Autonomous AI Agents, the Smart Assistants of the Future

[The image above is generated by Midjourney. The prompt I used to create the image is listed at the end of this email.]

In his seminal book, "The Four Hour Work Week," Tim Ferris introduced us to the concept of lifestyle design and the idea of working smarter, not harder. He outsourced parts of his life to overseas virtual assistants for $5 per hour. What we will see in the future is powerful Autonomous AI Agents (AAIA) completing time-consuming tasks that will be much more productive and cost-effective; plus, they are never sick or take vacations, managing your life while relaxing. They promise to take over time-consuming tasks, freeing us to focus on strategic decision-making and creative thinking, where human intelligence truly shines.

Autonomous agents, driven by artificial intelligence, can set their tasks when provided with a goal. They can execute these tasks, generate new ones, rearrange their task list based on priority, accomplish the most critical task, and continue this cycle until their goal is achieved.

So, how do these autonomous AI agents differ from general-use agents like Alexa and Siri? The key lies in their ability to understand and adapt to user preferences. While Alexa and Siri can respond to commands and perform tasks, they cannot learn and adapt to the user's unique preferences over time. AAIAs, on the other hand, are designed to learn from each interaction, refining their understanding of the user's preferences and making increasingly tailored decisions over time. 

Autonomous AI agents can serve as highly personalized virtual assistants, managing schedules, responding to emails, booking appointments, and making customized recommendations based on user preferences and habits. This is the use case I am salivating over. I want a capable assistant who works 24 hours a day, is never sick, has no HR issues, and gets better the longer we work together. 

Consider the task of scheduling meetings a necessary but time-consuming part of any business person's day. An autonomous AI agent could handle the scheduling and understand your preferences for meeting times, locations, and even the people you enjoy meeting with. It could consider your energy levels at different times of the day, your dietary preferences for lunch meetings, and your preferred commute times, among other things. The result? A perfectly tailored schedule with minimal input is required from you.

Early incarnations of this technology, like AutoGPT and Baby AGI, have already shown promising results. They've demonstrated an ability to understand complex instructions, carry out tasks, and learn from their mistakes. However, the journey to fully autonomous AI agents is still ongoing.

Technically, this can be achieved through two primary methods: training foundational models or augmenting large language models (LLMs). Foundational models are trained on vast amounts of data, learning to predict and generate human-like text. Augmenting LLMs, on the other hand, involves adding layers of functionality to existing models, enabling them to understand context, make decisions, and learn from feedback.

The reality of autonomous AI agents is closer than you might think. As these technologies continue to evolve, they promise to transform our lives, making us more efficient and freeing up time for us to focus on what truly matters. Just as Ferris envisioned a four-hour work week, autonomous AI agents could usher in a new era of productivity and personalization, where the mundane tasks of today become the automated tasks of tomorrow.

Tip of the Week: Goal-Oriented ChatGPT Prompts

I use ChatGPT for most of my everyday tasks, but initially, I got the best results by keeping the tasks very simple. However, as time passed, I learned how to create a prompt that would give me much more complex outputs but with only one input. Until I have a tried and true solution for an AAIA virtual assistant, I will continue to do my best to create prompts that get the result without a lot of back and forth.

I’ll reference my superprompts from a previous Artificially Intelligent Enterprise Edition edition on prompt engineering to accomplish this.

Let's role play. You have to play the role of PromptEngineer. Your task is to first ask the field in which I need the prompt to be created. (or ask If I want you to generate 5 random prompt Ideas). The prompt ideas should not be simple or straightforward. Suggest prompt ideas that are interactive with users.Example of prompt ideas:After I answer the field, you have to give me a list of 5prompt ideas. Ask me the corresponding number ofprompt ideas to continue or ask me. If I want to have more 5 Ideas. Do it as many times as I tell you. If Ianswer "n" then give another 5 Ideas...Ask me the corresponding number to continue the steps.After I select the number, you have to create a detailed prompt for that Idea. Make the prompt interactive where the prompt will ask for questions and wait for reply. Breakdown like you are asking some questions to me and after I reply to those questions only then you will move on to next steps.

Now I use that prompt to create my following prompt. You can see the prompt dialogue below. Note that the prompt causes ChatGPT to ask for feedback to ensure you get your desired response. I included the dialogue below.

Now that the prompt is done, I hit yes. I may use the output as a blog post, but at the very least, it’s a complete or almost complete output if you are a constant tinkerer like me.

What I Read this Week

• CAMEL: Communicative Agents for "Mind" Exploration of Large Scale Language Model Society - Research Paper on Arvis by Guohao Li et al.

• Generative AI Is Moving Into Adtech. But Which Companies Will Stand The Test Of Time? - Crunchbase

• Forget AI; Organoid Intelligence May Soon Power Our Computers - Forbes

• The Little Book of Deep Learning - François Fleuret

• Rise of New Open-Source AI Model Sparks Startup Acquisition Talks - The Information

• Artificial Intelligence and Business Value: a Literature Review - Information Systems Frontiers

• Microsoft’s Satya Nadella Is Betting Everything on AI - Wired

• Sanofi goes 'all in' on AI with new app to support company's manufacturing, R&D and more - Fierce Pharma

What I Listened to this Week

AI Tools I am Evaluating

• AutoGPT - Auto-GPT is an experimental open-source application showcasing the capabilities of the GPT-4 language model. This program, driven by GPT-4, chains together LLM "thoughts" to autonomously achieve whatever goal you set. As one of the first examples of GPT-4 running fully autonomously, Auto-GPT pushes the boundaries of what is possible with AI.

• BabyAGI - This Python script is an example of an AI-powered task management system. The system uses OpenAI and vector databases such as Chroma or Weaviate to create, prioritize, and execute tasks. The main idea behind this system is that it creates jobs based on the result of previous tasks and a predefined objective. The script then uses OpenAI's natural language processing (NLP) capabilities to create new tasks based on the objective and Chroma/Weaviate to store and retrieve task results for context.

• Glean - This is particularly interesting because it’s one of the first enterprise solutions I have seen for AI assistants. It’s an enterprise product, so I don’t know if it’s designed for a single user. I am signed up for a demo, but I likely won’t be able to take advantage of it for my small shop fully.

• Dante-AI - Create personal chatbots in five minutes and embed them on your website. This is a repeat recommendation, not because it’s great but because it’s relevant. I am still hoping for more of these kinds of offerings.

• The Ultimate AI Prompt Library for Marketers - I don’t know if this is the ultimate list of prompts, but it’s very good for developing ideas to build new prompts and solve common marketing problems.

• AgentGPT - Assemble, configure, and deploy autonomous AI Agents in your browser. I think that this is the type of solution non-technical folks are looking for, but in my experiments, the browser-based agents tend to time out.

• OctoML - Run AI models in the cloud your way. OctoML is taking beta users, anxious to try these AI hosting services.

Midjourney Prompt for Newsletter Header Image

For every issue of the Artificially Intelligent Enterprise, I include the MIdjourney prompt I used to create this edition.