In 2015, a black software developer said on Twitter that Google's photo service (Google Photos) labeled him and a black friend's photo as a "gorilla" tag, which made Google feel very embarrassed. Google claims to be "shocked and sincerely apologized." A related engineer said that Google is "researching for longer-term solutions."
More than two years later, one of Google’s fixes was to remove the tags of gorillas and other primates from the vocabulary of the service. This flawed solution shows that Google and other technology companies still face many difficulties in advancing image recognition technology, even though they want to be able to use these technologies on autonomous vehicles, personal assistants and other products.
Wired magazine tested the Google Photo Service with 40,000 images, many of which contained animals. Google Photos is a great performer looking for many animals, including pandas and poodles. But when using the words "gorilla", "chimpanzee" and "monkey" to search, Google photo feedback said "no results."
In Google Photo Services, Google conducted a search for the words "gorilla", "chimpanzee" and "monkey".
As a mobile app and website, Google Photos provides a place for 500 million users to manage and back up their personal photos. It uses machine learning technology to automatically group photos, such as lakes or lattes. Similarly, users can use this technology to search their photos.
In the test of Wired magazine, Google Photos did identify some primates. Searching with the words "ç‹’ç‹’", "gibbon", "monkey" and "orangutan" is not bad. You can find capuchin and apes as long as you use the term instead of the words beginning with M.
In another test, Wired uploaded photos of 20 chimpanzees and gorillas from the non-profit organizations Chimpanzee Paradise and Diane Fossey InsTItute. Some orangutans can be searched using keywords such as “forestâ€, “jungle†or “zooâ€, but it turns out that the rest are hard to find.
The conclusion is: in Google photos, ç‹’ç‹’ is ç‹’ç‹’, but monkeys are not monkeys. Gorillas and chimpanzees are invisible.
The Google Lens app on smartphones also does not recognize gorillas.
In the third test, Wired magazine uploaded a set of more than 10,000 images for facial recognition research. Searching with the word "African American", only one picture of an antelope appeared. Enter "black male", "black female" or "black", Google's system will return black and white images, sorted by gender correctly, but not filtered by ethnicity. For people with darker skin, you can use the words "afro" and "African" to search, although the results are mixed, not so accurate.
A Google spokesperson confirmed that after the 2015 incident, the "gorilla" was removed from the search and image tags, and today the tags "orangutans", "chimpanzees" and "monkeys" are still being blocked. The spokesperson wrote in an email: "Image tagging technology is still in its early stages, but unfortunately, it is far from perfect." He highlighted a feature of Google Photos that allows users to report errors.
Google's careful handling of gorilla images illustrates a shortcoming in existing machine learning techniques. With enough data and computing power, the software can be trained to classify images or transcribe them with higher precision. But it can't easily go beyond this experience from training. Even the best algorithms lack the ability to apply common sense or abstract concepts like humans to perfect their interpretation of the world.
Therefore, machine learning engineers must be vigilant about the “edge cases†that are not found in their training data when deploying their results in the real world. Vicente ord ez rom n, a professor at the University of Virginia, said: “Once your system is up and running, it’s hard to model it.†Last year, he participated. In a study, the study showed that machine learning algorithms applied to images can identify and amplify people's biases between different genders.
Photos uploaded by Google photo users were taken under various imperfect conditions. Given the number of images in the database, it is highly unlikely that one type of "humanoid" will be mistaken for another.
Google's parent company, Alphabet, and the broader technology industry face greater risks, such as the use of self-driving cars. Together with his colleague and software reliability expert Baishakhi Ray, Román is exploring how to limit the possible misjudgment of visual systems in scenes such as autonomous vehicles. Ray said that some progress has been made in this area, but it is still unclear how to control the limitations of these systems. “We still can't very specifically know what these machine learning models are learning,†she said.
If you want to know more about the products in FLOW Vape, please click the product details to view parameters, models, pictures, prices and other information about FLOW Vape. And deliver to the destination just cost 7-15davs.We are a legal e-cigarette trader in China. We can provide you with FLOW Disposable Vape. Here you can find the related products in FLOW Vape, we are professional manufacturer of FLOW Vape. We focused on international export product development, production and sales. We have improved quality control processes of FLOW Vape to ensure each export qualified product.
Sufficient supply and favorable price, If you need to pick up goods wholesale, don't miss us.
Folw Vape,Folw Disposable Electronic Cigarette,FLOW Disposable Vape ,Flow Cigarette Cartridge Kit,FLOW Vape Kit
Tsvape E-cigarette Supplier Wholesale/OEM/ODM , https://www.tsvaping.com