According to Google DeepMind Chief Scientist Jeff Dean, current AI models have likely already surpassed the capabilities of the average human in most non-physical tasks. The central argument is that while most people are not naturally skilled at a random new task, modern AI models are reasonably proficient across a wide range of activities.
## Key Points and Arguments
* **AI vs. the ‘Average’ Person:** Dean’s claim is specific: AI is better than the *average person*, not a world-class expert. He cautions that models still fail at many things and are not at a human expert level in specialized fields.
* **Accelerating Progress:** He believes we are close to a point where AI, through automated search and computation, will make scientific and engineering breakthroughs faster than humans in an expanding number of domains.
* **Avoiding the AGI Debate:** Dean steers clear of conversations about Artificial General Intelligence (AGI), stating the term lacks a clear and consistent definition. This contrasts with his boss, DeepMind CEO Demis Hassabis, who is more optimistic about achieving AGI within the next decade.
## Conclusion
The main takeaway is that a top expert in the field believes AI has already crossed a significant threshold, outperforming the general populace in a wide array of cognitive tasks. The future focus will be on leveraging this capability to accelerate human progress in science and engineering, rather than debating the ambiguous concept of AGI.
Mentoring question
The article distinguishes between AI surpassing an ‘average’ person versus an ‘expert’. In your own field, what specific skills or deep knowledge would you need to cultivate to stay in that ‘expert’ category where human value remains distinct?
Leave a Reply