domeet webmaster

“Brush your face to unlock your phone”, “Paint your face”, “Paint face verification and apply for a bank card”… Our innate “face” is gradually becoming the only one of our identity except for the image of the face value. Identifying the logo and gradually deepening into our daily lives has provided us with great convenience.

But when face recognition meets reality, there is no shortage of lying guns. When it’s terrible, like our protagonist today, Rekognition from Amazon.

Rekognition is an image analysis software developed by Amazon. Amazon has sold the technology to several US law enforcement agencies, such as the Washington County Sheriff’s Office, which aims to use software’s facial recognition algorithms to automatically identify and track suspects in real time; Amazon is also fighting to get this Technology is sold to the US government.

What is this magical algorithm, and you can see at a glance whether the other person is a suspect?

Rekognition requires a recognized face photo as the underlying database. The current base database is built primarily on Amazon’s cloud service AWS. When the camera captures a realistic face or a user uploads a comparison photo, Rekognition automatically compares it to the photos in the underlying database; if the similarity is high enough, Rekognition considers the person to be someone in the underlying database. In software, this similarity is called the “trustworthiness threshold” and its default value is 80%.

Using advanced technology to let criminal suspects run out of nowhere, imagine that the world seems to be getting better. However, this idea has also been opposed by many people, such as the American Civil Liberties Union (ACLU, a non-profit organization) that has been sung against Rekognition.

On August 14th, the ACLU conducted a test on it. The performance of Rekognition was astounding: Based on the basic database of 25,000 suspects, ACLU uploaded 120 photos of legislators to let Rekognition try. As a result, Rekognition proved that he was not good at using missing 26 legislators as the result of the suspect.

In fact, this is not the first time that Rekognition has “made a mistake.” As early as 2018, it has mistakenly identified 28 members of Congress as criminals. Unexpectedly, after one year of hard work, it has no hesitation. The road is farther and farther away on the way to identify government officials as criminals…

It must be said that this software provides us with a dramatic “answer”, but as a software that plays an important role in identity recognition, such a result cannot be rigorous and serious.Questioned.

San Francisco lawmaker Phil Ting proposed to ban the use of this software system, and said that such misidentification results will have a huge negative impact on people, when you are considered a criminal, “may affect Your ability to find a job, affect your ability to buy a house and rent a house.” This is far from being a ridiculous mistake made by a technology, but it is true that there are real people who are affected by this mistake.

So, in the face of another major mistake, what is the explanation for Amazon as a software developer? Forbes reports: “When using facial recognition to determine suspects in an investigation, law enforcement should use the recommended 99% confidence threshold and use only those predictions as an element of the investigation (not the sole determinant).” .

ACLU indicates that this explanation is not convincing. The 80% confidence threshold used in the test is actually the default setting provided by the software. Since the 99% confidence threshold is the recommended value, why doesn’t Amazon set the default confidence threshold to 99%? However, Amazon did not give a reply.

In fact, the software has been promoted and tested throughout the United States. In addition to problems in recognition accuracy, a number of other issues have been exposed in trials around the country. For example, due to accuracy, identification errors in colored people’s settlements may cause racial discrimination and cause conflicts between police and civilians. During the trial in Orlando, Florida, there are also problems such as network bandwidth problems causing software to fail to operate properly. Obviously, Rekognition is not ready for true promotion and application. In addition, the abuse of face recognition technology also violates the privacy and freedom of citizens to some extent?

In the future, where will Rekognition go, we cannot make a judgment at the moment.

As an important application for AWS, in addition to selling to government agencies, this image analysis technology, including face recognition, will also be sold to relevant big data companies through AWS, Amazon on July 27In the 2018 financial report, cloud services posted net sales of $6.105 billion in the second quarter, up 49% from $4.1 billion in the same period last year; operating profit was $1.642 billion, compared to $916 million in the same period last year.

When faced with problems, the serious application of the reality, and the temptation of huge profits, how will Amazon finally deal with this?

Technology advancement, artificial intelligence brings convenience and speed to our lives. It is self-evident, but technology can be used as a tool. In the end, it is still in the hands of people. To put it plainly, AI is just a child. There are still many things to learn, and the key lies in how our human “parents” can educate and guide.

References

[1]https://hackernoon.com/a-quick-introduction-to-aws-rekognition-8257d4777198

[2]https://www.orlandoweekly.com/Blogs/archives/2019/07/18/orlando-cancels-amazon-rekognition-capping-15- Month-of-glitches-and-controversy

[3]https://gizmodo.com/amazons-face-recognition-tech-once-again-pegs-politicia-1837215790

[4]https://www.vice.com/en_us/article/ne8wa8/amazons-facial-recognition-misidentified-1-in-5-california- Lawmakers-as-criminals

[5]https://www.forbes.com/sites/monicamelton/2019/08/13/amazon-rekognition-falsely-matches-26-lawmakers- To-mugshots-as-california-bill-to-block-moves-forward/#4c323d0c7350

[6]https://www.dailymail.co.uk/news/article-5995149/Amazons-face-ID-tool-mismatched-28-members- Congress-mugshots-ACLU.html

[7]https://www.icrac.net/open-letter-to-amazon-against-police-and-government-use-of-rekognition/

[8]https://www.cnet.com/news/amazons-rekognition-falsely-matched-lawmakers-to-criminals-again-aclu-says/