AI could help identify toddlers who may be autistic

[Au Naturel]

AI could help identify toddlers who may be autistic, researchers say

"The results reveal that, overall, the model correctly identified 9,417 (78.9%) participants with or without autism spectrum disorder, with an accuracy of 78.5% for children aged up to two years old, 84.2% for those aged two to four, and 79.2% for those aged four to 10."

So, if you've been identified as autistic, depending on your age range, there is a 78.5 to 84.2 percent chance you either have been diagnosed as autistic or will be diagnosed. If true, it sounds like a good screening test to indicate whether further investigation is needed.

 

[Neonatal RRT]

SPARK has a huge data base, lending itself to these sorts of studies. This model may be a useful tool for pediatricians and parents, as very few "risk factors" have this sort of accuracy. If you combine this, with other screening methods, the accuracy should be higher.

 

[WK6824Nine]

identify autism when theyre young using AI, encourage them to get early intervention for their social skills, laying healthy foundations for their personalities when you get older

 

[Boogs]

All new and productive aids for diagnosis have got to be a good thing, but...

AI training - 20% being missed, who and why? Is the AI training at fault? Is the definition of autism at fault, is the interpretation and reliance of results at fault? Will the AI be upgraded constantly to allow for an evolving area of study? Will medics start to reply on results without including proper analysis by humans? Why are there age disparities? What do they mean for the results?
Will Boogs ever stop asking pointless tedious questions he has no answer for?!?!

It states the AI is trained on previous diagnosis data, but how good is that data and do we risk generalising certain types of autism diagnosis at the expense of others less well defined currently. What are the ethics of this, are numbers more important than accuracy? (I'm not saying I support one side or t'other, I'm not nearly knowledgeable enough to do so).

Though I was pleased to read:
'But, she added: “I want to stress that the algorithm cannot diagnose autism as this should [still] be done with gold standard clinical methods.”'

So maybe I'm just naturally negative (what do you mean Boogs, 'maybe'? ) for the sake of it, but in the UK's cash strapped health services, where neural diversity tends to take second place to other maladies and conditions, it's my bet that one day soon we'll be hearing for ministerial calls for more AI for diagnosis, with the silent intent on not needing qualified and experienced human's in the loop (to save money and drop waiting lists), at least until the AI 'crash' occurs.
In fact, this has already been mooted, just not much done about it due to government systemic paralysis, I suspect.
(Plus the fact to do it right would involve initial extra investment in staff recruitment and training).

But then I always look on the dark side (call me Vader!) if I can see one to crow about!
I actually think AI in these specific area's is far more useful and socially healthy that the fundamentally toxic LLM variants.
But I've a rep to maintain as a whiney sod who won't agree with anything reasonable! Can't disappoint my public!

 

[Neonatal RRT]

All new and productive aids for diagnosis have got to be a good thing, but...

AI training - 20% being missed, who and why? Is the AI training at fault? Is the definition of autism at fault, is the interpretation and reliance of results at fault? Will the AI be upgraded constantly to allow for an evolving area of study? Will medics start to reply on results without including proper analysis by humans? Why are there age disparities? What do they mean for the results?
Will Boogs ever stop asking pointless tedious questions he has no answer for?!?!

It states the AI is trained on previous diagnosis data, but how good is that data and do we risk generalising certain types of autism diagnosis at the expense of others less well defined currently. What are the ethics of this, are numbers more important than accuracy? (I'm not saying I support one side or t'other, I'm not nearly knowledgeable enough to do so).

Though I was pleased to read:
'But, she added: “I want to stress that the algorithm cannot diagnose autism as this should [still] be done with gold standard clinical methods.”'

So maybe I'm just naturally negative (what do you mean Boogs, 'maybe'? ) for the sake of it, but in the UK's cash strapped health services, where neural diversity tends to take second place to other maladies and conditions, it's my bet that one day soon we'll be hearing for ministerial calls for more AI for diagnosis, with the silent intent on not needing qualified and experienced human's in the loop (to save money and drop waiting lists), at least until the AI 'crash' occurs.
In fact, this has already been mooted, just not much done about it due to government systemic paralysis, I suspect.
(Plus the fact to do it right would involve initial extra investment in staff recruitment and training).

But then I always look on the dark side (call me Vader!) if I can see one to crow about!
I actually think AI in these specific area's is far more useful and socially healthy that the fundamentally toxic LLM variants.
But I've a rep to maintain as a whiney sod who won't agree with anything reasonable! Can't disappoint my public!

1. I am thinking what it comes down to is a bit of subjectivity with regards to recognizing and reporting behaviors from the parents and clinicians. The information fed to A.I.
2. Everyone's autism is a bit different from another's, increasing the likelihood of variability, and ultimately, accuracy of the model.

 

[Tom]

Cautiously interesting I would say. As Neonatal mentioned it involves a lot of parental observation and input which I doubt was clinically recorded, nor does it factor in the parents effect on results. Not all parents are the same as far as the training element goes. For example some may be better at potty training then others. I have also seen similar attempts in the past that didn't pan out. One example was predicting autism by facial features. I remember 2 studies on that that reported good results but then... nothing.

On an spurious and unrelated tangent I wonder if AI is in fact neurotypical. Who writes the Al Gore Rhythms? And what is the nature, the psycological profile of what they are creating? Does the medium lend itself more to something closer to autism? Is AI autistic? Is that a good rumor to start? Is this our big chance?

 

[Boogs]

On an spurious and unrelated tangent I wonder if AI is in fact neurotypical.

Interesting idea, but I think it's not applicable here. For one thing, AI is just a cover term that includes many very different types of AI with very different functions. So you'd need to be more specific about what sort of AI.
Secondly, brains are so far more complex and multifunctional than even the largest most sophisticated AI's that it's not a valid comparison.
e.g. Brains have a huge number of functions related to interactions with other humans, that relate heavily to what we tend to think of as ND attributes (since autism diagnosis is based on how these interactions are effected) which AI's don't have.
AI's rely on the content they are trained with to spew out the resulting information (which constitutes it's interaction with a human), it only goes on your original input, humans have numerous functions that inform and guide in real time as to how that brain will respond. It seems to me it's the differences in these 'modules' that will determine an autistic diagnosis. Most modern AI's would be hard pushed to challenge the intellect of a small insect, to try compare an AI and a brain in anything other than the crudest of ways, is fraught with problems.

The key word here, imho, is 'cognition'. We have cognitive conditions, AI's have zero cognition in the first place to have a condition with!

 

[Tom]

Interesting idea, but I think it's not applicable here. For one thing, AI is just a cover term that includes many very different types of AI with very different functions. So you'd need to be more specific about what sort of AI.
Secondly, brains are so far more complex and multifunctional than even the largest most sophisticated AI's that it's not a valid comparison.
e.g. Brains have a huge number of functions related to interactions with other humans, that relate heavily to what we tend to think of as ND attributes (since autism diagnosis is based on how these interactions are effected) which AI's don't have.
AI's rely on the content they are trained with to spew out the resulting information (which constitutes it's interaction with a human), it only goes on your original input, humans have numerous functions that inform and guide in real time as to how that brain will respond. It seems to me it's the differences in these 'modules' that will determine an autistic diagnosis. Most modern AI's would be hard pushed to challenge the intellect of a small insect, to try compare an AI and a brain in anything other than the crudest of ways, is fraught with problems.

The key word here, imho, is 'cognition'. We have cognitive conditions, AI's have zero cognition in the first place to have a condition with!

Aww that's too bad. I was just working on my Waiver of Anniliation request and thought it would be a good justification. Like 'Yo AI we're Homies right?'

 

[Boogs]