That's interesting, and in some ways, not surprising... Medical diagnostics must be one of the most complex mental activities we can imagine...

A few years ago, I interviewed a computational linguist who was using Watson for a research project on automated speech recognition (ASR). The aim was to be able to do automated speech-to-text (STT) in "acoustically chaotic" environments. When there's a lot of noise (background and foreground) it creates two problems: the obvious one, which is that the background noise obscures the speaker's voice, and the less obvious one, which is that when the environment is noisy, the speaker tends to yell and their enunciation and prosody goes wonky. So STT AI, which is trained on people sitting calming talking into a mic in an otherwise empty room, can't handle the non-standard speech patterns that an acoustically chaotic environment creates.

One of the key applications of this technology is battleground medics. The person I interviewed said that on a battleground, someone is injured and they maybe get airlifted out. The person who gives them an evaluation before that will usually write on the person -- literally on the person, like on their skin with a magic marker -- what treatment they need or other info that a doctor can use to treat them. Well, as you can imagine, this really limits the kind of info that can be shared with the receiving physician.

So that application of the STT AI for acoustically chaotic environments was to enable an on-the-ground medic or any personnel to be able to make a voice recording that gets translated into text which can then printed out, or sent ahead of the wounded person, so the receiving doctors can be better prepared, and the person gets better treatment. As you can imagine, this is a technology that could really make a life-or-death difference.

I think that was maybe 4 years ago, well maybe 5 now.... I should look him up and see where his work is at, and whether he's still using Watson or not. ASR and STT AI are not the same as diagnostic AR, so it could that he's still using Watson...

Having said that, AFAIK, ASR and STT tech has yet to make the huge leaps that people have been waiting for for years -- like being able to do STT of a multiparty conversation of all same-gender, same-age speakers where everyone speaks English with a non-American, non-British accent.

I have some linguistic data that was recorded in an acoustically chaotic environment -- at an outdoor cafe in Japan, in the audio recording, there are car noises, motorcycles go by, the cafe has music playing in the background.... It's maybe two hours of data, I have been sitting on it because manually transcribing it will take forever. I keep waiting for the big ASR tech jump forward...

Sorry for rambling on!

Interesting. I was approached by IBM in 2012 looking to figure out financial services applications for Watson. They led the pitch with how they were reinventing health care and could do the same across any data-intensive industries.


Thanks for posting.