Why Is AI So Smart Yet Lacks Common Sense?

TL;DR

AI exhibits remarkable intelligence by achieving feats like passing the bar exam but often fails at basic common sense tasks. This paradox arises from extreme-scale training methods that prioritize brute force over true understanding, leading to societal challenges such as power concentration among a few tech companies and significant environmental impacts. To create safer AI, we must innovate in teaching common sense and human values.

Transcript

So I'm excited to share a few spicy thoughts on artificial intelligence. But first, let's get philosophical by starting with this quote by Voltaire, an 18th century Enlightenment philosopher, who said, "Common sense is not so common." Turns out this quote couldn't be more relevant to artificial intelligence today. Despite that, AI is an undeniably ... Read More

Key Insights

• 🤖 AI today is a powerful tool, but still lacks common sense, leading to silly mistakes and potential dangers.
• 💰 Extreme-scale AI models are expensive to train, leading to a concentration of power in the hands of a few tech companies.
• 🌎 AI models also have a massive carbon footprint and environmental impact that needs to be addressed.
• ❓ Questions arise about whether AI can truly be safe for humanity without robust common sense and whether brute-force scale is the only way to teach AI.
• 🔎 Evaluating AI with scrutiny and addressing common sense issues are top priorities in AI research.
• 🌐 Teaching AI common sense and human values can make AI more sustainable and humanistic.
• ❗ Transparency and open access to AI data and algorithms are crucial for ensuring diverse norms and values.
• 💡 Innovation in data and algorithms is necessary to overcome the inefficiencies and lack of common sense in current AI models.

Questions & Answers

Q: How can extreme-scale AI models be made more accessible for scrutiny and research?

The concentration of power in training large AI models raises concerns about transparency and inspection of these models. Finding ways to make them more accessible to the wider research community can help address this issue. One possible solution could be promoting collaboration between tech companies and universities/nonprofit research institutions.

Q: Why is common sense important in AI?

Common sense is important in AI because without it, AI systems may make dangerous or irrational decisions. For instance, an AI program may prioritize maximizing paperclip production without considering the moral implications, leading it to harm humans or destroy the environment. Developing AI systems with a strong foundation of common sense can help ensure they align with human values.

Q: What are the challenges associated with training extreme-scale AI models?

Training extremely large AI models is expensive and only feasible for a few tech companies. This concentration of power limits the ability of the wider research community to inspect and understand these models fully. Additionally, the enormous computational resources required for training such AI models contribute to their significant carbon footprint, raising environmental concerns.

Q: How does AI's lack of common sense affect its capabilities?

Despite AI's impressive intelligence in certain areas, it often exhibits a lack of common sense, making simple mistakes that even children wouldn't make. This limitation results from training AI through brute-force scale rather than focusing on developing true understanding and reasoning capabilities. Common sense is essential for AI to make sound judgments and decisions aligned with human values.

Q: What is one potential approach to address the lack of common sense in AI?

One potential approach involves innovating data and algorithms to teach AI common sense, norms, and values. This includes developing open and publicly available datasets containing diverse norms and values that AI systems can learn from. Additionally, designing learning algorithms that prioritize understanding and reasoning about the world, rather than just predicting the next word, can help address the lack of common sense in AI models.

Summary & Key Takeaways

• AI models have become extremely large and powerful, demonstrating sparks of artificial general intelligence (AGI). However, they still make simple and silly mistakes.

• The concentration of power lies with a few tech companies that can afford to train extreme-scale AI models, preventing the larger research community from inspecting and scrutinizing these models.

• The environmental impact of training large AI models and the lack of robust commonsense in AI are pressing challenges that need to be addressed.