Manual Artificial Intelligence

The current state of the art for artificial intelligence (i.e., large language models) is as follows:

  1. Input: Get lots of data, and ideally, categorize it into something less messy and more structured
  2. Analysis: Compress the connections between the data into numeric representations (i.e., an AI model with model weights)
  3. Output: Feed new data through those numeric representations (i.e., tokens in and out)

The problem with this approach is that each of those three major steps is manual. This is why companies like OpenAI, Google, and Anthropic compete for more data, talent, and compute. Because they need humans and machines to:

  • manually collect the data and clean it up (talent, partnerships, web scrapers),
  • manually figure out new mathematical formulas for compressing that data (talent, open source research papers, computer chips), and
  • manually use the tool (users, computer chips).

Automated Artificial Intelligence

But what if we could automate those three steps? Isn't that what nature figured out millennia ago? If so, we could create the foundation of Learning Loops. It works something like this:

  1. Input: Incentivize humans and machines to provide novel and proprietary data in real-time, so the information is constantly streaming in from various sources around the planet.
  2. Analysis: Constantly update the mathematical formulas as it receives new information to create, remove, or edit the connections between the data (i.e., change the architecture of the model and continuously change the weights in the model).
  3. Output: Constantly stream the output signals, as this will also be part of the input signal. This helps with understanding the impact of the change, enabling even further adjustments of the connections between the data in a tighter Learning Loop.

Biologic Intelligence

This seems out of reach since even the most well-funded companies have not released a product using this methodology. So is this possible?

Yes, it's called Biologic Intelligence. And it's been around a while.