Virtually Free Intelligence: What Lies Ahead for Data Systems?
A Revolution in Artificial Intelligence
In a world where the capabilities of artificial intelligences, such as GPT-4, are constantly evolving, the cost of AI continues to drop at an unprecedented rate. In 2023, the cost of utilizing these technologies dramatically decreased, reaching under a dollar per million tokens. This change has profound implications for how we design and use data systems.
The Challenges and Opportunities of New Data Systems
With this increasing accessibility, three major challenges emerge for data systems:
1. Data Systems For Agents
In the future, AI-powered agents will become the primary users of data systems. Unlike humans, these agents perform what is called "agentic speculation." This involves a complex workflow where each user request can translate into thousands of SQL queries. For instance, a question like "Why did coffee sales in Berkeley drop this year?" could require an in-depth analysis of numerous data sets.
To meet these needs, it is essential to rethink how data systems are designed. Query optimization, for example, could be improved by avoiding task duplication. By leveraging the properties of data systems, we could enable agents to make faster progress by reusing results from similar queries and providing approximate answers when sufficient.
2. Data Systems Of Agents
As agents take on an increasing share of knowledge work, it is imperative to create systems capable of efficiently managing a large number of these entities. These systems must be designed to maintain state over long tasks, coordinate agents, and manage failures. A robust architecture will be essential to ensure the reliability and efficiency of these systems in managing agent swarms.
3. Data Systems By Agents
With the advent of agents capable of synthesizing entire data systems, we are entering an era where custom solutions can be created for every new need. However, a major challenge arises: how to ensure that these systems match expected behaviors? Verifying systems created by agents will require innovative approaches to ensure trust in their operation.
Towards a New Design of Data Systems
One of the avenues to explore is the redefinition of the query interface. Instead of submitting one query at a time, agents could submit batches of queries with individual approximation requirements. This would better utilize their reasoning abilities and optimize the performance of data systems.
Another avenue would be to evolve how data systems interact with agents. Rather than simply executing queries, these systems could adopt a proactive approach. By providing results for related queries or latency estimates before executing costly queries, systems could guide agents toward more efficient solutions.
Conclusion: The Future of Data Systems
The rise of low-cost artificial intelligence offers unprecedented opportunities to reinvent our data systems. By adopting an agent-centered approach, we have the possibility to transform how we interact with data and significantly enhance the efficiency of our work. These changes are not merely technical; they also involve a reflection on how we can leverage the capabilities of agents to optimize our decision-making processes and analyses.
To discuss these developments and their impact on your business, Contact me.