In the contemporary scenario of privacy and the internet not going hand in hand, every other day presents cons
of AI as much as its pros. In the month of February, the government had issued rules called, The Information Technology Rules, 2021, to curb online violent content and find its originator. It was not possible earlier because of the end-to-end encryption of several platforms and the immunity provided to them under section 79 of the Information Technology Act, 2000. The issue of privacy is again in question, and in some aspects, is haunting the users.
Recently, a parliamentary standing committee headed by MP Shashi Tharoor was informed by Google’s representatives that the tech giant listens to some recordings between its users and the AI. Everyone is aware about the “Ok/Hey Google” feature of the Google assistant, where simply saying one of these keywords triggers the AI. It acts like a simple alarm for the software to start listening to the user for command. Now, the problem is a two-stage process of the system.
First, as responded to BJP MP Nishikant Dubey’s specific question, Google records the audio of the conversation between the users and its assistant. The important part is that the software does that without even activating the command “Hey/Ok Google”. In technical terms, it is known as a false accept. It means that the assistant has misunderstood any outside noise as the keyword and has recorded the conversation. So, the bottom line is that your voice assistant can listen to you without your permission whenever it is within a range to do so.
Also Read : Pick WhatsApp or Privacy
Second, these recordings are used for companies’ own use. One of the primary uses is to make its assistant more efficient by studying random recordings taken from the first stage and getting them analysed by different language experts. According to Google, 0.2% of these “audio snippets” are given to language experts. These language experts study several factors associated with the way of speaking of people with different dialects and accents of the same language. This puts the private conversations of the users in the hands of companies’ contractors. An infamous incident of a Dutch contractor leaking one of these recordings is a reason for the fear of its users. The explanation for the use of such recordings and safety of the data was stated by Google’s Product Manager, David Monsees in a blog post by the company itself.
The demand from the users and the authorities to not collect the recordings beyond the commands is not irrational. In order to make the software learn different ways of speaking different words, other ways can also be adapted. For example, there are companies which collect samples of recordings from sources, such as- call recordings of customer care for software related complaints, etc. Not every voice assistant records conversations to use it for such purposes. Siri, the voice assistant of Apple, only transfers that information to the cloud which involves direct instruction such as weather or match score. These voice assistants also do not wait for the command to come from the user and record your voice until that moment. Even if it gets recorded, it does not leave the device.
Also Read : Artificial Intelligence: Critical to Address Bias Concerns
In a recent experiment conducted by a team of experts, the authenticity of this allegation was tested. An android phone with the google assistant feature was used. The device was kept in front of two experts and they started a conversation in which they used a particular word. This particular word was an electronic gadget. After a conversation of a minute in which that word was used around 7-8 times. The google search on the device suggested the gadget with only the first word on the search bar. It indicated that even without triggering the voice assistant, it was aware of what was going on around it.
In the deposition conducted by the parliamentary panel, the Google team said that sensitive information is not heard in the recordings. It leaves the users and the authority in ambiguity, that what does the company terms as “sensitive”. Defining a piece of information as “sensitive” is subjective. It can be sensitive for one individual but not for the other. So, this takes us back to square one. The problem still persists as any information related to the defence of the nation could also be leaked if no changes are made.
This makes it a serious question to be asked in the courts of law of the country. Even though the company claims that an option is always available to its users to delete such “false accept” in the application itself. It is almost impossible to predict when the voice assistant has recorded something accidentally and then choose to remove it. Therefore, companies which are using software with voice recording of users need to be highly cautious of the data they are storing. The authorities on the other hand are expected to put a restraint on such practices for the welfare of millions of citizens.
