Many sci-fi films predicted this day would come, but who knew it would come so soon?Before long, AI-powered robots may take over, but not the world — more likely, people's households.Household AI-powered robots are a thing of the here and now as well as the near-distant future, since some are already on the market, others have just been announced to be released soon, and better, more advanced versions are predicted to be out in two decades, as the Globe and Mail reports.In fact, RBC Capital Market previously predicted the humanoid robot market would be worth $9 trillion USD by 2050..As many of these robots are being developed presently, developers are looking for data to program them with.One of the ways they are collecting this data is by paying people for doing many different monotonous household tasks. All people have to do is buy a head strap to attach their phone and start filming themselves doing things like watering plants, washing toilets, sorting home supplies, etc.This is work that one Nigerian woman, going under the pseudonym Omolara, told the Globe and Mail she does for an app called Atlas Capture. .Atlas Capture, which invites users to "Earn cash completing quick video tasks from your phone," according to the description on the app store.Omolara heard about the app from a friend.She was skeptical, but she signed up recording many home chores Atlas Capture requested: mopping the floors, cleaning the bathtub, and picking weeds.She found filming herself a bit stressful, since she could not sit down when recording and had to ensure both of her hands were visible at all times.Some jobs she recorded also seemed to require more footage than was necessary, or practical — like cleaning a toilet for 20 minutes. .Omolara would record in the morning and upload at night, and by the end of the week, her culminated 17 hours of recordings were worth $50 USD ($72 CAD) to Atlas Capture, about $2.94 USD ($4.19 CAD) per hour.Currently, Omolara is using this app and similar ones to earn money while she looks for a job, with many other users probably doing something similar, as the app claims to have over a million users.It also says it has paid out over $5 million to users overall.In return, Atlas Capture and similar apps take these videos and use them to train AI robots to become the ultimate housekeepers..Until recently, on the app store, Atlas Capture stated its copyright holder as Carbon Based Technology Corp, a registered Canadian company doing business for Mecka AI.Mecka has raised more than $120 million USD ($171 million CAD), with the company projected to make $300 million USD ($427 million CAD) in revenue by the end of the year.It is one of many companies providing data to perfect general-purpose robots to work in factories, warehouses, and to vacuum floors, and load dishwashers in homes — all you could ever want.Over the past few years, researchers working on programming these robots have discovered it is essential to collect first-person video to allow the robots to learn tasks..Similar to how AI's large language models (LLMs) such as chatbots ingest humanity's written and visual outputs, AI robots need to understand how to physically interact with the world."What we truly want to do is learn from the physical intelligence of humans," stated Danfei Xu, an assistant professor at Georgia Tech and an Nvidia researcher who worked with Mecka.This learning process can take a long time (predicted to take 20 years to perfect), since AI's ability to learn tasks and perform them autonomously in different environments takes considerable input.In 2022, researchers from Stanford University and Meta used a compilation of 3,670 hours of egocentric footage from 931 contributors, known as the Ego4D compilation, to power a robotic arm.This programming proved to be successful at training the arm to do tasks like folding towels, closing drawers, etc..The research highlighted the need for this type of egocentric video.Similarly, Nvidia, an AI company, trained an AI model with 20,854 hours of video, finding a clear relationship between video quantity and performance.Eventually, the goal for these companies is to build robot models that can generalize and learn new acts of prehension, manipulation, and dexterity, without massive amounts of training material."Once we have built the perfect general-purpose robot that has an innate learning algorithm similar to what humans can learn, we may not need companies collecting human data," stated Igor Gilitschenski, an assistant professor at the University of Toronto studying AI and robots.."It can easily be 20 more years, but who knows?" he added.As for the present, there are a few notable AI-bots which could be yours if you're willing to prepare your wallet for a huge incoming cost.As the food professor, or Dr. Sylvain Charlebois, tweets on X, "Early home robots are advertised at US$30,000, but capabilities remain limited, or $699/month to rent."LG's CLOiD is the company's first-ever AI-enabled home robot and has been announced publicly, though no release date has been revealed yet, but in the demo clips CLOiD folded laundry, grabbed milk from the fridge, and cleaned up a mess on the floor.No price has also been listed yet either.
