Various unpleasant incidents involving runaway AI modes causing damage to company infrastructure only emphasize the importance of testing the security of the latest developments. With real people doing the work, the ever-increasing workload has forced collective complaints of emotional exhaustion.
Image credit: Unsplash, Massimiliano Sarno
The publication provides comments from several experts who, due to their responsibilities, are forced to spend more and more time and energy testing advanced artificial intelligence models, while at the same time feeling a growing responsibility to society. In many cases, staff at the UK’s Artificial Intelligence Safety Institute, which independently evaluates the latest AI models, have been forced to take sick leave due to burnout and seek advice from medical experts. This is facilitated by the organization’s ever-increasing pace of research and its increasing burden of responsibility to humanity.
Strictly speaking, a series of voluntary dismissals from leading Western AI companies illustrates precisely the relevance of the problem described. AI experts complain not only of fatigue and burnout, but also of difficulties in completing the task of making AI safe. Sometimes, the problem is not a technical hurdle but an administrative one, as startup management is interested in accelerating the development of new AI models whose safety concerns have not previously been a priority.
Like Anthropic’s management, many artificial intelligence developers know nothing about the applications they design and feel psychologically uncomfortable. The Pentagon has made no secret of its intentions to use artificial intelligence systems in the military, and for many employees at AI companies, the idea of their involvement in creating lethal technology began to feel morally oppressive. Not only were employees calling in sick, they were quitting. Ethical rules and corporate standards in the field of artificial intelligence safety are still in their nascent stages, and some experts believe that certain boundaries of permissibility have been exceeded and that humanity is beginning to suffer from its development.
Cybersecurity and biotechnology are two major sources of this concern, as AI is advancing so much in this area that AI models are proving capable of wreaking havoc on entire company information infrastructures, and attackers are also capable of developing new types of bioweapons. Some participants in this process even feel that the coming global catastrophe is inevitable and that they are merely weak-willed witnesses to destructive trends. If the psychological climate within the company is poor and subordinates believe their leader is toxic, the situation will be even more serious.
The leadership of the British Artificial Intelligence Safety Institute said that the institute is committed to the well-being of its employees and that those who need help and support will receive it. Google DeepMind said that startup employees have the ability to raise issues of concern to management. Meanwhile, more than a dozen leading developers in artificial intelligence have left DeepMind, OpenAI and Anthropic over the past two years, disagreeing with management over priorities between the safety of the technology they created and the speed of development. This month, Jacob Coxon, who left Anthropic and OpenAI, said in explaining the reasons for his departure that these employers are putting the future of all mankind at risk.
Mrnank Sharma, an AI security researcher who left Anthropic, said “The world is in danger”and under the pressure observed, he will not be able to “Be guided by values in your actions”. Many developers involved in AI security testing believe that companies creating AI models are more interested in expanding their capabilities than worrying about security. That recognition appears to have prompted Anthropic executives to say this month that the pace of artificial intelligence development needs to slow down for the greater good. The race for artificial intelligence is starting to cost humanity more and more, not only in terms of material costs, but also in terms of security risks.
If you find an error, select it with your mouse and press CTRL+ENTER.