After leaving the company, Robinson wrote an article in The Atlantic. Its title is ‘I Quit OpenAI Because Its Culture Is Broken’. In this article, he has expressed concern about the way OpenAI works, rapid preparation of AI models and their security.
Questions raised on the speed of AI development
According to David Robinson, the speed with which OpenAI is working on another model after launching one model is making it difficult to pay necessary attention to the security of AI. He believes that at this time AI companies need to work more carefully.
Robinson also described OpenAI’s “move fast and fix things” approach as risky. He says that along with improving AI rapidly, its security should also be given equal priority.
‘AI companies are not taking enough precautions’
In his article, Robinson said he agreed with other former employees who believe that companies creating AI technology were not taking adequate precautions. He said that OpenAI is moving rapidly from one launch to another, but during this time it is not able to take as much caution as is necessary.
Robinson also questioned the way AI companies work. According to him, in the present times companies need more humility and caution. He said that it is not natural for people who work with a lot of confidence to take such caution.
Concern about AI more capable than humans
Robinson also expressed concern about the future of AI. He said that an environment where security issues related to AI could emerge is not suitable for creating AI systems that could be more intelligent than humans in the future. His concern is that future more capable AI models will always do what humans want them to do, not necessarily. Therefore, before making such models, it is important to pay more attention to their safety.
Two big changes suggested for AI industry
In his article, David Robinson has mentioned two important changes for AI companies.
- Learning from experts in other fields: According to Robinson, AI companies should make more use of the security-related experience and expertise they already have in other areas.
- New scientific understanding necessary: He said that before creating systems more capable than today’s AI models, there is a need to develop new scientific understanding to ensure that more capable AI models make safe decisions even when humans are not monitoring them.
What did OpenAI say about the criticisms?
On David Robinson’s words, OpenAI said that the company is increasing its work in collaboration with outside experts. Besides, the system of continuously monitoring the AI model is also being improved, so that wrong or unsafe behavior of the model can be detected quickly during training.
OpenAI said that it is taking care that its AI models are capable enough to be controlled safely. According to the company, if necessary, it stops the training of the model or stops its development for some time.
There have been changes in OpenAI’s safety team before.
David Robinson’s resignation has come at a time when there is constant discussion about the security of AI. The news also mentions the recent Hugging Face security breach and other security issues related to the misbehavior of AI models.
There have been changes in OpenAI’s safety team before. Apart from the departure of the company’s Safety Chief Johannes Heidecke, it has also come to light that three researchers have been removed for violating rules related to sensitive information.
David Robinson’s resignation and his article published in The Atlantic once again raise the question of how to keep the safety of AI models in mind in the race to make them more capable.