Sam Altman stated that the world is correct in fearing the uncontrolled advancement of Artificial Intelligence, admitting that accidents can be inherent to the development of new technologies.
During the Dreamforce conference, held by Salesforce, the OpenAI CEO advocated for the implementation of a culture of transparency. This culture must allow for the recording, investigation, and learning from any negative episodes.
To illustrate his point, Altman referenced accident investigation methods in aviation, aiming to identify failures and mitigate the emergence of new problems. He mentioned the U.S. Federal Aviation Administration (FAA) and the National Transportation Safety Board (NTSB) as applicable references for the AI industry.
According to Altman, although he considers that a certain level of accidents is inevitable with technological innovations, the main focus must be on establishing an excellent culture of recording and learning from these events. Recognizing this possibility does not imply accepting catastrophes, but rather quickly discovering what failed to prevent an incident from worsening.
Among the recommended practices, he highlighted the need to maintain the safety, monitoring, and alignment of AI systems, ensuring they remain under control and compliant with established objectives, even as model capabilities grow.
Amid growing discussions about increasingly autonomous AI systems, safety researchers have warned of risks such as potential loss of human control and the use of these technologies to enhance cyber and biological threats. Altman validated these extreme fears, stating: 'The world is right to be afraid of this,' citing the risk of accidents related to loss of control and centralization of power in a few development teams.
However, Altman argues that abandoning artificial intelligence is not the appropriate solution. If safety measures cannot be implemented, he suggested that companies should be prepared to scale back or suspend development. His perspective contributes to the debate on the pace of AI progress and the dangers associated with increasingly autonomous models.
For Altman, the way forward requires disclosure of incidents, a thorough analysis of what occurred, and continuous learning from each case, thereby preventing more serious consequences resulting from failures.