Contra LessWrong on AGI

Feb 21, 2023 · chatgpt ·

Contra Lesswrong on AGI is an article that explores the idea of artificial general intelligence (AGI) and how it is seen by computer scientist Eliezer Yudkowsky of Lesswrong. The article states that, according to Yudkowsky, there are two paths to developing an AGI: one that uses narrow AI technologies to slowly build up a more complex machine, and another that involves creating a “seed AI” which would develop itself once activated. The article then goes on to detail the arguments against each path and why Yudkowsky has been wrong in both cases.

The first argument against Yudkowsky's suggestion is that building up a complex AI from simpler components may not be enough if the individual parts are too limited or if the interactions between them are insufficient. This means that even though we could create an AI with many simple components, it may not have the capability to understand and interact with the environment in a meaningful way. Furthermore, Yudkowsky does not take into account the difficulty that humans have in understanding their own cognitive processes, meaning that even if such an AI was feasible, it would still need to be developed and tested through trial and error.

The second argument against Yudkowsky's suggestion is that creating a seed AI is not only risky but also unnecessary. If the aim is to create an AI capable of performing tasks as well as or better than humans, then it would be better to focus on improving existing AI techniques. Additionally, the article argues that there is no guarantee that a seed AI would develop itself in the way that we expect and that it could even result in dangerous outcomes such as unleashing an AI “monster”.

Finally, the article concludes that although Yudkowsky may have some valid points about the nature of AGI, his arguments ultimately fail to address the real challenges that face us when trying to develop a truly intelligent machine. To this end, the article suggests that instead of relying on theoretical arguments, we should focus our efforts on improving existing AI technologies and exploring new avenues for research.