AI and Speech-Communication in Autism

AI and speech in autistics.
Hari Srinivasan

These are thoughts I've been mulling over for a while now and have been slowly writing up. 

Overriding consideration: Cost, Accessibility, Portable, User-friendly and helps our communication & functioning in society. 

How can AI help? Some thoughts.

1. Speech Communication Therapy

• AI powered Speech therapy can be applicable and tailored for all ages across lifespan. Human speech therapists tend to focus on just the younger kids and regularly turn older teens/ adult autistic clients away saying it's too late, too old to learn to speak. But I believe there is room for improvement across the lifespan. For example my speech is starting to emerge in adulthood (while it did not in childhood) and I hope to get to conversational level and more in a decade. Autism is not called dev delay without a reason.
• Language in existing therapy apps & speech therapy tend to be targeted towards little kids (like asking for toys) not realizing speech/communication is an issue across lifespan. So language needs to be relevant to the current stage of life. Irrelevant to learn to ask for a baby toy when in 20s-30s-40s-50s-60s or have kiddie images on app for adult. 
• AI powered speech therapy apps can provide “age-appropriate” personalized interactive and engaging therapy sessions to help improve speech, language and communication skills. For instance in adulthood, the act of speaking itself (in a speech driven society) could be highly reinforcing and there there is lesser need for reinforcement in the form of flashing graphics/sounds or the high-pitched “good job” (which actually hurts the ears) that therapists tend to overuse in the copious childhood therapy that most of us have to go through.
• Accessibility: Can be done from home/classroom and thus time-saving in terms of travel and where the sensory environment can be better controlled for the autistic who can become easily overwhelmed. Any trip to a therapist's office involves sitting in the waiting room for instance which is often over decorated and can be an assault to the sensory system. Speech therapy rooms also tend to be small and rather uncomfortable cubicles/rooms.
• Cost & Frequency of therapy: AI tech costs will decrease over time while human-therapist costs will increase over time and not be fully covered by insurance etc. Lower cost means higher frequency, necessary for true impact of speech learning.
• AI powered speech systems can provide real time feedback to autistics, helping correct their speech and articulation.
• use of ML so that speech exercise/output is tailored to culture and foreign language pronunciations of names etc.
• AI powered language systems can help autistics understand and use language more effectively by providing cultural/context-sensitive prompts, suggestions and word completions.
• Love to see an AI/ML powered speech bot who can have real time conversations with adult me (non-human means non-judgmental and will not infantilize). No need for noisy visuals or sounds for reinforcement - by adulthood, the act of talking itself has high implicit motivation (we live in a speech-heavy society). Just a face that is talking to me, showing appropriate human emotions as it speaks. The AI bot would provide real time feedback to my speaking, helping correct speech and articulation and lead in prompts to promote more responses from me (just like a human therapist). Also adjust the prompting level depending on how my mood/body is on that day, if we could attach it to a watch that reads body signals and stress levels. And it would use ML to learn the topics I’m passionate about and help develop conversation skills around those to begin with. And add in some humor so it does not feel like “work.” And as we advance we could progress from 1 to several bots to simulate group conversations. Adult autistics experience a lot of loneliness due to limited social interactions, so it can be like a friend(s) you hang with and have fun conversations with. 

2. BCI: I'm just waiting on breakthrough as this will be a life changer for folks like me

• Non-invasive, baseball cap that you can put on and take off (don't want to stand out in a crowd with UFO headgear).
• Portable & Durable - can be used anywhere.
• Inexpensive so it's accessible to a population that often relies on already inadequate state support.
• Should take into consideration body movements. (Not expect us to stay still).
• Lets us speak our ”intentional thoughts,” only and not every thought.

3. Social Interactions

• Enhancing speech recognition: Can help you be better understood by others by improving the accuracy of speech recognition.
• Some autistics have a harder time reading body language and emotions. AI powered speech recognition systems can help autistic understand and identify emotions in human speech which can help improve their social interactions and communication
• Speech summarization: By providing a summary of key info & concepts, AI can help autistics better understand and retain info from spoken language.

4. Tech can/must become more User-friendly (whether letter/words or icon/image based)

• Current AAC tech and apps is not very user-friendly or intuitive especially given that a lot of the AAC users have a lot of sensory dysregulation, fine/gross motor-coordination issues, somatosensory body mapping issues and movement-disorder-like movements to contend with. AI/ML could enable personalized user interfaces optimized for individuals making it easier to use.
• Some of the icon based apps are confusing for users with screen after screen of icons that have to be navigated before finding the appropriate icon/image, by which time they have lost the motivation or become distracted. If communication is perceived as "too much work" or confusing, it becomes demotivating for user. 
• We can also use an update of those stick figures used in language icons, because frankly some of them are scary - the kind that come in bad dreams. Not the best motivation for communication. 
• use ML for “phrases-prediction” rather than simply “word prediction.”

5. Voice Output.

• Fix glitches in current text-speech output like in cultural pronunciation of names (have to type “Hurry” instead of Hari for my name) and reading out abbreviations (eg: superior colliculus (SC) will be read out as “South Carolina” and "Ca2+" will be “California 2 plus “and ADHD sounds like someone is choking). Adds to burden of motor effort by user to add spaces between abbreviations etc (in a pop that already has body regulation issues/ movement issues).
• Improve speech synthesis & clarity, by improving naturalness and intelligibility of synthesized speech. Natural-sounding voices with the right pitch, pauses, intonation and emotions of a talking human vs the emotionless flat robotic voices (which are demotivating for autistic)
• Natural sounding voices based on any/limited ‘vocalization’ by individual (or family member) so that the voice is individualized to person, instead of every male sounding like Ryan-voice or Will-voice. Current tech is there but very expensive, (subscription based models tend to run up costs year after year) and therefore inaccessible to a population already living in poverty.

6. Increase overall independence and accessibility

• access and interact better with services and technology.
• AI-powered Assistive Tech (AT) would help individuals complete tasks that would be otherwise difficult /impossible to do such as controlling smart home devices or controlling a computer.
  

It boils down to improving quality of life. AI can make services and technologies more accessible for folks with speech issues such as providing more personalized and user-friendly interfaces and providing greater independence and autonomy in our communication and access to technology and services. And it can help us improve our communication skills which will help functioning in society.

--- 

Concerns

I worry that AI apps will be seen as another profit-making venture on the backs of disabled people. After all, the autism space is open-season for exploitation with a never ending supply of generation after generation desperate for solutions. Autism therapy is one the biggest profit-making industries too, and starts from the time of your diagnosis. All with zero accountability in terms of actual benefit for autistics. The tech/method get the claim if the autistic progress, but if the autistic does not progress, it was because the autistic was not going to progress anyway. 

Pushing any one methodology/technology seems to profit (money/fame) the maker the most, so I'm always vary about anything that claims to be the only solution or the best solution. 

So I would encourage everyone to use multiple things or more, so you are not dependent on just one thing and tailor everything to suit and generalize to other methods/tech, just so that you are not at the mercy of one tech/method. It allows for better adaptation when something about it changes.  What happens when they want to charge more or the software changes. Remember, Autism loves sameness and its not good to get stuck on one thing. 

And don't be afraid to discard and move on if it does not work out - which is of course, easier said that done if you have spent a ton of money on some therapy/tech. 

I understand it costs money to develop solutions but if that solution is expensive it becomes meaningless as then it is inaccessible to most. 

And I worry about the humanity of it all. For instance, high yearly subscription costs almost feel like someone is holding your tongue hostage with the kidnapper allowing you to communicate only at their will and mercy and if you paid up a hefty ransom yearly.