Get excited! The future is here: machine learning and AI allow you to automate the ineffable! Sounds great… if only you knew how to spot a good use case.
To help you take advantage of this amazing technology, I have a neat trick you can use to identify tasks that are perfect for machine learning and AI. It’s a sort of guided meditation designed to bring out your aha! moment.
Imagine AI is a hoax
I want you to imagine for a moment that there is no machine learning. It’s all a hoax; there’s just an island in the middle of the ocean somewhere with a bunch of my friends behind computer screens pretending to be an AI. When you send them an input, they quickly send back a decision.
For example, you send that cat photo and they sneakily type ‘cat’ for you so it looks like cool machine learning.
Here’s the fun part. I’m going to let you borrow my island to use for your business. Free of charge. There’s only catch: you can’t write them any instructions — my friends are weird like that. You can only teach them with examples, but they’re fast and numerous. It’s a pain to spend time teaching them if you have a one-off task, so focus instead on the repetitive drudgery you’d like to cut out of your life. What would you use this island for?
What repetitive drudgery would you offload?
By answering that, you’re on your way to the right applications for machine learning and AI. (You can get started without knowing the difference between ML and AI.) This is thing-labeling and once you get creative with that idea, you realize there’s so much you can achieve with it!
Here are just a few examples of machine learning labels that I’ve seen in real life:
- Bottle of wine: will I like it or not? (Things the author does for fun.)
- Video game: joystick motion?
- Sound recording: correct text captions?
- Song lyrics: language?
- Toilet user: did they wash their hands?
- … list continues at the end of the article…
These are all labels -small decisions- and the island can learn to do all these things for you… if you give it enough examples first. The workers are willing to learn and they have all the time in the world.
How drunk are these people?
But wait! Before you offload all your work to this island, consider: How drunk are these people? Can they even do your task? Don’t just trust them with important work. Blind trust is a terrible thing! Force them to earn your trust by checking that they actually perform your task well enough.
Blind trust is a terrible thing! Machine learning should earn your trust by performing well when you test it.
You can’t do this unless you can express what it means to do your task right. A common mistake businesses make is to assume machine learning is magic, so it’s okay to skip thinking about what it means to do the task well.
You need to know what it means to do your task right
Hang on, you care about getting your task done right and yet you can’t say whether it actually was? Sounds like a serious problem! Before you’re even ready to think about data, make sure you’ve thought about how you’d know whether a unit of work was performed correctly. In other words, whether a label the island gives you is correct.
Unless the business decision-maker responsible for your project is able to articulate how to score performance, machine learning is a nonstarter for you. You’re not ready to dive into a serious machine learning project until you are in possession of a document that outlines:
- What does it mean to do your task correctly?
- Which mistakes are worse than which other mistakes?
- If 1000 units of work were done — some imperfectly — how do you give that pile of work a score? (That’s entirely up to your business leader.)
AI projects should start with the business decision-maker
I hope you see that you don’t need a PhD in machine learning to produce such a document. Instead, you need to understand what’s important for your business. Every data science project begins with a business decision-maker, and machine learning is no exception. Don’t worry, I’ve got your back. I’ll soon be sharing guides to help business decision-maker get started, design performance metrics, and set criteria.
A short rubric for getting started
In the meantime, grab a pen and paper. Forget about machines and imagine a (drunk?) island of workers. Which repetitive tasks would you have it help you with? Can you express the recipe for doing the task? If yes, just have a software engineer translate those instructions into code for you. If no, can you say how you’d score 1000 imperfect units of work? If no, keep meditating. If yes, welcome to machine learning!
(Since this is always the first step in a machine learning project, if anyone asks whether you’re doing ML/AI, you can say “Yes!” with a clean conscience. Congratulations!)
For more inspiration
Here are just a few examples of machine learning labels that I’ve seen in real life:
- Baby food ingredient: safe or spoiled?
- Patient: ideal medication dosage?
- Email: spam or ham?
- Recorded phone call to call center: issue topic?
- Bottle of wine: will I like it or not?
- Steering wheel: left or right?
- Photo: which animal?
- Game piece: which location on the board?
- Start of a sentence: end of that sentence?
- Stock: tomorrow’s price?
- Transaction: legitimate or fraudulent?
- Data center cooling system: warmer or cooler?
- Machine: when will it need maintenance?
- Inventory: when to restock?
- Scene description: pixels in a visual rendering?
- Today’s temperature: tomorrow’s temperature?
- Auction: how much to bid?
- Movie: will you like it or not?
- Live lecture: text captions?
- Poem: what does it sound like out loud?
- Image of an invoice: total amount in dollars?
- Service request: waiting time?
- Expense report: budget category?
- Sound recording: correct text captions?
- Song lyrics: language?
- Sentence in English: same meaning in Chinese?
- Form incorrectly filled out: correct fields?
- Clothing item: skirt or blouse or …?
- Video: which actors?
- Video game: joystick motion?
- Toilet user: did they wash their hands?