A Council of AI Advisors—Turning a Mixture of Experts into Your Personal AI Advisory Board
When the President needs advice, he has his Cabinet, his Council of Economic Advisors, the Joint Chiefs, and so on (I wouldn’t be surprised to learn there’s an advisory council to identify advisory councils). Similarly, CEO’s have boards they can go to for advice or to brainstorm ideas. A reliable, available council of advisors is a valuable resource that most of us don’t have access to. Sure, we can talk to our friends or whomever happens to be available but that’s not quite the same.
How Can AI Help?
An article in MIT Technology Review1 finds that popular uses for AI are brainstorming and asking for advice. That’s only a start: AI provides a powerful opportunity to create your own personal council of advisors. An AI can be trained, or prompted, to mimic a human expert in a variety of fields. Not unlike a real expert, an AI never gets tired of showing off its knowledge. Although a set of AI models may not replace a council of human advisors, for those of us who don’t have access to such a council, AI can provide a reasonable alternative—provided it’s used appropriately.
Although we call AI “artificial intelligence,” a more accurate term might be “imitation intelligence.” AI Large Language Models (LLMs) are designed to mimic human behaviors, and they are very good at it. However, there’s no there there. No matter how human-like their behavior, LLMs are just very sophisticated computer programs and prompting is really programming with natural language. An AI LLM has no awareness and no context other than what you provide when you write a prompt—LLMs are probability engines that do an extremely good job of choosing the most likely response from a given set of options.
The prompt you give to an AI LLM determines the range of possible options it will use to generate its responses. The larger the set of options, the more generic the response; thus, a generic prompt yields a generic response. A more focused prompt usually yields a more focused result. Of course, an AI can also yield a nonsensical or hallucinatory result and figuring out when that’s happening isn’t always that easy: one lawyer famously got into trouble when he failed to notice that the AI he was using for his research included references to non-existent court decisions.
A Mixture of Experts
Put together enough different AI models and you have a mixture of experts—at least, that’s the idea. However, successfully tapping into that expertise requires very good prompt writing skills, the ability to manage a bunch of different LLMs, and the ability to keep track of which one you’re speaking with. Otherwise, it’s easy to end up with chaos. Just imagine a roomful of experts nattering on incessantly, interrupting or talking over each other and you won’t be far off, although the experts probably won’t randomly hallucinate.
However, AI Infrastructure Software turns a mixture of experts into a highly useful council of AI advisors.
From Chaos to Useful Advice
AI Infrastructure Software provides a framework for managing multiple AI models—usually, though not exclusively, LLMs. It manages communication with the AI models, automatically enhancing prompts and eliminating the chaos of trying to manually keep track of which AI you are talking to at any given moment. Infrastructure software brings AI to the data, sharply decreasing data movement and avoiding potential security risks. It enables you to distribute processing so that your systems are not overwhelmed by spikes in demand and enables you to scale AI initiatives. In addition, AI infrastructure software provides:
- Ability to connect to your choice of AI models
- AI-enhanced natural language search capabilities
- Enhanced prompt engineering
- Retrieval Augmented Generation (RAG)
- Context aware relevancy ranking
- Personalized advice
Infrastructure software lets you have a natural conversation with your Council of AI Advisors. It makes it obvious which AI you are talking to and enables you to cross-check answers from one AI against another. Infrastructure software helps you verify that different AI experts are giving you useful advice, not generic nonsense or hallucinations. Infrastructure software will personalize your interactions with your AI advisors by including information from any local data sources you wish to use (e.g. email, Slack, Webex, and so on) to further contextualize your prompts and ground results in facts.
A Council of AI Advisors
While an AI may not be creative in the human sense of the word, AI models have access to a lot of information. They can help you process information, explore possibilities, and consider likely outcomes of a decision. Like good human advisors, they are very good at helping you identify opportunities and options that you would never have thought of on your own. Assuming, of course, that you can manage the interactions in a way that returns useful results and doesn’t drive you nuts.
With AI Infrastructure Software, you can have your own personal council of AI advisors, and you can populate it with the mixture of experts you need when you need it. The infrastructure software does the heavy lifting of managing interactions and enhancing prompts. If you need more information on a topic, infrastructure software will use AI to find you useful, context relevant information. There is no question that infrastructure software is the missing link that makes AI truly useful.