Are there confidentiality issues with automated transcription?
February 26, 2020 2:53 PM   Subscribe

I'm interviewing someone and helping him write a memoir for his family. I'm a slow typist and transcribing the interviews is taking me a long time. I'd like to try Temi, which is AI transcription, but are there any issues with confidentiality?

The guy I'm working with is a very, very, very minor public person, but public enough that confidentiality is critical. He's entrusted a lot of intimate stuff to me, and I don't want to compromise that in any way. Mostly I think I should just suck it up and do it myself. Thoughts?
posted by gigondas to Technology (11 answers total) 2 users marked this as a favorite
 
Many if not most AI transcription services have humans involved(generally contractors in places like the Philippines or India or via services like Mechanical Turk.) Sometimes it’s just QA, other times there is no AI involved at all, it’s just regular old I with a bit of shiny tech wrapper.

These kinds of startups are also often terrible at security, and will leave entire datasets wide open on the public web by poorly configuring s3 buckets.

So this random guy on the internet says eh, it’ll probably be fine, but I would trust my secrets to some AI transcription service.
posted by rockindata at 3:14 PM on February 26 [3 favorites]


Would not trust my secrets!

In case you are curious, here is one of many articles about fake AI.
posted by rockindata at 3:22 PM on February 26 [2 favorites]


It is very bad practice to pass along confidential data to third party companies for processing without having both an agreement with the data owner that allows for it, and a contract with the third party which lays out the terms of the confidentiality.

Temi, for what it's worth, says in their FAQ that they will sign an NDA if asked.
posted by Jairus at 3:23 PM on February 26 [1 favorite]


Would he be identifiable based on what he says? If it's really sensitive, I wouldn't risk it. But I do use AI transcription all the time and I'm not overly worried about it. I use Otter (free up to 600 minutes a month)- here's their privacy policy.
posted by pinochiette at 3:38 PM on February 26


AI - fake or otherwise - isn't the real issue here.

You have a relationship with your client that, presumably, also involves a commonly understood definition and expectation of privacy. Therefore, you need to look at the terms of any third-party service you're thinking of using from the perspective of your client: based on your understanding of the services you're considering outsourcing to, can you guarantee that you will be fulfilling your obligations to your client with regards privacy?

If you're willing to take on the work to ensure that a third party has the same expectations and guarantees of privacy as your client has with you, and you are willing to take that on as a liability in your relationship with your client, then go ahead and use them.

If not, then you are taking a risk, and it's basically down to you how comfortable you are with that. But, given that "[you] don't want to compromise that [confidentiality] in any way", then I wouldn't.
posted by parm at 3:51 PM on February 26 [1 favorite]


This depends on what you've promised your client in terms of confidentiality, and what terms the company can offer. (I'd be a lot more comfortable with a transcription service that specifically addresses security for IRBs, HIPAA-related data, etc. than something as generic as Temi's website, for example.) But really - this depends on your client. Ask him, and don't make him any promises about security that you can't back up via a contract with the third party. (A real contract, not website terms of service that could change without any remedy for you.) Maybe he's fine with it, if only for some of the less sensitive stuff, but there's no one-size-fits-all answer here.
posted by Stacey at 3:57 PM on February 26


Based on these replies, which pretty much corroborate my own instincts, I'll just slog through and do it myself. Thanks.
posted by gigondas at 4:02 PM on February 26 [1 favorite]


Why not use the dictation tool available in MS word or Google docs? I use that when I want to take notes, and the just go through and spend some time making corrections myself. Saves a lot of time and energy
posted by shaademaan at 4:07 PM on February 26


I use Temi for my doctoral research. It is common and well-used among my professors and peers. I sought and obtained explicit approval from the institutional review board to use the service. A one hour interview, about 15-17 single spaced pages, takes ~8 minutes to transcribe.
posted by maya at 5:34 PM on February 26 [1 favorite]


If you trust Google (or are committing this manuscript to a machine where Google already has access anyway), Google Live Transcribe works astoundingly well for American accents (not so well for mine), and happens in real time so you know for sure humans are not involved: https://play.google.com/store/apps/details?id=com.google.audio.hearing.visualization.accessibility.scribe&hl=en_AU

OtterAI is another AI transcription program that I don't believe has humans faking it (as you can see the transcript immediately) and its privacy policy seems really good to me. Although of course, there's nothing to stop it NOT doing what it promises in the privacy policy, so I wouldn't trust it for anything that really is super sensitive.
posted by lollusc at 5:38 PM on February 26 [3 favorites]


All o can say is I work in a highly confidential field and we use Speakwrite without problem.
posted by purenitrous at 9:01 PM on February 26 [1 favorite]


« Older In search of specific Mexico City lodging...   |   Collaborative Todo List Apps/Programs Newer »

You are not logged in, either login or create an account to post comments