Remember when we were horrified at ”Das Leben der Anderen” and then all installed an Alexa at our home? Yeah…
Phelim Kine, a China Watcher with Politico, just published a piece on how transcription app started asking questions about a conversation he had with an Uyghur that fled a surveillance state. Out of the frying pan…
Well, don’t take my word how bad of a practice that is, I have a vested interest in the competition. What you can take my word for is how complicated it is to design a product with privacy embedded to it.
First of all modern A.I. -products are driven by data, and lots of those, especially computer vision and natural language tasks typically revel in data. Just borrowing that from your customer is eh, well eh, very eh, convenient. *insert it’s free real estate meme*
Not having your data automatically means that you’ll have to
buy rent a proprietary model. If you want your service to be any good, that is. It also means that you can’t sell that data to other A.I. companies, or put it on balance sheet and pitch it to V.C.’s.
Other types of data, such as your behavior and identity (Plain old profiling, I don’t think that your profile is being harvested from audio, even though the rumors are plenty), are even more liquid and can be sold via ad networks.
For stenos we have implemented a quaint old business model that asks money from the user in exchange for value. All jokes aside: That has become a hard sell.
Another vulnerability for the well meaning startup is the cloud based data storage. Not storing any data of users is a pretty good way of avoiding leaking it. I know first hand that keeping user data is quite a responsibility, and peace of mind is a great thing. However, not having the data is also a hurdle, for instance when offering backups for your clients, or syncing that data between devices.
A third piece of this puzzle is the customer interaction. This dawned quite lately to me in this particular venture, even though this is not my first rodeo. Not having any customer data means not being able to e-mail your customers with reminders or questions, which can be vital to your learning curve and retention metrics.
To me personally it has been something of a hobby to minimize data requirements. I have been involved in quite a few projects involving data protection at RWS (the Dutch Directorate-General for Public Works and Water Management). Most of these projects involved existing systems. Designing a system bottom up was very liberating for once. Most startups however will have to tackle these challenges an other way, as long as we don’t change the rules to game.