The intersection of AI and blockchain - Part 1 | Dabble.AI #9
The Dabble.AI Project #9 - Exploring autonomous business models
Keeping up with the ever-growing number of AI models and tools being released has become impossible. So, it's time to zoom out and start thinking strategically about the business applications for AI. A lot is happening so there is a lot to consider. But topping my list are advances in AI orchestration and the intersection of AI and blockchain technologies - AI decentralization.
This will be a multi-part post because there’s a lot I want to unpack. I'm starting with AI orchestration. I'll discuss the intersection of AI and blockchain technologies in a future post.
Let’s begin with the end in mind - a future when highly autonomous organizations are commonplace. For the sake of this discussion - I’m not talking about solopreneurs running AI lifestyle companies. I’m talking about complex organizations that today would require hundreds or thousands of employees, across multiple departments and divisions, with many millions or billions of dollars in revenue. To see that happen, we’ll need more advanced AI. So let’s start there.
The two paths to more advanced AI
In the simplest sense, there are two paths to more advanced AI - scaling up and scaling out. Scaling up is about making a single AI model more and more capable. This is the current trend (generally) with the push for bigger and bigger LLMs - large language models. However, hardware constraints are a major obstacle to continuing on that path. Just look at NVIDIA's stock performance over the past few years. That illustrates part of the problem - costs being driven up by supply and demand. But at some point, it goes beyond simple economics and becomes a matter of physics.
The short version of the physics constraint is that microprocessors generate heat and everything melts at some point. Advances in quantum, carbon nanotube, and other new computing technologies might help overcome current limitations. But scaling out is the natural evolutionary path in nature and I suspect it will be for AI also, even if we see significant hardware improvements. This is why I think AI orchestration technologies are so important. They are likely the key to significantly more advanced AI systems and potentially the enabler of AGI - artificial general intelligence.
One AI model/provider will never be enough
Even if we could scale up, a single AI model or service provider will never be enough. No single model or provider will do everything better than all the other options. Plus, the best AI for a given task will change over time, as existing systems improve, or new options become available. Also, costs, performance, data privacy, IP protection, and other business-specific considerations will drive decisions to use multiple models, providers, and tools. But whatever the reasons, as the demand for more advanced AI applications increases, staying competitive will require using multiple AI models and tools. That’s where AI orchestration comes in.
So what is AI orchestration?
AI orchestration combines and coordinates multiple AI models, tools, systems, workers (AI and human), and other related resources to accomplish goals that no single AI system or group of humans could accomplish independently. Said another way - it’s about organizing and managing AI and human resources to survive and stay competitive.
Terminology and key concepts
The terminology isn’t standardized, but the key concepts are easily understood. The best way to wrap your head around the general concept is to consider a traditional company with just human workers. In a traditional company, you have employees with different roles and responsibilities. Those employees (or contractors) use their skills, experience, tools, and other resources to accomplish tasks that align with the company's goals and objectives.
AI orchestration is similar. Just swap roles for employees and contractors with roles for AI workers/agents. And just like in a traditional company, an AI agent might be specialized in a particular area or a generalist. And again like in a traditional company, AI workers will have varying levels of skill, specialization, costs, responsibility, and decision-making autonomy.
An AI agent is usually more than a single AI model or tool. It's typically a collection of components, including models, tools, prompts, code, and operating procedures. They are combined within the context of a single AI agent for a defined job or role. So an AI agent is somewhat analogous to a human employee or contractor, it has a role, responsibilities, and skills, and uses tools or other resources to accomplish tasks based on the work it is responsible for.
The most important thing to know is that AI agents can collaborate with other AI agents - and human workers. So to accomplish higher-level objectives, AI agents can work together - just like humans do.
Multi-agent systems and AI agent swarms
As a traditional company grows, it's common to see the workforce grow and become more specialized. Also, when headcount increases, the company becomes more complex and requires more management. The same is true when the multi-agent AI systems or swarms grow. The agents generally become more specialized and become increasingly more complex to coordinate. But, AI agents can also manage and organize the work of other AI agents.
So AI agents don't necessarily need to be directed by humans. Instead, one or more AI agents can be managers of other agents (and humans) with knowledge of their skills or capabilities. They can then reason through the best ways to organize and manage other AI agents and humans based on a given objective. The specifics on how this works are beyond the scope of this post but just like with human employees, it mostly depends on the decision-making atonommy the AI agent is provided with.
Different AI orchestration platforms have different ways of implementing multi-agent systems and again, not all of the platforms use the same terminology. But the general concepts are the same. AI agents can be made aware of other AI and human resources, and then they can reason through how and when they need to collaborate. At this point, AI agent discovery and collaboration is mostly a function of the orchestration platform you decide on. However, protocols like Agent Protocol are emerging to make it easier for AI agents to discover and tap the capabilities of other agents. So, it's a pretty safe bet that soon, AI agents will be able to find and collaborate with other AI agents across different platforms.
The main point is that AI orchestration platforms are more than simple workflow automation systems. Their overarching goal is to enable AI agents to reason through complex workflows - without necessarily needing pre-defined (read: human-defied) steps.
Emerging tools and platforms
There are several emerging AI orchestration tools and platforms. The oldest of them is LangChain which was released in October of 2022. Others include AutoGPT, AutoGen, BabyAGI, and CrewAI to name a few. Plus more to come I’m sure.
I've been dabbling with a few different AI orchestration tools. While each provides its unique features and benefits, they all serve a similar purpose: to make it easier to combine multiple AI models, tools, and agents to accomplish things that a single AI model or system can't. I won’t cover specifics today. I'll save that for another post. But, getting familiar with these tools is a must.
A real-world example of AI orchestration
In previous posts, I've discussed AI for writing and publishing books.Book publishing isn't overly complex. It involves several steps, but the steps are generally well understood. So let's look at a few steps to illustrate the general idea behind AI orchestration.
In a traditional case, we might have a single human author who is responsible for both researching and writing a book. Or, the author might work with one or more human researchers. Human subject matter experts could also be involved to ensure the content is accurate and relevant. Then, when a draft is written, a human editor might be used to review and improve the content. There are many more steps, but you get the idea.
In the AI case, we might have one AI agent do the research, another to write the content, and another for the editing. Or we might have a human in the loop for the editing. The AI agents might use a combination of language models, prompts, contextual data, search engines, and other tools or resources to accomplish their respective tasks. Further, each agent might be given explicit instructions or reason through the process itself. In either case, the agent is responsible for a defined purpose, but the specific steps it takes will depend on how the agent is designed and the context in which it operates.
Building multi-agent systems vs traditional business building
The more I think about AI orchestration, the more I see parallels between building multi-agent systems and traditional company building. In both cases, you have defined goals that requires assemblimg a collection of workers to accomplish those goals. In both cases, workers need clear objectives and access to the resources necessary to do their job.
To align with the company mission and vision, the workers need to be able to work together to accomplish the overarching goals. In the same way, AI agents need to be able to work together to accomplish higher-level objectives. And in both cases, the optimal workflow needs to be reasoned through.
You also have to decide which workers to hire and how they should be organized. Weather to define standard operating procedures or let workers reason through processes on their own. And in both cases, you'll continually evaluate and adjust as needed to improve performance.
So, AI orchestration is much more than a technology consideration. And in the same way that assembling the right people, and positioning them in the right roles is essential to building a successful company; assembling the right AI agents and positioning them in the right roles is essential for building a successful multi-agent AI system.
That's it for Now
I think that's enough for today. In a follow-on post, I’ll expand and discuss how blockchain technologies play into things.