The AI that Codes Itself

We have seen tremendous progress in AI over the past few years. However, all of those projects were developed by human teams. This is a slow process. The future AI leader will be a company that creates a self-coding application.

Estimand is at the beginning of applying causal AI to reduce defects in software code. The current aim is to work alongside human programmers, but the gap between improving code and writing it is small. This is not the same as an LLM rehashing human examples.

Why it Matters

Intelligent machines currently rely on mathematical models inspired by psychology. This aspect of AI is progressing well. The issue is that for every 1000 lines of code, there are over 100 defects [1]. While that statistic is from a few years ago, it has worsened.

Today, apps are slower than before, with an exponential increase in size [2]. However, the provided value has stayed the same. These inefficiencies cause popular AI applications like ChatGPT to require more electricity than 130 homes use in a year [3].

The Inefficient Programmer

Development teams spend 20% of their time fixing bugs [1]. The root cause of this issue is a disconnect between logical machines and nonlogical humans [4]. Zooming out, we see that there is also a significant difference…