AI is learning from humans Many human beings

0
30
AI is learning from humans Many human beings

Namita Pradhan sat at an office in Bhubaneswar, India, about 40 miles from the Bay of Bengal, to watch a video recorded at a hospital across the world.

The video showed someone inside the colon. Bradhan was looking for polyps and a small growth in the large intestine that could cause cancer.

When he found one – it was like a disgusting and angry Magic – he marked it with the mouse and keyboard in his computer, forming a digital circle around the little bulge.

She was not trained as a doctor, but she was helping to teach an artificial intelligence system, which could eventually become a doctor’s job.

Pradhan was one of dozens of young Indian men and women who lived on offices on the fourth floor of a small office building.

He was trained to explain all kinds of digital images, noting everything from stop signs and pedestrians to satellite images to factories and oil tankers.

Aye, most people in the technology industry will tell you, whether their industry has a future, and they are improving rapidly thanks to something like machine learning.

But technical officials rarely discuss the labor-intensive process involved. Artificial intelligence learns from humans. Lots and lots of humans.

Before learning the AI ​​system, one has to name the data they have given. For example, men should refer to polyps.

This work is important for building artificial intelligence such as self-driving cars, surveillance systems, and automated health care.

Technical companies remain silent about this work. They face growing concerns from privacy activists about the large amounts of personal data they store and share with outside companies.

Earlier this year, I took a look at the curtain that Silicon Valley processors rarely give. I traveled all over India and stopped at the facility across the road from Superdome in New Orleans.

Overall, I visited five offices where people do the repetitive work required to teach AI systems, run by a company called iMitit.

The head and experts like intestinal surveyors used to describe a good cough of a bad cough.

It was the language experts and street scene IDs. What is a pedestrian? Is it a double yellow line or a white dotted line? One day, you will need a robot car to see the difference.

What I saw was not like the future – or at least automatic that you can imagine. Offices may be call centers or payment processing centers.

One of these buildings was a pre-built residential building in West Kolkata in the middle of a low-income residential neighborhood with pedestrians, chariots and street vendors.

Like the facilities I saw in Bhubaneswar and other cities in India, China, Nepal, the Philippines, East Africa and the United States, tens of thousands of office staff stared at teaching machines around the clock.

Thousands of workers, and independent contractors usually work at home, also unveiled data through crowdfunding services such as Amazon Mechanical Turk, which will allow digital work to be distributed to independent workers in the United States and other countries.

Workers earn some money for each sign

Based in India, iMerit is called Data for the biggest names in the automotive and technology industries. She publicly rejected the names of these agents, citing confidentiality agreements.

But it was recently revealed that more than 2,000 employees in nine offices around the world are contributing to Amazon’s online data naming service called SageMaker Ground Truth. Previously, Microsoft was listed as a client.

One day, who knows when, artificial intelligence can make the labor market. But right now, it creates relatively low-paying jobs. The data rating market is over $ 500 million in 2018 and will reach $ 1.2 billion by 2023, according to research firm Cogniletica.

The study showed that such work accounts for 80 percent of the time spent building artificial intelligence technology.

Is the work absorbent? It depends on where you live and what you work on. In India, it’s a middle class ticket.

In New Orleans, it’s a good job. For someone who works as an independent contractor, this is often a dead end.

There are skills to learn – such as showing signs of illness in a video or medical examination, or keeping a steady hand when creating a digital car around a car or tree image.

In some cases, when the work involves medical or pornographic videos or violent images, the task is severely changed.

When you see these things for the first time, it is very annoying. “You don’t want to go back to work,” said Christie Milland, who has spent years working on data naming at Amazon Mechanical Turk and becoming a worker on behalf of workers on duty.

“But for those of us who can’t go back to work, you can work for them,” Miland said. Before traveling to India, I tried naming photos on crowdsourcing service, drawing digital boxes around the Nike logo and identifying “unsafe to work” images. I was in pain.

I had to pass a test before starting work. Even this was disappointing. The first three times, I failed.

Labeling images so that people can instantly search a website for retail goods – not to mention the time NSFW spends to identify initial images of naked women and sex toys – has never been inspiring.

AI researchers hope they can build systems that can learn from small amounts of data. But for the future, human action is essential.

“It’s an expanding world, hidden under technology,” said Mary Gray, an anthropologist at Microsoft and co-author of Ghost Work, which explores the data classification market. “It’s hard to get the guy out of the loop.”

The city of temples

Bhubaneshwar is called the city of temples. Ancient Hindu temples thrive in roadside markets on the southwestern edge of the city – where huge stone towers were built in the first millennium.

In the city center, many streets are approximate. Bikes, cows and trucks roam between cows and wild dogs.

 

LEAVE A REPLY

Please enter your comment!
Please enter your name here