Making AI Useful for People

Current AI Models based on Large Language Models (LLM) have a major drawback in that they cannot solve specific problems in a determinstic manner e.g. "Calculate 2 + 2" or "What is the current temperature in Celsius for Hong King". One improvement for this gap in functionality is the use of MRKL modules. Visit the web site of AI21 to view more about one of the pioneers of MRKL technology.

MRKL modules are designed to combine the strengths of neural computation with symbolic computation, to solve problems that are difficult for either one alone. By using modular reasoning and external tools, MRKL systems can leverage the power of existing technology and knowledge bases, while also incorporating natural language understanding to enable communication with humans.

A MRKL system is composed of a set of modules (e.g. a calculator, weather API, database, etc.) and a router that decides how to 'route' incoming natural language queries to the appropriate module.

A simple example of a MRKL system is an AI that can use a calculator app. This is a single module system, where the AI is the router. When asked, "What is 100*1028?" the AI can extract the numbers from the prompt, and then tell the MRKL System to use a calculator app to compute the result. This might look like the following:

What is 10*1028?

CALCULATOR[ 10 * 1028 ]

The MRKL system would see the word CALCULATOR and plug 100*100 into the calculator app. This simple idea can easily be expanded to various symbolic computing tools.

In the example above, the AI Model acts as a router to identify the right module for the given query, which in this case is a calculator app. The router can also handle more complex queries by breaking them down into sub-tasks and routing them to the appropriate modules. For example, if asked "What is the weather like in Hong Kong today?", the router could identify the weather API as the appropriate module, extract the location "Hong King City" from the query, and then ask a weather API for the current weather in that location.

MRKL systems have the potential to enable more intelligent and flexible interactions between humans and computers, by combining the strengths of both neural and symbolic computation.

See here for a nice explanation of MRKL and some examples which can be copied and retried.

See here for an example of a MRKL built using Langchain