Your gut has no understanding of probability and the inevitability of coincidences. I am well aware of the issues involved in speech recognition on mobile devices - I have written software to do exactly that. Google are not listening to you. Some eastern european hackers may be, but they are not too bothered about the legality of such things.
I understand that it's feasible but it just doesn't make any sense as a strategy. As part of my training we did modules on marketing and demographics and in my job I have a lot of experience of market segmentation, branding and understanding how to get complicated messages across to individuals and groups. Here is why I think this is nonsense.
A short history of targeted advertising:Companies have been targeting advertising since long before Google and the internet. You place your ads in publications or in the breaks in TV programmes that fit the target demographic. The National Readership Survey developed the A, B, C1/C2, D & E demographics to classify readers and that has been used to target advertising for years. The Acorn classification is even more detailed and is broken down into 6 categories, 18 groups and 62 individual types:
https://upload.wikimedia.org/wikipedia/en/7/79/CACI's_Acorn_Structure_%28Uploaded_June_2014%29.jpg.pdfLoyalty cards were developed in the 1990s for loyalty (obviously) but also to allow stores to aggregate customer data. It creates individual profiles showing what beer people drink, the food they buy etc and that allows stores to focus their offer in particular stores based on what people actually buy and they can also send out individual offers to certain types of customer.
With the advent of online shopping and using cookies and purchase history it's possible to create an individual advertising profile for everyone. I am acutely aware of it because I buy things for my personal use but also for specific projects and jobs.
My recent profile on Amazon assumes I want to buy a load of computer cables, batteries, a four hole punch, archive boxes, football books and videos about crime and punishment (from my interest in the Steven Avery case). That's a mixture of work and personal searches and purchases.
My YouTube profile recommends football (including Arsenal Fan TV because it's so funny when they lose), TYT, Stewart Lee, British Pathé etc...
All of which make sense based on my
choices - either searches, purchases or browsing history. And the choice factor is critical.
Because on a practical level, using a random selection of words picked up by my computer or phone. I rarely speak when I'm at my computer but will often have either music, or more usually, a film or tv programme playing on the iPad. If Google is listening to the microphone on my computer it will be hearing Tommy DeVito saying: "I'm funny how?... funny like a clown?... or "Go and get your shine box..." or Jimmy McGill talking about cucumber water and vibrating chairs at the nail salon.
And when I'm talking it will only occasionally have any relevance to the choices I may make when buying things. And even when I do how does the speech recognition know what was relevant? I've been discussing my summer holiday and the options included walking holidays in Scotland, a cottage by the beach in Cornwall, a house in France up in the hills, or possibly somewhere near Nice, or Normandy?... maybe Spain... Italy... or we might go to see my brother in Denmark...
And that example is why it makes no sense to me and would represent a huge backwards step in terms of targeted advertising. The whole emphasis has been to narrow down from very broad A, B, C classifications to more and more focused advertising based on actual choices. Using chance eavesdropping of partial conversations that are as likely to be about speculation and aspiration as actual choices will create a profile that is less exact than one produced through searches and actual purchases.
There is absolutely no point basing advertising on what people say they would like. Your computer eavesdropping while you're watching Top Gear will assume you're in the market for a Bugatti while your internet searches and purchase history will make it patently clear that you can actually afford a used Peugeot 208.
It would be simple to set up a blinded experiment to see whether this is real or not. But people seem to prefer to rely on anecdote (about the most unreliable type of 'evidence' their is) to get their knickers in a twist and moan than get off their arses to prove it one way or another.