Artificial intelligence (AI) is presumed to be the future technological advancement of human life with its near-limitless applications and highly effective methods. To name a few, its applications in the technology, healthcare, and automotive industries are well within development. Take healthcare, for example, where AI can be used to predict diseases, monitor critically ill patients, and even forecast diagnoses (What Is Machine Learning? | IBM, 2016). AI, although within its infancy, can have promising benefits for members of the Deaf and Hard of Hearing community. For instance, some students developed a Zoom-linked AI that can detect and translate American Sign Language (ASL) and automatically turn it into captions for individuals who don't know sign language. However, although its applications are promising and exciting, many populations, such the Deaf and Hard of Hearing community, may endure negatives from its implementation. Therefore, it is crucial that we recognize and address discrepancies within AI. Specifically, the implementation of machine learning (ML) may result in a reduction of Deaf culture and associated practices. That said, the overwhelming concepts of ML promise a more accessible and inclusive environment for Deaf and Hard of Hearing individuals.
AI is the concept of replicating aspects of human intelligence into technology. One of its subsets, ML, learns from examples by extracting relevant features (What Is Machine Learning? | IBM, 2016). For instance, let's say you wanted a machine to distinguish between photos of lions and tigers, you would upload numerous pictures of lions and tigers that are labelled with their respective names. The algorithm could then start to pick up the features that make a lion a lion and so forth. Then, when you upload an unlabelled picture of a lion, it can predict that it's a lion. While the concept seems so simple, its applications are ubiquitous (What Is Machine Learning? | IBM, 2016).
Deaf AI is a new tech startup that bridges the gap between ASL users and non-users by translating English words into ASL using AI (Deaf AI – Experience the Digital World the Right Way, 2022). The model uses voice detection from many partnered social platforms and directly turns it into a signed phrase. A demo of this is seen below. This platform serves as a more accessible alternative to real person interpreters. That said, the implementation of this project could lead to a reduced need of human interpreters, ultimately leaving many hearing, Deaf, and Hard of Hearing individuals jobless.
Description: In the demo video, the machine learning model distinguishes between signing and non-signing based on hand expressions, facial expressions, and body posture.
Alongside Deaf AI, several other startups have worked to assist in making communication and media consumption easier for the Deaf and Hard of Hearing community. For example, Robotica is a UK-based startup made by a team out of Norwich, UK, that brings sign language translations (facial expressions, hand gestures, and body movements) from hundreds of streaming services. It uses this data to train a model to make consuming media and news easier for Deaf and Hard of Hearing individuals (Robotica, 2022). Robotica’s AI algorithm is already trained in British Sign Language and is being introduced in ASL and Italian Sign Language (Robotica, 2022).
Another aspect of AI technology applications for Deaf and Hard of Hearing people is education. For example, Microsoft developed an AI-powered translator technology that utilises advanced AI that translates lecturers into high-quality captions with accurate punctuation (AI Technology Helps Students Who Are Deaf Learn, 2018). Microsoft is partnering with the National Technical Institute for the Deaf to deliver this technology (National Technical Institute for the Deaf | RIT, 2023). To add, it is already implemented in the Rochester Institute of Technology, which houses 1,500 Deaf and Hard of Hearing students, as observed in the image below (AI Technology Helps Students Who Are Deaf Learn, 2018).
That is not to say that AI is not free of limitations. AI, and in particular, machine learning, is well within its infancy. For one, there remain significant discrepancies in AI accuracies across the different models. Research studies are warranted to assess current AI models available for the Deaf and Hard of Hearing community and to determine the best one. AI also has several ethical, legal, and human interface limitations. For example, the technology is prone to hacking, which could lead to revealing sensitive data, such as patient data, if applied in healthcare. On top of that, certain services, such as healthcare, video surveillance, and automotive vehicles, use AI for high-risk tasks, such as predicting disease or detecting certain individuals. Added that there is a lack of legal responsibility, issues could arise if AI makes a mistake. That said, although AI's implementation for the Deaf and Hard of Hearing community is promising, its limitations must be addressed before moving forward. One way to do this is by equipping ML techniques with high quality privacy to prevent any means of hacking. Despite these limitations, AI holds much promise in providing a more accessible and effective environment for Deaf and Hard of Hearing individuals.
References
Comentarios