The Rise of AI Agents
Extending Your Own Capabilities
“There is no greater love than a maker’s for the things she makes. Especially when they grow to be something she never imagined.”
― Hannu Rajaniemi, The Causal Angel
Note: Substack and Twitter are currently in a dumb fight and the ability to add tweets to a Substack post or provide a Substack link on Twitter are disabled. There are a lot of twitter threads about the projects mentioned here.
AI is moving fast. Part of the reason for that is how AI systems can be tested quickly and improve themselves. The other part is from AI improving or unblocking areas that haven't seen as much progress in the past. Let's talk about the latter with the current rise of AI agents. AI agents are virtual entities that can perceive and interact with their environment. Two projects got released very recently and are already being improved at a rapid pace, AutoGPT and BabyAGI. These projects create an agent framework for obtaining user provided objectives by leveraging large language models (LLMs) to determine and execute tasks. In layman’s terms, these agents are intelligent robots provided with a goal which are using the latest technology not just return a value but to interact intelligently with their environment and take actions based on the responses they receive. The advancement here is a logic flow for creating a task queue for an objective, checking between a task queue and what’s happened in memory, and then determining the best move forward for executing tasks, performed almost entirely with just an LLM in creative ways. The architecture for BabyAGI is below in Figure 1.
Figure 1. The architecture for BabyAGI
What’s great here is that now we have a working framework to easily extend and test different agent designs. This enables very fast iteration. These projects have only been around for a few weeks and already people are building upon them. Frankly, it feels like people are being one-upped every hour. If you haven’t seen them yet, I suggest playing with them and thinking about how you can use the projects to upgrade yourself and your capabilities. Here are a bunch of people who have taken the projects to new heights:
https://twitter.com/adamcohenhillel/status/1645560223334293504
https://twitter.com/LatentLich/status/1645564783784067073
https://twitter.com/SullyOmarr/status/1645205292756418562
https://twitter.com/Radio_poodle/status/1643212279163539458
The potential applications of AI agents are vast and transformative. For instance, Stanford and Google just released this paper and demo that shows the use of generative AI to create a video game where the NPCs go about their daily lives with their own objectives and actions (think Free Guy or the Sims). This seemingly small-scale application highlights the potential of AI agents to tackle real-world challenges. History has shown that games and virtual environments often serve as proving grounds for AI models before they transition to real-world applications. One example is DeepMind’s progression from mastering games to revolutionizing biology.
So, what’s missing? Well, quality actuators. Actuators to have these agents actually execute tasks. Some of them are straight forward and already exist, such as providing a search capability or spinning up and running code. However, as always, the devil is in the details. You may have an agent realize that it needs to buy 100 widgets but does it have an upgraded directory, with the ability to pay, a model to correctly assess the right widget, and the ability to properly negotiate? Will you need a separate agent to accomplish each of these? Or is the API to an application enough? Again questions of distribution control will become paramount.
What about ecosystems implications? Depending on who you ask, around half or more of all internet traffic comes from bots. In the next five years, it’s possible that a quarter or more of all traffic will be agent driven. Agents will be gathering information, exchanging information with one another, and interacting with the world. People will begin to use agents to augment their abilities. Imagine every individual having a team of executive assistants that can recruit their own teams to accomplish tasks. That’s the potential of these smart agents.
As agent technology advances, entire markets will emerge for who creates the best agents. Just like assembling teams now, whether internally or through contractors, companies will find the best agents for their budget to accomplish the tasks they need. The best agents will have verified processes that mitigate risk and have higher execution achievement. While the lower performing agents will have less guarantees of performance or ability to obtain desired outcomes.
As AI agents become more sophisticated and integrated into our lives, we can expect to see several trends emerge:
Personalization: As AI agents become more commonplace, they will cater to an individuals' unique needs and preferences, driving innovation across industries and revolutionizing sectors like education through personalized learning experiences and real-time feedback.
Collaboration: The future will see an increased emphasis on collaboration between humans and AI agents, requiring the development of new communication and cooperation methods to achieve a more symbiotic relationship, ultimately transforming how we work and interact with AI technology.
Marketplaces: The emergence of AI agent marketplaces will enable users to find the best agents for their needs, driving competition among developers to create more advanced and efficient agents. This will lead to AI agent specialization for particular tasks or industries, fostering a diverse ecosystem of AI agents with unique expertise and capabilities.
Security and Regulation: As AI agents become more integrated into our lives and critical systems, ensuring their security will be paramount. This will require advanced security measures and protocols to protect AI agents from cyberattacks. Additionally, there will be a growing need for regulation and oversight to address ethical concerns and ensure the safe and responsible use of AI agent technologies, including the development of industry standards, best practices, and regulatory frameworks.
As these agents become more sophisticated, they will play a pivotal role in shaping the future of AI technology and the world around us. The accelerating pace of AI agent innovation promises to bring about a new era of AI-driven advancements, reshaping industries, markets, and our daily lives. Even though people have been building the foundations for these technologies for decades, we have barely begun the first inning and already the potential is increasing. As is repeatedly becoming clear, we live in very interesting times.