How the robots from Redmadrobot Data Lab tell us about news, cases, and approaches in AI, which can be transferred to our reality and actually used.
Amazon robots as a cause of employee injuries
In 2014, Dave Clarke, Amazon's senior vice president of international operations, inaugurated the company's new warehouse, where small orange robots worked alongside people.
At the time, it looked like the embodiment of bold science fiction predictions. "It's the best for everyone," Clark said.
In fact, everything was not so bright. After 6 years it turned out that the number of automated loaders is directly proportional to the number of injuries suffered by workers.
The load on the workers remained intense, despite the seemingly robot-assisted work process. While previously the norm of scanned parcels at the old warehouses was 100, at the robotic warehouses it was already 400.
The work for employees became monotonous. Before the orange assistants appeared, they used to walk around the warehouse themselves, and now the goods are moved by robot loaders. Workers began to get tired more: standing, doing the same movement hundreds of times in a row is not easy.
Despite the fact that the company has declared the critical importance of safety in warehouses, and the founder and head of Amazon, Jeff Bezos said: "Nothing is more important than the health and well-being of our employees," in 2019, in the order processing centers were registered about 14 thousand injuries. This is almost twice the industry standard.
This data is confirmed by the accounts of Amazon warehouse workers and former security specialists. The latter claim that the company used robots to increase production quotas to such an extent that employees could not cope without harming themselves.
Managers began to realize that something was going wrong, but it was already too late. Management was excited about the capacity, customers were excited about sending the order on the day. And the workers were forced to listen to advice on how to spend their weekend after a 60-hour workweek: just "recharge your battery" and "laugh more".
The workers could carry bags of dog food and cat litter all night long - and sometimes it was tens of thousands of pounds. And it had to be done quickly. Actually, under constant pressure, the workers decided to go the short way.
Instead of climbing the stairs and taking something heavy from the top shelf, they stood on tiptoes and pulled. One hand grabbed a bag from the conveyor belt, and the other pressed a button on the computer screen. Extra hours, tiredness, and stress aggravated the problem.
Last year, the Amazon Security Group took the initiative to "drastically reduce" the number of lost-time injuries. These are serious injuries that will require a weekend off for a worker. However, this initiative was not to prevent injuries - it simply did not allow them to be recorded as lost time.
Many employees who could not do their work in the warehouse due to injuries were transferred to other positions. For example, they were sent to mark up photos to help teach Amazon algorithms.
Why doesn't Amazon do anything about it? Probably because of their obsession with customers. Tim Bray, the former CEO of Amazon, tried to explain the company's actions with one of Amazon's 14 principles: "When we create something new, we realize that we may not be understood for a long time.
AI will build cities.
If 10 years ago someone said that with the help of Minecraft it is possible to solve the issues of modern metropolitan development, you would probably be at least surprised.
But years have passed, and there is an algorithm competition for the construction of cities in Minecraft (Generative Design in Minecraft Competition) - and this is a new approach to designing a quality urban environment.
Today, this competition is held solely for entertainment, but the methods studied by participants could be used in the design of real cities.
Impressions of 2020 GDMC AI Settlement Generation Challenge in Minecraft
The competition is held for the third time already. Participants create a program that can build a realistic-looking settlement on an unknown site in the game.
And if in the early years they looked too artificial - with the same buildings, repeating elements - now they are settlements with a plausible layout adapted to different terrain. Roads cover the slopes of hills, bridges cross the rivers, and furniture appeared in the houses. To achieve this, AI uses different methods: pathfinding, cellular machines, machine training.
Judges evaluate the participants on 4 parameters: adaptation in a particular place, the efficiency of layouts (presence of bridges and roads between different areas), aesthetics and history of the city (whether there are ruins or pits, from which were extracted building materials for houses).
