Business Advisory Systems
Business advisory systems are technology-driven platforms designed to assist companies in making informed decisions and enhancing their operational efficiency. They have been around for a long time and come in all shapes and sizes. Many managers in planning and operations often rely on advisory systems to provide relevant and timely guidance in order to make the best decisions. These managers are typically very knowledgeable about the business but usually less knowledgeable about technology.
They typically rely on information technology departments for development and support which results in these systems being expensive to develop and difficult to modify and adapt in a rapidly changing world. The personal computer brought spreadsheets to every desk and allowed managers to get up close and personal with data further amplifying productivity, but proficiency using a spreadsheet required the ability to "write code".
Everything changed in 2023 with OpenAI's ChatGPT and the coming-out party for AI.
ChatGPT has a dialog in, dialog out interface (e.g. keyboard in, text out) to a very powerful generative large-language-model AI, OpenAI's GPT-4. It's easy to get lost in conversations with a very smart virtual friend. I'm not sure if GPT-4 passes von Neumann's Turing test but, to me, it's scary close.
The Era of AI Assistants
Then, last month, OpenAI released a beta version of their new Assistant API (Application Programming Interface) and I was able to peak into a new breed of personal advisory system. (https://platform.openai.com/) The API allows anyone to write code that can orchestrate sending data to, and getting data from, GPT-4. This code becomes an "AI Agent" that is very smart while being integrated into the world in which it "lives".
The Rise of AI Agents
Using the Assistant API, the GPT-4 "AI Agent" can automatically be fed information. This information can include numerical data because the API enables GPT-4 to generate code during its analysis of the data. In response to the analysis, the agent can call functions that send information to where it "desires", such as notifications, alerts, alarms and even commands. It can evaluate new data periodically and can access private reference information that can help "it" make better and more relevant decisions.
Once the data has been identified and the agent connects the "sources" and "sinks", the behavior of the agent is controlled by the text prompts. Changing the text prompts is about talking to the AI in our English language, not in computer code. It becomes easy to change what the agent strives to achieve by just changing the description of what the agent is to do, without writing code. As GPT-4 gets smarter and more capable, the text prompts can be changed independent of the software agent code. This increases the agent's ability to adapt to change, without (I'll mention again) writing code.
I coded a Python agent toolkit to experiment with the tools, or rather, play with the toys. I'm building agents that analyze electrical data and weather data as well as any data in the Matlab Thingspeak cloud (https://thingspeak.com/).
I see a huge potential for these agents to lend a helping hand as "personal advisory sidekicks". Agents that have access to the real-world and can help us understand and make better decisions.
While the tools available today are very impressive, there is still a long way to grow.