Tech behemoth OpenAI has touted its artificial intelligence-powered transcription tool Whisper as having near “human level robustness and accuracy.”
But Whisper has a major flaw: It is prone to making up chunks of text or even entire sentences, according to interviews with more than a dozen software engineers, developers and academic researchers. Those experts said some of the invented text — known in the industry as hallucinations — can include racial commentary, violent rhetoric and even imagined medical treatments.
Experts said that such fabrications are problematic because Whisper is being used in a slew of industries worldwide to translate and transcribe interviews, generate text in popular consumer technologies and create subtitles for videos.
More concerning, they said, is a rush by medical centers to utilize Whisper-based tools to transcribe patients’ consultations with doctors, despite OpenAI’ s warnings that the tool should not be used in “high-risk domains.”
Why is generative AI even needed for audio transcription? We’ve had decent voice recognition tools for years even on cheap consumer grade stuff.
We also used to have decent web searches!
Because with normal algorithms you have someone to blame.
AI is a trick to hide when you steer the results the way you want.
Whisper really is a lot better when it works, and it’s free. The problem is that it refuses to produce gibberish or give up when it doesn’t work. You’ll always need an editor.
The toaster oven I just invented works much better than a traditional one. It reheats French fries perfectly, you can dehydrate in it, makes succulent roasted chicken, and about 2.5% of the time it burns down your house. You’ll always need to keep an eye on it to make sure that doesn’t happen. Remember though, much better than a traditional one.
You need an editor for traditional transcription tools too :) and it’s A LOT more work. They don’t even do punctuation or names.
No, we really haven’t had on-device voice recognition that meets any definition of “decent”. Anything reasonable phones out to “the cloud” for decent voice recognition.
So? I’d rather have my software talk to a server than be downright wrong just so another business can climb onto the AI bandwagon.
You can’t do that with personal information like the ones doctors needs transcribed. It has to be local.
Reality is more nuanced than this. You can absolutely be HIPAA compliant while using “cloud” servers as long as they are sufficiently isolated and secured. The requirements are definitely insufficient to protect your data from a Motivated State Actor™ but they are good enough to keep your data away from an abusive family member or crazy ex. I have worked on systems that handle patient data as well as other systems with restrictions I can’t discuss and I can assure you patient data is much easier to move around and handle compared to state secrets.
Edit: funny story, I just got back from a doctor appointment where they asked me to sign a consent form for recording and transcription of the visit by a computer system. It’s definitely happening, in practice.
Because AI!
Some examples
In this example, the speaker said, “as the um, the, her father dies not too long after he remarried….” while the program transcribes that as " It’s fine. It’s just too sensitive to tell. She does die at 65….”
In this example, the speaker said, “and after she got the telephone he began to pray” while the program transcribes that as “I feel like I’m going to fall. I feel like I’m going to fall, I feel like I’m going to fall….”
Wow, that’s bad. I thought it would be more of a “confusing a sentence for a similar sounding one” type thing but from the above and the article it’s just generating semi-believable text and sticking them into the transcriptions.
It’s actually extremely good at figuring out confusing text. It gets weird when the audio quality is bad.
I use it for generating subs for obscure movies.
No one is good with bad audio. My wife did some transcription work for a little while, it can be pretty painful, especially for doctors, and all the medical terms.
This one was wild:
In an example they uncovered, a speaker said, “He, the boy, was going to, I’m not sure exactly, take the umbrella.”
But the transcription software added: “He took a big piece of a cross, a teeny, small piece … I’m sure he didn’t have a terror knife so he killed a number of people.”
From picking up and object to mass murder lmao. Not even close!
But it gets the spirit right
/s
Sounds less like transcribing word for word, and more like attempting to summarize and parse meaning on the fly. AIS have notoriously little grasp on reasoning and logic, so it’s interesting how the output holds up in a court of law.
LLMs in medicine. What could go wrong?
Whisper isn’t a large language model.
It’s a speech to text (STT) model.
Which has the same concept as the LLM under the hood, hasn’t it?
As someone who uses Whisper fairly often, it’s obvious that they’ve trained off of a bunch of YouTube videos.
Most of the time it’s very accurate, but there have definitely been a few times in long transcription sessions where it will randomly hallucinate that someone is saying “Don’t forget to like and subscribe!” When nothing was said anywhere near that.
That’s hilarious. I just love how AI is basically like a 6-year-old who weaves his favorite new expressions into everything without fully understanding what they mean.
Microsoft teams has some automatic transcript capabilities that are so hilariously bad, it’s hard to believe Microsoft released it.
I guess they use the same service.
Ml is currently best used as a tool for helping a real human do work. It has to be managed and governed and checked after. It’s still capable of amplifying the amount of work we can do, But it’s kind of like having to have an architect look after engineering work, you can’t just pass the buck and pretend that it’s going to be fine.
maybe that’s just the ai’s internal monologue leaking through?
Oh word?
