TL;DR: The more artificial intelligence (AI) spreads across everyday life, the more opportunities there are to get paid to train computers. Data annotation, aka data labeling, is a form of data entry you can do from home. It involves labeling text, images, audio, and more, and can pay more than traditional data entry work. Here’s how to avoid scams and get hired somewhere legit.
What are some popular AI jobs?
As artificial intelligence (AI) takes over more of our everyday lives — from Siri and customer-service chat boxes to self-driving cars — humans are still needed to help our future robot overlords AI systems understand the world. You know the “I’m not a robot” ReCAPTCHA tests you take before submitting a form? When you select four photos with crosswalks (or trees or bikes…), you’re actually training an AI. Developing computer systems to adapt based on training and algorithms is called machine learning (ML).
Many of these jobs for humans fall under the umbrella of data annotation. Often used interchangeably with the term data labeling, data annotation involves labeling text, images, video, audio, maps, search results, and ads so AIs can learn what written and spoken words mean, and what static and moving objects are. From there, the AI will be able to recognize them in the future. Chatbots, for example, will want to know that the acronym LMK means “let me know” in certain situations.
Data annotation/labeling is a form of data entry — a type of work that’s been around long before AI and machine learning. Medical billing and coding is an example of old-school data entry, which can pay $15 to $20 per hour on the lower end. Meanwhile, the pay band for AI data annotation specialists tends to be a bit higher, from about $25 to $35 per hour. With increased experience (and even a security clearance), you can get paid more. Some jobs, like online map quality analysts, pay by the task, so rates can vary wildly, from $10 to $50 an hour depending on the ease and simplicity of the assignment.
What’s it like working in AI data annotation?
As you already know — there’s a lot of data out there. So AI data “labelers” and AI annotation specialists have a lot of work to do, across a variety of media. It’s not just text review anymore.
Text annotation
In text annotation, you’re training a computer to understand the written language. That might involve identifying text within images, linking together different words that have similar meanings, and labeling the writer’s underlying feelings and intent behind the words, slang, and phrases written. For example, “Yo, who’s up for some ’za?” might need translating into “Who wants pizza?”
Text annotation can go even deeper into a territory called semantic annotation or tagging, where you attach metadata about related concepts to a text document (or other medium). This could involve highlighting a famous singer’s name and linking it to their date of birth, their most popular songs, and quotes — which are all interlinked in a database. When you flag relationships between things, this helps an AI to recommend related topics, books, music, and movies to users.
Video annotation and image annotation
In video annotation and image annotation, you’re getting paid to help AI recognize specific images and other visuals. This job often involves drawing “bounding boxes” around specific objects, animals, plants, landscape features, and people, and defining them in a way the computer can understand. If you’re working in video, you would have to track the thing you’re identifying throughout the video — either continuously as a stream or frame by frame.
In addition to helping computer models learn what’s inside certain visuals, your annotation work can also be used for accessibility; the words you add to an image’s metadata can also be read aloud for visually impaired people who can’t see what’s on the screen.
Audio annotation
Virtual assistants and customer service robots wouldn’t exist without audio annotation. An AI you’re working with might have basic speech recognition, but it still needs a deep understanding of the customers and lingo specific to its industry. So data labelers working in audio annotation will find themselves transcribing audio, evaluating quality on recordings, and creating audio recordings of their own. And just like with text annotation, they’ll be training the AI about the meaning and intent of various words, phrases, terms, and slang.
Other kinds of AI data annotation and data labeling include:
Social media evaluator: Evaluating the quality and relevance of social media ads targeted to your demographic.
Translator: Translating written and/or spoken words from one language to another.
Search engine evaluator (aka “internet assessor” and “search evaluator”): Measuring the quality of search results and overall user experience with search engines.
Map quality analysts: Evaluating the relevance and accuracy of online maps — including search results, business names, addresses, and pin locations.
Looking for an AI data labeling job? Don’t get scammed
Many data labeling jobs let you work from home and often don’t require a degree or a lot of experience. That means these jobs are in demand, and scammers use this to their advantage.
There are a lot of fraudulent job listings out there for data entry and data annotation, with the intention of getting you to reveal your social security number, bank account info, and other sensitive data. Some scammers will also ask for money during the “interview” process.
To make sure you don’t get burned by a possible job, first look closely at the job description. If the listing is vague and full of misspellings, and the job poster has an unofficial (like from gmail) email address instead of a business email address, it’s probably a scam.
Of course, don’t forget to google the company, even using the terms [company name] + scam or fraud to see if anyone has called them out. Other red flags include not conducting face-to-face ( in person or on a video call) interviews while only communicating via email, and not offering an employment contract before demanding work get started.
Then there are more gray areas. Some AI data entry jobs aren’t technically scams, but they do pay shockingly low wages. For example, some tasks offered by Mechanical Turk will only pay you literal pennies.
How do I get started?
Like we mentioned above, AI data labeling can usually be done remotely from your home and on your own computer, and often don’t require a lot of experience or a college degree.
But there are specific skills and equipment needed, and even location requirements. For starters, you’ll have to be able to sit in one place for an extended amount of time while concentrating on one thing. For video and image data annotation, you’ll need good hand-eye coordination in order to precisely outline the thing (like a jogging person) that you’re labeling. AI companies will want you to work quickly so you can get through huge datasets, so they’ll likely require you to have devices like a computer with a secured high-speed internet connection and/or an iPhone or Android smartphone that’s less than three years old. Other requirements can include a high word-per-minute (WPM) typing speed, cultural awareness and familiarity with current events (for social media evaluating), fluency in written and spoken English, and that you’ve been living in the US for at least the last three years.
The interview stage often involves a test. The company you’re applying to will send you a large document of processes to study, and they’ll give you a set amount of time to complete an evaluation that mimics some of the functions of the job. Passing the test is of course a key to getting hired. But once you pass, you’re off to the races. Have fun and get to labeling!
