Computer Software Decodes Emotions -- AND Detects Autism!

Post Reply New Topic

http://digg.com/d31QJOm

YES!! !!
I've been thinking about this for years. Now, after all the work I've done on studying math [ed:sic] just for this purpose, somebody goes and beats me to it. Now, if they had something like this for real-time conversations, extended to video processing as well, we'd all basically be cured. Is this really possible? No more autistic social isolation?!?!


We now have ways to predict ASD with statistical accuracy?!?! The hits just keep on coming! (J.C.! If only this existed when I was a kid!)
Treatment must logically come next. This has to happen now.


In other breaking news, scientists have released new data that suggests that the sky is blue and that water is wet.
It's rather ironic that NTs need to be reminded about this so often. I guess that something so amazing that it shatters NTs social dominance has to be met with confusion.

But seriously, this has just made my day! I'm not sure how to respond to this right now, but I know for certain what I'm doing over the summer: reviewing everything eXaudios publishes!

Last edited by AdmiralCrunch on 07 May 2010, 10:50 am, edited 1 time in total.

interesting.

the program is neutral, but society can still use the results of it for good or for ill. So really, Im now worried about society being a ***** and scanning everyone, and just doing s**t to those who the machine deems autistic.

I had that reaction at first, too. But there's no reason to think that there's no way to counteract this. If the algorithm can be reviewed (not necessarily easy, since it's likely proprietary), then it's possible to be beaten. In fact... this may be the single best thing to ever happen to autism.

It's the same situation with "Deep Blue" and other computer chess programs. In the 90's, they seemed unbeatable, but after a while, mathematically-oriented chess experts realized that the computers were only good at certain types of game play. Today, well-trained chess masters can easily beat computers far more powerful than Deep Blue by forcing the game away from any points where the computers have the advantage, setting the computer theorists back to the drawing board.

If the same type of analysis on emotional decomposition software can be made by mathematicians, especially us autistic ones, it could be reverse-engineered. Then, it just might be possible to train ASs to sound indistinguishable from NTs. Of course, we'd have to do it using the software itself; imagine a language lab where you use audio feedback to adopt the accent of a foreign language, only you'd be doing it to alter the emotional resonance of your voice. Once trained, ASs would basically be indistinguishable from NTs in virtually every meaningful way. Social interaction would become much easier. Throw in some video feedback training, like I really want to create, and you have a way to train yourself to be just like an NT.

Consider that: a functional treatment for autism. I shudder in anticipation!

Against what independent standard has this program be compared to validate the program?

ruveyn

The man they interviewed, Dr. Bonneh, looks like he had a background in CS, so it's likely Bayesian inferencing. Also from the way the article was presented, that seems likely.

The software's algorithm is probably using frequency analysis and neural networks to detect specific artifacts as features. Testing a target group versus a control would give a usable percentage output. The 85% success rate they give sounds reasonable, if low. This function will probably be optimized in the future to give a better result.

The good news is that, as I was saying before, there's no reason to expect that this detection algorithm would be impervious to reverse-engineering. To my knowledge, speech therapy has been used to change voiced artifacts like this in the past -- people who get cochlear implants need to re-learn how to speak. If the above is all true, I think it's likely possible to use speech therapy to train ASers to defeat the algorithm. A nice result, of that and rather the whole point, would be significantly improved social interaction potential.
But, then again, maybe my line of reasoning is flawed. Regardless, I'm going to be investing my time into researching this.

This is my main concern.

WTF? Cured how? Misreading or failing to read social cues is only one aspect of AS, and to my mind not the most negative. I've been around the block enough times now that I now easily recognize pretty much all the social cues I missed when I was younger, but that doesn't magically give me the mental resources to respond to them appropriately, or in a timely manner.

And I'd lay down money that that machine can't identify my speech patterns as Autistic.

I mean, it's all very interesting, but I wouldn't rejoice that it's going to eliminate Autism as a handicap. Psychological Stress Evaluators have been touted for decades as lie detectors and as I've read, have a much higher success rate than the traditional polygraph, yet they haven't replaced the polygraph in commercial use, and neither of them is admissible as evidence in a courtroom (in the US). So you can probably still get in on the second and third generations of this technology and tweak it to your heart's content.

That's a good point. I had always assumed that, like me, everyone with AS had communication as problem number one. I guess I'm tipping my hand here.


I don't know if it's my mathematical brain, but I see the two as corollaries: analysis and construction are basically the same thing. Since we live in a universe with entropy, it's always easier to analyze than construct, but enough analysis will inevitably lead to construction. There's only a finite number of things to say in any situation that are funny, witty, etc., and these all tend to be based on major themes. Look at enough situations and patterns start to emerge. Consider: there's only so many bases for jokes that eventually they all start sounding the same.

The article included this gem: the system then suggests various tactics to a customer service representative, depending on the needs of the company.
Imagine creating software that could let you know, based on probabilistic modeling, what the mostly effective topic or riposte would be in any situation. I posit that given enough computing power, this is a real possibility.


Reasonable. But I'd lay down money that not everyone with AS has your linguistic skills. I know I certainly don't.


Maybe I am reading too much into this. I've just been imagining something like this since I was a little bullied kid in grade school: the computational machine that was so powerful it could tell you how to beat anyone in verbal sparring, or at the least let you hold your own with everyone else. To me this is worth investing some research time.


Yes, these are still based on subjective human experiences and so can always be beaten by the human mind -- and it's not a lie if really do you believe it. But the authoritativeness of competitive criminal trial is not what I'm personally yearning for here; just the ability to keep up conversationally with NTs. That's got to be a lot less demanding.


You're right: the future for this is a lot more meaningful than the present. Hell, I'm just happy to know that there's something for me to do with my life from now on -- if not a career, at least as a hobby.

This could've been implemented years ago. Most people on the spectrum speak in a relative monotone, and that's relatively easy for machines to differentiate.

Kind of reminds me of the device used in 'Blade Runner' to detect the replicants.

Focusing on oneself and failing to understand other peoples needs/emotions is another aspect of Aspergers

Cool software though.