How machine learning can support people with disabilities
Carlos Pereira from Livox speaks to Project Literacy
As part of the Project Literacy Lab initiative we have been working with some of the world’s best social entrepreneurs who are dedicated to closing the global literacy gap by 2030. In order to showcase some of the great work the Lab entrepreneurs are doing, we are kicking off a series of short interviews with our Lab alumni to inspire our supporters with stories of innovation, resilience and a deep passion for learning. For the first one, we invited Lab 2016 alumnus Carlos Pereira from Livox. Carlos in one of the brightest entrepreneurs in the world and has won many prizes for his work with Livox, including the Tech for Global Good Awards last November from the Tech Museum of Innovation.
What is Livox?
Livox is an alternative communication software that enables non-verbal people with disabilities to communicate and to learn. What makes it unique is its intelligent algorithms that makes it adjust itself to a wide range of disabilities as well as the usage of Machine Learning and Natural Language Processing to enable people with disabilities to communicate up to 20 times faster.
Why did you decide to found a start-up?
I’m a father of a 10-year-old girl named Clara that has Cerebral Palsy due to a medical mistake during my wife's labor. My company is one of many initiatives that we started in order to give my daughter a better life. Quickly, thousands of other people with disabilities started to use our solution and now we are changing the lives of many people with disabilities.
What is the most valuable lesson you've learned as an entrepreneur?
Actually I never think too much about this. I just had to do it because my daughter needed it. However if there's a key lesson that I learned is that it takes a long time to be a success! :)
Entrepreneurship, social innovation, and having a "start-up mindset" are all hot topics at the moment. Do you have any advice for aspiring entrepreneurs (or intrapreneurs working in large corporations)?
Yes. I do believe that social innovation must be stimulated. There are so many good opportunities waiting for a passionate entrepreneur to make them happen. Deep commitment is the key component for any entrepreneur.
What's next for you and Livox?
For decades, alternative communication devices have just emulated the same paradygm of paper cards grouped in sets of similar meaning. However, due to the complexity of communication, any device can become extremely complex with dozens of items on a screen. This is a major issue for people with motor or cognitive impairments since they have a hard time to finally find what they want to communicate. We call this, the Reciprocity Gap. It takes a long time for people with disabilities to either initiate a conversation or to answer questions.
We solved this issue with a new technology that uses Machine Learning and Natural Language Processing. We received a grant and technical support from Google to improve this technology. In order to allow people with disabilities to initiate conversations faster, we use Machine Learning to bring the right options for conversation depending on the time and location.
In order to allow people with disabilities to answer questions faster, we use Natural Language Processing. This means that you can actually talk to the person with a disability, Livox listens to the conversation and then it brings possible answers related to the spoken context. Livox just starts to listen the conversation after listening to a keyword, similar to what Apple and Google did with "Hey Siri" or "Ok Google" keywords. In Livox, we recommend the keyword to be the name of the person with disabilities. So you can something like: "John, how many spoons of sugar you want in your coffee?", Livox understands that the answer for this question is a number and then it brings number so "John" can say: "I want 3 spoons of sugar".
Since we use Machine Learning, there's no need to memorize any questions. Currently we trained our models with over 1.4 million questions and our accuracy rate (this rate means how many times Livox brings the correct content after a question) is around 95%. In our studies we can see that with Machine Learning and Natural Language Processing, non-verbal people with disabilities can communicate up to 20 times faster than using other alternative communication devices.
We plan to continue pushing forward the limits of technology in order to improve the lives of people with disabilities and we are deeply committed to this!