On speaking to AI

🌈 Abstract

The article discusses the contrasting approaches of Apple's Siri AI and OpenAI's ChatGPT in the realm of conversational AI assistants. It highlights the differences between the "Copilot" philosophy of Siri, which prioritizes privacy, safety, and security, and the more powerful but potentially riskier "Agent" approach of ChatGPT's voice capabilities. The article explores the trade-offs between these two philosophies and the implications for the future of AI interactions.

🙋 Q&A

[01] Siri AI vs. ChatGPT

1. What are the key differences between Siri AI and ChatGPT in terms of their approaches to AI?

• Siri AI is built around a small, specialized AI model focused on specific tasks, with an emphasis on privacy, safety, and security. In contrast, ChatGPT utilizes the more powerful and generalist GPT-4 model, which allows for more natural and open-ended conversations, but also carries greater risks.
• Siri AI is designed as a "Copilot" that assists users with specific tasks, while ChatGPT represents an "Agent" approach, where the AI system can engage in more autonomous and potentially unpredictable interactions.
• The author suggests that Siri's approach trades off power for security, while ChatGPT embraces the potential benefits and risks of a more powerful and less constrained AI system.

2. How do the technical differences between Siri AI and ChatGPT contribute to their contrasting approaches?

• Siri AI uses a small, 3 billion parameter AI model that can run on the user's device, allowing for privacy and security. In contrast, ChatGPT utilizes the much larger and more powerful GPT-4 model, which requires specialized hardware to run.
• The smaller size of Siri's model means it is less capable than the generalist GPT-4, leading to a more limited but more predictable and controllable AI assistant.
• The author notes that a future update to Siri may allow it to leverage larger cloud-based AI models when needed, blending the Copilot and Agent approaches.

[02] Implications of the Contrasting Approaches

1. What are the potential benefits and risks associated with the Copilot (Siri) and Agent (ChatGPT) approaches to AI?

• The Copilot approach of Siri aims to make AI less "weird" and more predictable, sacrificing power for security and safety. This can lead to more reliable but less transformative AI assistants.
• The Agent approach of ChatGPT embraces the full power and potential of large language models, but also carries greater risks of unpredictable or harmful behavior.
• The author suggests that companies will need to carefully balance these trade-offs, deciding when to prioritize low-risk, less capable systems versus high-risk, high-benefit AI agents.

2. How might the different approaches to voice interactions impact the future of AI-human interactions?

• Voice interactions, as demonstrated by ChatGPT's voice capabilities, can make AI assistants feel more natural and human-like, with implications for how people emotionally engage with them.
• The author suggests that voice and multimodal (visual, auditory) interactions will become more common and appealing to a wider audience, changing how people interact with AI systems.
• However, the risks and unpredictability of agent-based voice AI, like ChatGPT, will need to be carefully managed, in contrast with the more constrained Copilot approach of Siri.