Compound AI Systems

What are Compound AI Systems?

AI systems have long been dominated by single, monolithic models. These models often are massive in size, trained on large datasets, and designed to perform specific tasks. While these single-modeled AI systems have showcased remarkable capabilities, they also revealed significant limitations. They lacked versatility and adaptability and often failed to adjust to dynamic environments. 

The concept of compound AI systems has emerged to tackle the limitations. A compound AI system typically consists of multiple components working together. These components may serve different purposes within the system, such as handling different stages of a task, providing complementary functionalities, or handling different kinds of data, while interacting with each other and collaboratively completing the task. A recent paper by Berkeley Artificial Intelligence Research shows that instead of focusing on ever-larger models, developers are increasingly building compound AI systems. At its core, compound AI systems represent a profound shift in AI system development, moving away from large standalone models towards more dynamic and collaborative compound systems. 

What kind of Components are in a Compound AI System?

A compound AI system comprises several key components, each playing a significant role in the system. The components can vary based on what kind of tasks the system tackles. Let’s take an example of an AI system that generates creative images based on textual user inputs (e.g., MidJourney). It may combine the following components to achieve high-quality artistic outputs:

• Large Language Model (LLM): an LLM component analyzes the user's text description, understanding the desired content, style, and artistic elements.
• Image generation component: this component leverages the LLM's interpretation to generate a series of candidate images based on a vast dataset of existing artwork and artistic styles.
• Diffusion model: this is likely used in a text-to-image system to refine the initial image outputs, progressively adding details and improving the overall quality and coherence of the final image.
• User feedback integration: it allows users to provide feedback on generated images through text prompts or by selecting preferred variations. This feedback loop helps the system refine subsequent image iterations.
• Ranking and selection component: it uses ranking algorithms to select the best image from the generated options, considering factors like user preferences and adherence to the original description.

A question-and-answering system is another example that can benefit from a compound AI system to achieve a high level of accuracy and trust:

• LLM: a powerful pre-trained LLM is used to generate potential answers to user queries.
• Information Retrieval (IR) component: it retrieves relevant information from a knowledge base or corpus based on the user query and the task domain.
• External knowledge base: it serves as a repository of structured information that can supplement the system's knowledge.
• Ensemble component: it allows the system not to rely on a single answer from the LLM, instead, it could generate multiple potential solutions and then use an ensemble method to select the most likely and accurate answer based on the agreement among the generated options.
• Fact verification: this could involve integrating external tools or databases to verify the factual accuracy of key points within the LLM's generated answer.

What are the Advantages of Compound AI Systems?

• Improved performance for complex tasks:  for many complex tasks, a well-designed compound AI system can gather the capacities of its components that excel at diverse purposes and this collaborative outcome can usually outperform a single, large model. 
• Increased flexibility and adaptability: compound AI systems are more adaptable to different situations and tasks. For example, individual components can be replaced or adjusted to optimize performance for a specific scenario. 
• Dynamic integration of new data from different sources:  compound AI systems can incorporate new data sources or tools more easily. For instance, a real-time data stream or a specific database can be connected when needed.
• Enhanced control and trust:  compound AI systems offer greater control over the model's outputs. Filters, output validation, or human-in-the-loop approaches can be included to ensure the system operates according to desired requirements. 

Summary

Compound AI systems represent a significant step forward in AI development. By leveraging the power of collaboration between diverse components, they offer the potential to tackle complex problems that were previously beyond the reach of traditional AI.  As compound AI system technology continues to evolve, we can expect to see even more innovative applications emerge, shaping the future of various industries and scientific fields.