Task-oriented Autonomous Agents
A task-oriented autonomous AI agent functions independently to accomplish specific tasks or achieve predefined objectives, adjusting priorities, learning from past experiences, and executing tasks without constant human oversight.
These agents are engineered to perceive their surroundings, make decisions, and take actions based on their programmed directives and acquired data.
For instance, Yohei Nakajima released the BabyAGI project as an example, featuring AI agents interacting with each other. The BabyAGI includes three distinct agents:
- Task Execution Agent, responsible for handling tasks sequentially.
- Task Creation Agent, generating new tasks based on previous outcomes to reach specific goals.
- Task Prioritization Agent, organizing tasks according to their importance.
Agents involve an LLM to perform the following steps [5]:
- Decide which action to perform, based on the user input or its previous outputs.
- Perform the action.
- Observe the output.
- Repeat the first three steps until it completes the task defined in the user input to the best of its abilities.
Balance AI POCs
We have conducted several Proof Of Concept(s) utilizing most promising libraries. We have created POSs using the following Agent frameworks:
AutoGPT is a generalist agent, meaning it is not designed with a specific task in mind. Instead, it is designed to be able to execute a wide range of tasks across many disciplines, as long as it can be done on a computer.
BabyAGI 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 tasks 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
Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.
We have utilized CrewAI framework to build a simulation of financial analysts team using AI agents.
The goal of the team is to perform professional research on a given cryptocurrency token and the project behind it.
The designed system (based on [3]) consists of the following agents:
- Financial Analyst
- Staff Research Analyst
- Private Investment Advisor
They utilize the following tasks:
- Research
- Financial Analysis
- Announcement Analysis
- Recommend
The tasks use CrewAI and Langchain [4] capabilities to create and execute tasks as well as search information on the internet including parsing websites for relevant information.
As the BRAIN of the system we are using LLM (Large language model) [6] model. We have experimented with:
- OpenAI — gpt-3.5-turbo (via API)
- OpenAI — gpt-4 (via API)
- Llama2 (locally)
The system spawns agents crew (set of autonomous agents) based on the crypto token entered as prompt.
The agents performs various tasks based on their description and systems comes back with the full report.
The example report for Bitcoin prompt:
We have decided to open-source the example POC described here. It can be found at our repository at: https://github.com/balancedao/task-oriented-agents-poc
Balance AI Context
We are planning to use similar approach in our Agents VM (as part of V2 upgrade).
We are planning to use few types of agents one of them being:
- Task-oriented Autonomous Agent
Utilizing similar technology stack as aforementioned frameworks we can achieve the agents execution within our Agents VM.
While storing the agents description on-chain (or off-chain as encrypted IPFS data), we would allow to control the agents creation and execution.
We are working on a specific protocol to describe the agents environment (including roles, tasks, goals and tools). The protocol would allow to store that info on-chain and allow execution within our Agents VM.
Sources:
- [1] https://attri.ai/generative-ai-wiki/task-driven-autonomous-ai-agent
- [2] https://github.com/yoheinakajima/babyagi?tab=readme-ov-file
- [3] https://github.com/joaomdmoura/crewAI-examples
- [4] https://python.langchain.com/docs/get_started/introduction
- [5] https://analyzingalpha.com/langchain-python-tutorial#Agents
- [6] https://en.wikipedia.org/wiki/Large_language_model
Disclaimer:
Any analysis, information or explanation we give to you about operations and performance on your trading account is not intended to be and should not be considered as advice. We do not provide investment, financial, legal or tax advice, especially we do not provide advice to conclude any transactions. This tool shall only analyse your trades, explain how it works, what is the trend, what is the possible risk etc. If you decide to use the information provided at the service, you do so at your own discretion and risk. The decision to conclude any transaction, including buying, selling, trading in securities or any other investments rest solely on you. Any trading or investment transactions involve a risk of substantial loses and shall be made based on the personalised investment advice of qualified financial professionals. We are not liable for any loss or damage that you, or any other person or entity incurs, as a result of any trading or investment transactions based on any information provided at the service.
To stay updated with our latest news and developments, be sure to follow us on our official channels.
