Elon Musk Thinks AI Will be “Smarter Than Any Single Human” by Next Year
Elon Musk has long been a vocal AI pundit. Photo by Mat Hayward/Getty Images for Prime Video).
Key Takeaways
- Elon Musk said that AI will probably surpass human intelligence by next year.
- By 2029, he said AI could be “smarter than all humans combined.”
- As model performance increases, the way artificial intelligence is measured may need to change.
The machine learning models’ ability to complete intellectual tasks has long been benchmarked against humans. Now, given significant advances in model performance during the last 2 years, many leading figures in the space believe AI is on the cusp of surpassing human intelligence.
In recent comments, Elon Musk speculated that that moment could occur as soon as next year and that artificial superintelligence isn’t far off either.
When Will AI Overtake Humans?
In many respects, AI has already surpassed human cognitive ability. If intelligence is perceived as nothing more than a memory game, computers did that decades ago and AI assistants can already search and filter vast repositories of information no human could ever hope to remember.
But only in the past few years have large language models (LLMs) been able to complete complex intellectual tasks that require more advanced reasoning.
Musk’s statement was made in response to a video he shared featuring Ray Kurzweil, who in 1999 famously predicted that AI would achieve human-like intelligence by the year 2029. The Tesla founder, however, said that it would probably be “smarter than all humans combined” by the year 2029.
How Artificial Intelligence is Measured
Ultimately, there are limits to the analogy between artificial and human intelligence as they are currently conceived.
Contemporary ML benchmark tests are the equivalent of exams where there are only wrong or right answers and students have memorized the textbook in advance.
In these kinds of tests, depending on the topic, the most advanced AI models already perform between middle school and college level and it isn’t unreasonable to assume they will outperform most humans within a year or 2.
Looking forward, AI researchers are developing more sophisticated benchmarks to assess models’ ability to solve problems outside the controlled environment of existing tests.
For example, researchers at the University of the Witwatersrand in South Africa have proposed using Minecraft to measure AI performance.
In the end, Musk’s statement and Kurzweil’s predictions reduce intelligence to something that can be determined quantitatively. But as AI advances the way people use it may end up resembling a Minecraft game more than a high school exam. There is no way to win Minecraft. Instead, each user must set their own goals, decide for themselves why they play and measure their success or failure accordingly.
AI Minecraft Challenge: Measuring Artificial Intelligence Through Gameplay
ChatGPT Update Made it Worse? OpenAI’s Performance Dips Amid User Complaints
