AI outperforms humans at Amazon: surveillance becomes impossible

In an interview with the media outlet The Register, Eric Brandwine, distinguished engineer and vice president of Amazon Security, shared a sobering observation: humans are no longer able to keep up with the pace of AI at Amazon, particularly during the step-by-step verification phases of the work performed.To justify excluding humans from this process, Eric Brandwine cites the theory of the normalization of deviance.

At Amazon, humans would no longer be able to keep up with the pace of AI

Artificial intelligence agentic is capable of performing dozens of actions per minute, a pace that humans struggle to keep up with. That is Eric Brandwine’s assessment; he is an emeritus engineer and Vice President of Amazon Security. For him, humans would no longer be able to hold their own against AI; their exclusion from certain work processes would therefore be inevitable, particularly during the process of validating the work performed. He explains: “We like to think that we are all very good at our jobs and that we hold ourselves in high regard. But in reality, humans are not very consistent”. Brandwine also believes that humans tend to be “a little too attached to other humans”.

At Amazon, each agent deployed internally has its own identifier. Each of its actions is then logged as being carried out on behalf of a specific employee. However, the human no longer oversees the entire work process. Although safeguards are indeed in place, they are primarily dynamic policies that govern the work.In other words, humans tell the AI in real time what it can and cannot do, but they no longer perform step-by-step validation work. And ultimately, the artificial intelligence trains itself.

The theory of the normalization of deviance

To support his point, Eric Brandwine mentions the theory of the normalization of deviance, formalized by Diane Vaughan in 1996. At first, a group has the time and resources to perform tasks. Everything goes well. Then, the group has less time and fewer resources. The quality of the work declines, and the various people begin to somewhat circumvent the strict rules put in place in order to complete the tasks on time. Finally, the time and resources are drastically reduced. The quality of the work then becomes poor, and people normalize deviance in order to complete tasks, as long as no serious incident occurs. Eric Brandwine explains: “This is a trap to which everyone is subject, and one of the most harrowing stories I have read in this field involved emergency services. You are surrounded by machines that beep constantly. On the first day, you flinch at every alarm, even when the patient is fine. It is a false alarm. You return to your station, sit down, and after a while, after enough false alarms, enough repeated beeps with no real consequence, you let your guard down and stop responding. And eventually, a tragedy occurs.”

For Eric Brandwine, the normalization of deviance theory can be fully applied to artificial intelligence embedded in the workplace. Human beings, no longer able to keep up with the frenzied pace of AI, are beginning to adopt risky behaviors. Amazon Security’s vice president concludes by saying: “This is why, at Amazon, we are not strong supporters of human intervention in the loop.” Step-by-step human oversight therefore seems to be a thing of the past at Amazon. Meanwhile, Tesla had to recall more than 14,000 Model Y vehicles due to a missing label. The reason: a computer vision tool malfunctioned. The carmaker has since introduced a manual check. In other words, human beings struggle to keep up with the pace of certain technologies, but they can also prevent serious problems.

👉🏻 Follow tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.

Source: The Register