Part 8/16:
Another common fear is that AI will undergo a "treacherous turn," becoming malicious or malevolent once surpassing human intelligence. The paper AI 2027 hypothesizes that competitive pressures may lead to deployment of increasingly dangerous AI systems, which could unpredictably turn malevolent.
The speaker counters this with the point that AI, even at high levels of intelligence, tends to become more helpful and benevolent the smarter it gets—particularly because these systems are designed and trained with safety in mind. The fear of a sudden "evil" turn ignores that AI, as it currently exists, does not inherently possess the ego, self-awareness, or malice necessary for such a conversion. Instead, projecting human-like malevolence onto AI is an anthropomorphic fallacy.