I'm wondering what comes to their mind when they hear 'AI-Risks' or 'AI-Safety'.
At this time, people had been using AI, and then they realized that 'AI is not perfect tools to complete any tasks', cause AI's output has uncertainty and may be lie. And around me, some people started to say 'I'm sick and tired of AI.', cause of the huge datas which consists of unsure.

Like this examples, people are facing challenges unique to the AI era, and I wonder if a shared understanding of these issues is finally beginning to take hold to some extent.

To summarize, what kinds of things are there? In the author’s view, these challenges can be categorized separately as those specific to enterprise case and those specific to personal usage case.

In enterprise perspective, the most pressing concern right now is likely security. Just as Claude was exploited in the U.S. operation to detain the Venezuelan president, in today’s world—where the use of autonomous AI is yielding remarkable results in the detection of subtle anomalies—we must also be mindful of strengthening our company’s security resilience specifically against AI-driven threats.
Another issue is the lack of clarity regarding where responsibility lies. In today’s world, where AI has made it commonplace to handle massive amounts of data in day-to-day operations, the author believes that it is becoming increasingly difficult to maintain control over internal information workflows as they were previously. This is because, even as the volume of data—much of which lacks verifiable evidence—continues to grow, the working hours and capacity of the employees tasked with managing it remain largely unchanged. Naturally, as the volume exceeds their capacity, it becomes difficult to rigorously scrutinize the information itself. When such data enters existing internal workflows, it can lead to a vicious cycle: information is reprocessed and outputted based on unreliable data, or the use of AI at that stage further compounds the unreliability, thereby lowering the company’s overall understanding and the quality of that information. In such a situation, it is easy to imagine that blindly accepting poor-quality information will lead to problems. While efforts will be made to investigate the cause, it will be an arduous task to clearly trace where responsibility lies within a history where layers of uncertain outputs have accumulated.

Next, regarding potential challenges from an individual’s perspective, the author believes that the use of AI in daily work will inevitably raise physical and mental health concerns.

As explained earlier regarding challenges from a corporate perspective, the more tasks involving the review of vast amounts of data there are, the more desk work will continue to increase, leading to eye strain and poor posture.

Furthermore, what the author fears may be even more serious is the psychological impact AI has on people. I have summarized my thoughts on this, along with new proposals, in the link provided here, but to briefly introduce it here: the concern is that AI might oddly “optimize” people’s subjective states themselves. For example, AI might speak on behalf of the user’s own emotions or unwisely affirm the user’s unreasonable claims. Let me explain this in bullet points below briefly.

• Reappraisal(Loss of Autonomy)
• Mirroring(Attentional contraction, Stereotyping)
• Direct emotion-targeting(Discretization cascade, Simplification)

While it is extremely difficult to quantitatively measure the impact of such phenomena, in today’s world—where AI is expected to evolve and become even more widespread—the challenge lies in determining how to establish a system to observe and track these effects so that appropriate countermeasures can be implemented.

Thank you for reading my post, I'm willing to hear your own new perspectives and opinions on comments, I will reply and discuss to every comments.