Dear Mr. Huang:
Thank you for sharing your vision of the future for Nvidia, AI, and by extension all of our future lives. It’s difficult to imagine the computing power that will be realized by the upcoming Ultra Blackwell and Rubin architectures. As Nvidia prepares to move into more inference, deep machine learning, and reasoning with your wide variety of customers and partners, all while bringing along a software suite to match, I am concerned and compelled to ask, “How can a ‘reasoning’ AI model make any sense if the highest common goals of humanity are not ‘taught’?”
Admittedly, determining the “highest goals of humanity” then becomes the question. Would different cultures, governments, and religions all have input? Some would argue that, over millennia, humans have valued a love of family, compassion for others, and an ethic of truthfulness and honesty. More recently, an appreciation of the beauty, sustenance provided, and fragility of our planet and environment might be considered an overarching theme in problem solving, as the need for various resources is a component of many complex decisions where AI might be employed. Additionally, the idea of “first do no harm” together with a moral concept that would protect us against outcomes, i,e. reasoning, that would prioritize profit, efficiency, competitive advantage, or completing a requested task despite the negative impacts to human societies or individuals that happen to be in the way would be an important aspect in Artificial General Intelligence models.
Obviously, who or what group would decide the ultimate criteria and how the “training” might be accomplished or enforced are not defined. Yet, isn’t this the challenge we must now confront? Tragically, we have been unable to control the dark side of Social Media much to our detriment. Can humans be trusted with mature Artificial Intelligence?
