Discover more from Addition
Teaching AI to think it's way through complex, knowledge-based tasks
In some of our previous posts, we've explored how AI systems can perform complex tasks by sequencing multiple generations together. For example, back in October, we demonstrated how to generate ads by breaking the project into roles, including a synthetic "Creative Director," a "Synthetic Art Director," and a synthetic "Production Artist."
You can think of this as the Ford Model-T plant of AI. Each link in the chain has a defined job to do and a fixed order in which it's performed, and in the end, it's able to output something more complex and novel.
Lately, we've been exploring a new paradigm called agents. Agents expand on the assembly-line concept in a fundamental and transformative way. Instead of executing a predefined sequence, AI learns to improvise across various tools and resources to achieve an outcome. This is accomplished by teaching AI to generate “thoughts” after each task, and using this thought process to determine the next course of action. By teaching AI to think it’s way through complex tasks, AI systems can evolve from something that resembles a Model-T factory to something that resembles a knowledge-based company in the present day.
Iterating Across Tasks
One massive benefit of "agents" is that they can accomplish complex tasks that require iteration. This makes them particularly well suited to dealing with imperfect data sets and tapping into the AI brain's analytical and creative sides.
Imagine a system capable of performing research (an analytical task) and brainstorming (a creative task) to generate an output based on a user's prompt.
This system will include two functions that it can iterate through.
Research: Our AI system will be able to recursively research a given topic by leveraging a suite of tools such as News Search, Google Search, and web-scrapping.
Brainstorming: We'll also teach our AI how to generate ideas in response to its research findings. This will allow it to make creative leaps versus simply recycling the information it finds.
Rather than going through a linear process to determine an insight, the AI system recursively iterates through its capabilities until it feels like it's ready to generate a final output.
Example: Data-Driven Insights
To demonstrate how our autonomous AI performs in the real world, let's put it to work on a complex task: coming up with creative insights.
Creative insights are notoriously challenging. Large brands often spend months researching and millions of dollars pursuing them. And for a good reason. Insights have to be rooted in credible information. But they must also reveal a new perspective on a given subject, thus requiring them to be creative and unconventional. This is a great use case for our autonomous AI.
To teach our AI system how to generate insights, we will train its "brainstorming" capability with some examples of high-quality creative insights from Advertising Strategy Director Julian Cole's Linkedin Posts like this one.
With this style of output in mind, we can give our AI system a prompt, and it will research the subject and use its research findings to come up with insights. It will iterate through this process until it feels like it's completed the task.
Let’s see how it does:
Me: Come up with an insight about our relationship with Artificial Intelligence.
The AI begins combing through sources like this one via it’s search capabilities.
And after a few cycles it generates an insight for me.
“AI isn't just a tool, but a partner that must be nurtured and cultivated in order to reach its full potential.”
Let’s try a few more…
Me: Come up with an insight about Skittles.
“Skittles have transcended time and cultures, showing that it's more than just a candy - it’s a universal symbol of joy and happiness.”
Me: Come up with an insight about TikTok culture.
“TikTok is more than just a platform for entertainment and self-expression. It's shaping our lives in ways we don't always realize, creating an ecosystem of influence that extends far beyond the app itself.”
Prompt: Come up with an insight about gamers
“Gaming is often seen as a male-dominated activity, but it can be a powerful platform for female empowerment if given the right attention and support.”
The ability for AI to improvise across tools, capabilities and resources unlocks a range of new use cases. For example, imagine a brand that personalizes all aspects of their marketing to make it more relevant and engaging for consumers. Accomplishing this requires analysis: understanding a customer and what available sets of products, services, and information might appeal to them. But it also involves creativity: the ability to deliver a message with emotion, humor, or provocation. AI can achieve this and more by tapping into its full range of capabilities with autonomy.
PS – if you’re interested in trying our “Insights Generator” we will be releasing it as a free tool. Drop me a line by replying to this email if you’d like early access in exchange for your feedback.