Artificial Intelligence (AI) to Diagnose and Treat Autism

Possibilities and Limits of AI for Autism

For several decades, researchers have explored the idea that artificial intelligence (AI) could be used to diagnose autism and help people on the autism spectrum to improve social, communication, and emotional skills. Diagnosis of autism through the use of AI is now a reality (though not the norm); AI-based therapies are in development and show promise. While some types of AI therapies (such as those that require the use of an interactive robot) are not yet available at a reasonable cost, AI-based apps are now downloadable for any smartphone user.

Defining Artificial Intelligence

The term AI is frequently applied, both correctly and incorrectly, to a wide range of programs and apps. It is usually used to distinguish "ordinary" programming from a type of programming that learns as it interacts. In theory, therefore, AI-based programs and apps are more human-like than ordinary algorithm-based programs and apps.

The journal Business Horizons defines AI as "a system’s ability to correctly interpret external data, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation." In other words, AI can respond to an individual person's particular abilities and challenges with specific outcomes in mind—just as a teacher or therapist might.

Artificial intelligence is not, at this point, anywhere near the level suggested by science fiction. In other words, no robot or program can use AI to pass the Turing Test, developed by the famous cryptographer Alan Turing. The Turing Test states that "if a person cannot distinguish whether he/she is talking to a human or a machine, the machine exhibits intelligent behavior."

Why and How AI Is Used to Diagnose and Treat Autism

It's not always easy to spot signs of autism, especially when the person in question is very bright and/or high functioning. That means it can take longer than it should to get a diagnosis—and a late diagnosis means a delay in getting therapies and services that should be available in a child's earliest years.

There are several reasons why delays occur. There is no single obvious sign of autism, and some signs of autism can also suggest other, unrelated disorders or personality differences. Evaluators may be uncertain whether a particular behavior is part of an autistic pattern or just a personal idiosyncrasy, and many evaluators and parents are unwilling to pin a label on a child until they're absolutely certain the label is correct.

According to the publication Spectrum News, a form of AI called "deep learning" is sometimes better able than human beings to spot relevant patterns. Deep learning is a type of machine learning that is actually based on artificial neural networks, and these types of programs may be a good way to provide evaluators with confirmation of a diagnosis or suggest the need for further evaluation.

There are a few companies pioneering methods to diagnose children with autism using AI and AI-like technology:

Behavior Imaging

Behavior Imaging, a Boise, Idaho company, uses a system called the Naturalistic Observation Diagnostic Assessment. This tool is an app which allows parents to upload videos of their children for observation. Initially, clinicians watched the videos to make remote diagnoses; more recently, however, the company has started training AI-like algorithms to observe and categorize behaviors. The algorithms would not diagnose the children but might point clinicians to specific behaviors that might otherwise have been missed.

Cogna

Another use of AI-aided diagnosis is an autism screening tool created by Cognoa in Palo Alto California. This tool is a mobile app that parents can use without the involvement of a trained evaluator; it reviews answers to multiple-choice questions as well as videos of the child.

So far, while there is interest in and some use of AI as a tool for supporting diagnosis, there is little support for the notion that AI alone can provide a reliable diagnosis of autism.

Robots to Treat Autism

People with autism are often overwhelmed by the demands of human interaction. Social expectations, sensory challenges, difficulty with expressive and reception speech, and attentional issues can all interfere with optimal outcomes. To circumvent this problem, a number of innovative groups have started exploring ways to use AI to teach and engage people on the spectrum.

One of the most intriguing (and expensive) approaches to using AI in therapy involves creating and training robots to interact with autistic children. Their purpose is to give autistic children practice with identifying facial expressions, interacting socially, and responding appropriately to social cues.

SoftBank Robotics

SoftBank Robotics NAO humanoid robots are about two feet tall and look like science-fiction-style androids. They are capable of expressing emotions by changing the color of their eyes, moving their arms, and changing the tone of their voice. Children with autism often respond more positively to NAO than to a human therapist, perhaps because NAO (and other robots for autistic children) have unlimited patience and are able to repeat the same cues in the same way over and over again without variation. Many children on the spectrum look forward to their time with and, in some cases, show NAO affection with hugs.

