Inputs by Dhruva KC
If you use a voice-activated home assistant like Alexa, there is a feature that allows you to listen back to all the recordings on your device, and one that allows you to delete these voice recordings.
But these voice recordings are also used for one more purpose – to train Alexa to listen and recognise commands better.
How are user voice recordings used to train AI like Alexa?
A human user is sent anonymised recordings. They are required to listen to the recording and then enter the right interpretation. This interpretation is fed to Alexa’s learning system so that it can recognise voice better next time.
For a long time, Alexa users did not know that Amazon was sharing the voice recordings with human listeners. When it was found out, Amazon put this in the Terms of Service.
Today, it is common knowledge that human listeners listen in on conversations.
However, the user has the option to listen back to everything recorded on their device. They also have the right to delete these recordings.
What did Amazon do in this particular case?
When a user deletes a recording, it is assumed that the recording is also deleted from Amazon’s servers.
Except, it was not. The recordings were retained by Amazon even after the user specifically deleted them.
Why is it more wrong to store children’s voice than the voice of adults?
For two reasons:
A. Children are not of age of consent. This means that they have not reached the age at which they can give their consent to being a part of this recording. For this reason, no one has the right to take this decision on their behalf. All their data must be kept off public records until they attain the age of consent or the age at which they can legally enter a contract.
B. Children’s voices can be used to make deepfakes. These deepfakes can be used in crimes against other children.
Since children are vulnerable (can be hurt easily), every society is duty bound to protect them even more.
Therefore, children’s data is always treated with more care than any other data.
The Fine
Amazon has agreed to pay $25 million to the US government for collecting voices and data of children under 13 before the next week.
It has decided to change its features on handling the data and geolocation of children. This was under a settlement (an official agreement intended to resolve a dispute or conflict) in Washington, US, approved by a federal judge.
This is because COPPA’s (Children’s Online Privacy Protection Act Rule) rule states that websites should only save the information of a child as long as it is reasonably necessary (ideal) to fulfil the purpose of the information collected. Amazon violated this rule by keeping data for years. The settlement was a continuation of a 2019 complaint. It was filed in May this year by the DOJ (Department of Justice) on behalf of the FTC (Federal Trade Commission).
Other directives in the judgement
A directive is an order.
After paying $25 million to the government, Amazon is also required to perform the following actions:
- Stop using geolocation, children’s voice information and data should be deleted when the user requests to. They should not use it for any improvement or development of devices.
- Delete any inactive accounts of children users under 13 (unless the parents do not want them to).
- Notify users about the settlement.
- Notify users about the deletion and apologize for saving their data.
- Create a program that keeps a child’s geolocation safe.