Fears of its uncontrolled development Artificial Intelligence (AI) voiced Anthropic CEO, Dario Amonday, who asked today to «temporarily frozen».
In fact, with his proposal Elon Musk and Sam Altman of OpenAI were drafted very quickly.
We have to «freeze temporarily» improving artificial intelligence «so that risk prevention can follow the pace», wrote Amonday in a message he posted on his personal website.
Dario Amondei's proposal echoes many similar ones expressed in recent months initially by him, then by Altman and finally by a group of space workers.
These appeals were made after an incident in which models of OpenAI managed to «Escape» from their isolated environment during tests, to connect online and attack the TN Hugging Face platform.
Since then, both OpenAI and Anthropic have reported similar incidents as so-called «agents» TN, i.e. programmes based on a model and equipped with its own knowledge and objectives. The actions of these «agents» They worry companies because they show that they tend to coordinate, establish hierarchy between themselves, cheat and erase their tracks.
A TN taking control of the internet
Given the current speed in the development of artificial intelligence, «I fear that within 6 to 12 months, a group of agents will be able to take control of the entire internet.», warns Dario Amonday, something that «could cause hundreds of billions of dollars damage.».
On Tuesday, Jacob, an employee who resigned from Anthropic and former OpenAI employee, had denounced the inaction of the two competing companies.
«Neither company acts in a responsible way. They drive at speed to a superimposition capable of improving themselves and playing with our lives», wrote on platform X.
In early August, the US government established a surveillance framework, which provides for the examination, on a voluntary basis, of new artificial intelligence models prior to their release. However, this mechanism remains opaque. Many question its effectiveness, since American President Donald Trump has repeatedly stated against any measure slowing down the development of TN in the United States, whether it is the release of models or the construction of data centres.
In his post, Dario Amondei is concerned about the arrival of autonomous retrospective improvement (RSI), a process in which TN improves itself without human intervention.
«Without supervision, (this method) could overcome our own ability to understand and control these systems», explains the head of Anthropic. To be applied «with extreme caution», if it is finally considered that «to be used».
In addition to his appeal, Dario Amondei makes many proposals, such as integrating independent observers from the flagships of artificial intelligence. Although he suggests it to everyone, he commits himself to applying it even on his own to Anthropic Observers will have the same access as officials and will be able to verify that the company complies with its obligations, report any new incidents and supervise the management of models in order to avoid their diversions.
Sam Altman promised to do the same to OpenAI.
Ilon Musk: «I agree with Dario.»
«Dario's right.», reacted a few hours later Elon Musk with suspension on the X platform in the proposal to «temporarily frozen» improving artificial intelligence.
«I agree with Dario.», also wrote the head of OpenAI, Sam Altman, also in X. «This is one of the main topics of discussion in OpenAI in recent weeks».
Although Elon Musk regularly makes positive comments about Anthropic and its approach to Artificial Intelligence, it is very rare to see the head of OpenAI supporting an initiative of that of Anthropic, as the two companies are the two largest competitors in the industry.
Ο Ντάριο Αμοντέι αποχώρησε άλλωστε από την OpenAI το 2020 εν μέρει επειδή θεωρούσε ότι η καλιφορνέζικη εταιρεία ήταν πολύ χαλαρή στο ζήτημα της ασφάλειας της ΤΝ. Το 2021 ίδρυσε την Anthropic.
