Often human work behind AI
Artificial intelligence on the outside, human on the inside
By Max Bourne
April 3, 2024, 3:46 p.m
Listen to article
This audio version was artificially generated. More info | Send feedback
A supposedly AI-based service whose 3D models are actually designed by humans, seemingly autonomous delivery robots that are actually remote-controlled: It is not only at Amazon that human work is primarily behind some of the supposed artificial intelligence. This often has a very banal reason.
Amazon, one of the largest technology companies in the world, has made a laughing stock of itself: The supposedly revolutionary technology that made supermarket shopping possible without a checkout was actually based less on artificial intelligence and more on the fact that more than 1,000 workers in India monitored customers while they shop using cameras and recorded all goods by hand. AI (Artificial Intelligence) on Amazon probably means “Actually, Indians” or “All Indians,” users scoff at X.
For Amazon, cashierless technology is obviously a failure and is now largely being phased out. However, the fact that a high proportion of human work is hidden behind a supposed AI system is not an isolated case. In general, AI systems need a large amount of human support to learn certain skills. And not just during development, but also permanently, during the use of these systems.
OpenAI, the company behind the voice AI ChatGPT, and other technology companies now employ tens of thousands of people in low-wage countries – including Africa. In Kenya and Uganda, for example, an entire industry has emerged from companies that provide AI companies with employees to train AI models. Amazon itself operates the “Mechanical Turk,” a platform for conveying tasks that only human employees can perform. These are often simple tasks, like adding descriptions to images to train AI models. Most workers come from emerging or developing countries. There are often complaints about poor or completely non-existent pay.
AI more expensive than humans
In some cases, however, the work of human employees goes far beyond training AI models. Last year, the startup Kaedim caused a stir because it said it had developed an AI-based technology that could convert two-dimensional images into three-dimensional models. In fact, a significant portion of this work was done by human employees working around the clock in various countries. According to a report by technology portal 404 Media, they had to create models within 15 minutes to maintain the appearance of an automated process.
In another high-profile case, the company Kiwi-Bot gave customers in the US the impression that its delivery vehicles, known as robots, were driving autonomously. In fact, employees in Colombia controlled the small vans carrying food orders or other deliveries via camera. This was probably not only the more reliable solution, but thanks to an hourly wage of two dollars, it was also probably the cheaper one compared to autonomous vehicle control.
So far, it is not just the case that AI is not yet able to fulfill many of the tasks that companies had long hoped for. This includes completely autonomous driving. Due to the high computing power required, operating AI systems is expensive. Many of the companies that have presented spectacular AI technologies in recent years, for example for image and speech generation, are accumulating losses worth billions.
How these companies plan to reduce costs and generate profits in the foreseeable future is unclear. Because the ever larger and more powerful chips and data centers they require are currently becoming more expensive, not cheaper. This means that sometimes using AI is more expensive than using humans for the same task. This is especially true when this task can be completed thousands of kilometers away in low-wage countries.