The False Dawn of AGI

🌈 Abstract

The article discusses the concept of Artificial General Intelligence (AGI) and the challenges in defining and achieving it. It highlights the gaps between the promises and expectations of AGI and what can realistically be expected.

🙋 Q&A

[01] Assumptions Underlying the Definition of AGI

1. What are the three critical assumptions underlying the definition of AGI proposed in the "Levels of AGI" paper? The three critical assumptions are:

• The ability to perform physical tasks is not a necessary prerequisite for achieving AGI
• Tasks can be neatly divided into two categories: cognitive and physical
• AGI can be defined in terms of the successful accomplishment of a set of cognitive and metacognitive tasks

2. Why does the author consider these assumptions to be flawed?

• The assumption that physical embodiment can be regarded as a simple afterthought is doubtful, as physical embodiment has profound implications for an intelligent agent.
• The assumption that tasks can be neatly divided into cognitive and physical categories is contradicted by everyday experience, where most tasks require a high level of both physical and cognitive competencies.
• The assumption that AGI can be defined in terms of task accomplishment is flawed, as human-level intelligence cannot be demonstrated by simply passing a set of tests, but requires the ability to accomplish human-level tasks that are profoundly human.

[02] Limitations of the "Levels of AGI" Definition

1. What are the author's concerns regarding the "Levels of AGI" definition? The author argues that the "Levels of AGI" definition, which proposes a "broad suite of cognitive and metacognitive tasks" as a test for AGI, is flawed because:

• It presumes that human-level intelligence can be demonstrated by carrying out a set of tasks, which is a strong claim that goes to the heart of what it means to be an intelligent human.
• The author believes there is a distinction between "narrow" and "general" human intelligence, and passing a set of tests may not be sufficient to qualify as "intelligent" in the general sense.
• The author argues that true human-level intelligence can only be proved by accomplishing human-level tasks, such as being a statesman, poet, or parent, which appear virtually impossible for an AGI to achieve.

2. What are the potential consequences of the gap between the expectations and actual abilities of AGI? The author warns that the enormous gap between the expectations of AGI and its actual abilities, if not addressed, will likely result in another crisis of confidence in AI and another long, dark AI winter, similar to the previous AI winters experienced in the past.