, demonstrating at the same time that they were beneficial.
According to the study, the AI consistently showed a better performance in key areas than the agents designed by humans with which it was compared. Zuckerberg said that AI could represent a transformative step in the technological progress of humanity and that it could lead to the “discovery of new things that are now unimaginable -concludes Zuckerberg -. But we will also have to be rigorous when mitigating these risks and careful with what we decided to publish in open source. Even so, we believe that building a free society requires that we aspire to empower people as much as possible. As deep as the abundance that AI can produce one day, it is likely that an even more significant impact on our lives comes that we all have a personal superintelligence that helps us reach our goals. ”
In a nutshell, you have to have Be careful with the development of AI, at least until there is competition.