Concerns Raised by Geoffrey Hinton Regarding Profit-Driven AI Development

2023-06-13

Two pioneers of artificial intelligence, Geoffrey Hinton and Andrew Ng, recently came together for an interesting discussion on the threats and risks of AI. Hinton, who recently left Google to discuss the threat of artificial intelligence, supports Elon Musk, Steve Wozniak, Yoshua Bengio, Gary Marcus, and other AI experts who signed an open letter calling for a pause in the development of AI after GPT-4. However, Ng does not agree with the idea of a "pause". While Hinton now expresses concerns about the dangers of AI, he previously overlooked the ethical issues raised by the Google team. He compared the potential risks of AI to the development of atomic bombs during World War II, emphasizing the dangers of profit-driven AI development that could lead to AI-generated content surpassing human-produced content and threatening our existence. Hinton calls for unity among AI researchers when faced with differing opinions. He suggests that researchers need to reach a consensus, similar to climate scientists reaching a consensus on climate change since the late 19th century. This is important because policymakers and decision-makers seek technical advice from researchers. Another important point of discussion is whether language models like ChatGPT or Bard truly understand what they are saying or if they are just statistical structures. While some believe they understand, others disagree. Resolving this issue is crucial for reaching a consensus on AI-related issues. The challenge in assessing understanding lies in determining appropriate tests within the system. Large-scale language models and AI models seem to be constructing a world model, indicating a certain level of understanding. However, this is a subjective viewpoint. Further discussion and the development of a shared understanding among researchers can promote more consistent reasoning and improve the AI community's understanding of AI risks. This discussion involves statistics, as we all recognize its crucial role. However, some people who view it as merely statistics tend to think from the perspective of programming or the co-occurrence frequency of words. "We believe that the process of creating features or embeddings and the interactions between these features go beyond mere statistics; it involves understanding," he adds. By predicting the probability of the next word based on the complex interactions between features, we can predict the next symbol. I personally believe that this process represents understanding, similar to what our brains do. However, this is a topic that needs to be discussed in the research community to convince others that these systems are not just statistical structures and to reach a consensus on addressing AI-related risks. In conclusion, gaining a better understanding of the ability of AI systems to understand may bring the research community closer to reaching similar conclusions. Source: https://analyticsindiamag.com/geoffrey-hinton-raises-concerns-over-profit-driven-ai-development-urges-caution/

RECENT AI NEWS

RECENT AI TOOLS