To make a settlement on an invisible map in Minecraft is not difficult even for a child. But the AI is hard to cope with. Therefore, even the winners have failures: for example, in one settlement there were houses covered with sand before the cornice. Perhaps, this was due to the fact that the algorithm wanted to rely on a solid foundation, so it "drowned" the buildings until they touched the stone.
Neural network vs forest fires
California firefighters use AI to fight forest fires. Cal Fire's Predictive Services helps predict the spread of fire and manage resources wisely.
This year in California more than 3.8 million acres (more than 15 thousand square kilometers) of forest were burned as a result of fires. And since mid-August, 29 people have died.
According to Jeff Marshall, head of Cal Fire's Predictive Services, fire forecasting tools help officials understand how big fire can be, where it can go, and how quickly you need to start evacuating the population. Cal Fire's Predictive Services also simulates seasonal trends in fire propagation.
This is not the first use of neural networks in predicting weather disasters. For example, they are used to decipher mysterious earthquakes in California and to predict floods in India.
AI in agriculture
Artificial intelligence is capturing more and more new industries. And if before neural networks and machine training were used mainly in the modernization of agricultural machinery, now they are used to breed salmon and plant strawberries. But let's take it in order.
Norwegian company Telenor is testing models of AI, created to reduce costs and improve the efficiency of breeding salmon, one of the country's main export goods. These models are used to optimize feeding, keep fish clean and healthy, and help companies make the most effective decisions regarding business operations.
For example, one of the algorithms helps fish workers understand how to feed salmon.
According to Björn Taale Sandberg, head of Telenor Research, their technology can analyze camera images to determine when the behavior of a flock will change. This, in turn, will alert the observers that the fish are no longer hungry. This is an important function since about 40% of the cost of maintaining a fish farm is food. And if the food is given too much, it falls to the bottom, which can cause environmental problems.
Another application helps to identify individual salmon. The fact is that salmon on the farms are cleaned of lice (this is really a big problem for fish health).
But, according to Sandberg, fish that are cleaned too often can be stressed, and this will lead to other problems. In addition, some farms use cleaning chemicals, the excessive use of which is harmful to the environment.
In addition, the company is developing small handheld devices with machine training for remote fish farms, where the Internet may not be good enough to support connections between ocean or fjord farms and operators onshore.
In such devices, the neural network will be trained locally and the tool will be able to automate farm solutions without connecting to the shore.
Meanwhile, in China AI has taught remote strawberry growing.
China is hosting the "Smart Agriculture" competition, organized by Pinduoduo, the country's largest e-commerce platform. As part of the competition, a team of researchers decided to apply AI technology for remote cultivation of strawberries in automated greenhouses.
They came into confrontation with ordinary farmers to understand whose actions would bring the greatest economic benefit.
How is the competition going? A team of researchers formulates and optimizes AI solutions based on data on greenhouse growth and conditions obtained from IoT devices, cameras, and sensors.
AI improves the picture in video calls
NVIDIA developed the Maxine platform, which improves bandwidth for video communications.
Developers have refused to stream each pixel of the image. Instead, Maxine analyzes the features of the speaker and "builds" this face in the video of his interlocutor. This significantly reduces the amount of data transferred during a video conference.
The platform improves picture quality, removes or blurs the background, directs the speaker's gaze into the camera, adds subtitles, removes noise, and allows you to create your own avatars.
One-line
Microsoft received an exclusive license for GPT-3. Musk outraged;
Ministry of Education, Ministry of Internal Affairs, Ministry of Emergency Situations, Rosmolodezh, Ministry of Health, FOMS, Ministry of Transport, and Ministry of Agriculture presented projects on the implementation of AI in their agencies;
How would you look if you lived in Africa, Asia, India, or Europe? Insert your photo into the Pix2Mix neural network and check!
Fun Al
How would you look if you lived in Africa, Asia, India, or Europe? Insert your photo into the Pix2Mix neural network and check!
If you liked it or didn't like it, write in the comments. After all, you can't retrain the robot without feedback. The latest news about AI and not only on our Telegram channel.