Massachusetts Institute of Technology

Researchers at MIT, wanting to take the interactive robot a step further, required a robot to integrate information about individual children using data from video, audio, and measurements of heart rate and skin sweat. Using this information, along with information about expected and appropriate behaviors, the robot can make sense of and respond to a child's behaviors.

Manatee

Manatee, a Denver startup specializing in AI apps for people with autism, is working with a company called Robauto to developing a robot called BiBli that can talk children through challenging interactions without judgment—at the child's own pace. Manatee co-founder and CEO Damayanti Dipayana recognizes both the benefits and limitations of a technology like BiBli: "I don’t think AI can provide all kinds of therapy, but it’s a scalable way to provide care for kids who wouldn’t get care," she tells Verywell. "And it's much more accessible emotionally, too. Many kids with autism or anxiety disorder find it easier to talk with the screen or the robot. In the long run, the information collected by a robot or app can be analyzed and shared with a therapist to provide a therapist with insight into what issues are challenging."

AI Apps for Autism

AI-based apps are less costly and easier to integrate into ordinary homes, schools, and therapists' offices than high-end robots. There are many autism apps on the market that support behavioral therapy and learning, but most are relatively simple logical tools for following a set of rules and earning points for doing so.

"The difference between AI and tech logic is that interaction may start with a standard response but then the model starts moving," says Dipayana. "The AI app uses a series of exercises to help the user calm down or respond appropriately and then, depending on the mood of the child, the model offers exercises and then learns how the child responds. Instead of coding with logic you give it a framework within which it can learn; ultimately it starts thinking more like a human being."

The Manatee app is one of the first AI apps offered as a simple, no-cost iPhone download. "The goals are written by clinical psychologists," says Dipayana. "It's recommended that kids do the activities with parents first. There is a step-by-step list that takes it from easy to more advanced skills; the app is intended to be supportive by offering guidance and help with lots of focus on parental involvement."

Limits of AI for Treating Autism

AI is a new tool for treating autism, and, so far, research is limited on its outcomes. AI-based robots and apps, while they do have the ability to support children as they learn, have a few shortcomings. For example:

  • While robots are, undoubtedly, extremely cool, they are very expensive to make and use.
  • Children who can use apps must be able to read and follow instructions. They must also be motivated to comply with a program that offers "virtual" rewards for a job well done. In other words, even semi-independent use of any app requires a level of functioning and motivation that is well above that of many children with autism.
  • Apps are intended to teach specific skills such as appropriate social communication, facial expression recognition, and eye contact. While some children are more willing to interact with a robot than with a human, it's not yet clear that those children will be able to transfer their interactive skills to human playmates.
  • Apps are not yet integrated into most typical settings. While some therapists and some schools are beginning to embrace the technology, there is a long way to go.
Was this page helpful?

Article Sources

  • Dipayano, Damayanti. “Interview with Damayanti Dipayana Co-Founder and CEO of Manatee .” 20 July 2019.

    Conversation with Damayanti Dipayana Co-founder and CEO of Manatee, a corporation dedicated to creating AI-based tools for treating autism.

  • Hall, Heidi. “Artificial Intelligence Thinks like People with Autism, Used to Develop Educational Tools.” Vanderbilt University, Vanderbilt University, 5 Sept. 1970, news.vanderbilt.edu/2017/09/05/artificial-intelligence-thinks-like-people-with-autism-being-used-to-develop-educational-tools/.

  • Ham, Becky, and MIT Media Lab. “Personalized ‘Deep Learning’ Equips Robots for Autism Therapy.” MIT News, 27 June 2018, news.mit.edu/2018/personalized-deep-learning-equips-robots-autism-therapy-0627.

  • “Can a Computer Diagnose Autism?” Spectrum, 16 July 2018, www.spectrumnews.org/features/deep-dive/can-computer-diagnose-autism/.

  • “Siri, Siri, in My Hand: Who's the Fairest in the Land? On the Interpretations, Illustrations, and Implications of Artificial Intelligence.” Business Horizons, Elsevier, 6 Nov. 2018, www.sciencedirect.com/science/article/pii/S0007681318301393?via=ihub.