Silicon Valley has been buzzing lately over whether one should be polite to artificial intelligence. The spark was AI company Anthropic updating its terms of service this week to add a provision barring users from "repeatedly subjecting our models to cruel behavior." In plain terms, the rule prohibits verbally abusing the models. The provision has ignited online debate about whether machines can have feelings, and Elon Musk's stance clearly backs the "they can feel" camp.
The first to weigh in was Aaron Levie, chief executive of cloud storage firm Box. On a social platform, he said that while he does not believe large language models possess consciousness, banning abusive behavior might still be a good thing. His reasoning: these models learn from interactions with humans, and no one wants future models trained on abusive content. "Models only understand the data they were trained on, or what they encounter in interactions," he wrote, "so if we want safe and aligned models, we probably want the AI's training data to contain a lot of good interactions."
Ban on Cruelty Toward AI Models Sparks Wide Discussion
He went on to frame it as a simple "Pascal's Wager" style choice. Pascal's Wager is a theological concept arguing that even if you don't believe in God, claiming belief costs nothing but avoids the potential cost of being wrong. Applied to AI, it means it's better to be nice to AI.
Elon Musk then replied to that post, though his response suggests he may have missed Levie's point. "I think this is the right approach," Musk wrote. "It is wrong to be cruel to an entity that believes it is experiencing suffering." Musk has made similar remarks of late. His characterization of humans as a "biological bootloader" is drawing increasing attention. On Oct. 4, he posted: "Spreading superintelligence to the stars is a good success condition for the biological bootloader." He has also said repeatedly that he wants to extend the "light of consciousness" beyond Earth.
That phrasing is seen as implying that, in his view, saving humanity itself may not be the most important goal.