---
title: "Teaching a Computer to Ghoti"
description: "Discover how AI's unsupervised learning revolutionizes complex domains like legal \u0026 translation by mastering rules without explicit programming, unlike traditional coding."
categories: ["general"]
keywords: ["unsupervised learning","AI in legal","AI in translation","text-to-speech technology","complex systems","programming rules","Tomasz Tunguz","Theory Ventures","natural language processing","content generation"]
ai_summary: "Explore how unsupervised learning enables AI to master complex domains like translation and legal without explicit programming."
date: 2023-08-15
lastmod: 2026-07-28
canonical_url: https://www.tomtunguz.com/teaching-a-computer-to-ghoti/
author: "Tomasz Tunguz"
---

Classical computer programming is about rules. What does a program do when a user inputs 4 + 7? Or a credit card number? Or clicks on a button. I spent years as a programmer translating human rules into rules the computer would understand.

One day, with enough experience, the number of exceptions to rules dawns on a programmer. It's no different than english. 

Imagine writing a text-to-speech program : 

Home & tome rhyme. So does comb. But not come. 

gh in weigh & daughter is silent but in laugh the gh sounds like an f. 

Compare ti in titanium with ti in celebration.

Taken to the extreme, we can spell fish in English as [ghoti](https://en.wikipedia.org/wiki/Ghoti). The gh from laugh. The o from women. The ti from celebration. 

One of the key innovations in this wave of AI is these systems is to learn the rules of a system without having to be taught. 

That's why AI's impact in complex domains like legal, translation, text-to-speech, & content generation has been so immense. 

These domains have their own intricacies which would be an onerous task for a team of programmers but the unsupervised learning techniques that have been developed over the last decade learn the right pronunciation for each word.

AI teaches computers to ghoti.