In a recent podcast, Demis Hassabis, co-founder of DeepMind, shared his views on the current state of artificial intelligence (AI) development. He believes that the capabilities of AI are currently overhyped. Despite the high public interest in AI, this has also led to a lot of noise in the field. It is a surreal experience to see the public gradually understanding the future importance of AI, he feels.
Image source: The image was generated by AI, provided by the image licensing service Midjourney
Hassabis noted that despite the over-exaggeration of existing AI models, he is still amazed by their ability to infer real-world information solely through language learning. "They learn in an abstract domain, and thus can infer some real-world content, which is truly astonishing," Hassabis said.
When discussing short-term hype versus long-term potential, he pointed out that many people and startups claim that AI can do things that are currently beyond the reach of technology. He calls these ideas "crazy notions" and criticizes some venture capitalists for chasing these non-existent technologies.
However, Hassabis believes that the long-term potential of AI, especially in the realm of Artificial General Intelligence (AGI), is still underestimated. He is confident that AGI could be achieved within the next decade. DeepMind's original goal was a 20-year project, which is set to expire around 2030. "I wouldn't be surprised if AGI arrives within the next decade," he emphasized, believing that AGI could bring immense potential to humanity, such as curing diseases, developing clean energy, and addressing climate change.
At the same time, Hassabis also mentioned that as the capabilities of AI systems grow, it is crucial to address the risks of human misuse and失控. He called for international cooperation in AI regulation and predicted that this technology would be difficult to confine within national borders. In the final stages of AGI development, Hassabis envisioned a scientific "Avengers" team of top global AI researchers, working together to face future challenges.
Key Points:
🌟 The capabilities of current AI are exaggerated, with many claiming functions that cannot be realized.
🌍 The potential of AGI is still underestimated; Hassabis believes AGI could be achieved within the next decade.
⚖️ Strengthening international cooperation is necessary to prevent the risks of AI misuse and失控.
