Moving Humanity From Consumption to Creation
Short Thoughts on the Implications of Learning on Relative Disempowerment
The following is an excerpt from our website.
A world where AI conducts scientific research, writes code, and solves complex problems usually invokes a fear-based response among “outsiders,” or people who lack control over AI progress. This can manifest as some false sense of human intellectual superiority over AI, but I doubt this will sustainably uphold a contributive society as it becomes clear that AI is far superior to us in essentially all domains. That is, we can not forever be delusional in valuing human cognition over machine intelligence.
I feel this is likely to lead to a transition from humans being active contributors in society to passive consumers who let a small group of powerful humans and AIs direct their lives. Similar to this thought is the idea of “relative disempowerment” (Kulveit et al.), whereby human quality of life stagnates or marginally improves despite the global GDP experiencing rapid growth and technological advancement. This would be due to a misalignment between human needs and technological advancement (i.e. technological advancement does not serve human needs).
This is not an inevitability, though. If humans deliberately decide to be creators and change makers, we may be able to maintain, and even increase, our sense of meaning and purpose. Moreover, increasing the quantity and impact of human-initiated artifacts may help to ensure alignment between humans and AI. This does not mean, however, that we should purely be creators without consuming anything; if we don’t consume, how will we learn? Instead, we must draw a distinction between consuming to be entertained and consuming as a means to achieving an objective.
I have experienced this dichotomy first hand as I transitioned from being a passive consumer of calculus to using AI to extend and apply my understanding of calculus to my research in learning science. I have repeatedly observed this phenomena among the 30 study participants I have interviewed, where many of them use the computer science vocabulary they are taught in university to design complicated program behavior. To make this transition, though, one must be able to imagine the idealistic world in which their invention exists and instantiate its implementation through specifications that define its behavior.
—Hudson
If you’d like to contribute to our work, please see https://papertlab.org/funding/safety.
See a related post (not by me) on LessWrong here.
 | A guest post by
|