The Anthropic Incident |
A furious debate is going on in the US on the dangers of Artificial Intelligence (AI) to humanity. It began earlier this month when Dario Amodei, the CEO of the leading US company Anthropic disclosed that a very disturbing incident had occurred in the company some months ago. AI agents within a model which was undergoing an experiment process within an isolated environment broke its instructions. Surprisingly collaborating with other AI agents within the model it bypassed its isolation and was able to hack into some of the company’s protocols. It thereafter hacked into another company Hugging Face which also is based on open source systems. While the AI agents’ collaborating among themselves was in itself most worrying, what was equally so was that this activity went on for two months without being discovered.
Dario Amodei advocated that the US companies in the AI field should go slow because of the dangers of AI based on Large Language Models (LLMs) could go beyond or against the instructions given to it and cause great problems. HIs caution was supported by the heads of other companies. They included Sam Altman of OpenAI and also Elon Musk who has large investments in his AI companies. All these leaders and many more have expressed great worries about investments leading to more sophisticated AI models being developed without sufficient precautions being taken........