Humans scale better than AI

One question we get a lot is “if you’ve got humans in your process, doesn’t that mean you’re unscalable?”.

AI is expressed through software, but that doesn’t mean that AI has the same characteristics of “classic” software. In fact, the incredible scalability of classic software is much, much harder to replicate in many applications of AI.

This is because:

  1. AI critically depends on the data used to train it. Domain-specific data is often required, which by definition is not easily applicable elsewhere.
  2. AI is probabilistic. In applications that require the kinds of reliability only on offer by deterministic methods, you will have to fill in the gap somewhere.

The importance of point 2 cannot be overstated. If you are using AI to build a contract database, then there is only one way you can rely on the output. A person must be involved. Ensuring that every data point is correct. Every ambiguity is flagged and dealt with. Every document is correctly organised into the correct agreement.

AI-first solutions gloss over this point and their customers end up having to clean up the mess. And the mess that AI couldn't handle is always the hardest part. Remember: for a task that cannot be 100% done by AI, there is no substitute for doing the remaining work. Nomio centralises this work and does it so much more effectively than if each customer did it themselves.

We have forged tools and processes that drastically improve efficiency. Each of our Document Programmers can do 50x the work of others. We’ve also done this so many times that we’ve seen all the pitfalls that individual customers haven’t. The output is so much better as a result.

The final secret? We’ve never met a customer who wants to do all of that work themselves. So it doesn’t get done. But we love it.