Tag Archives: artificial intelligence

Robots Can Help Children with Autism Learn

About 1 in every  160 children globally has autism spectrum disorder. Most of them face developmental delays such as behavioral challenges and difficulties with social interaction. This makes learning new skills a serious challenge for them, especially in traditional schools.

It has been reported that socially assistive robots can help autistic children learn, but only if the robot can accurately interpret their behavior and react appropriately. In 2020, researchers at the University of Southern California developed a personalized learning robot called Kiwi for children with autism.

Kiwi, a personalized learning robot for autistic children. Source: kcet.org

Kiwi Teaching Math and Social Skills

Kiwi uses math games and an algorithm that monitors the child’s math performance to provide appropriate feedback and change the games’ level of difficulty accordingly.

While the content of the game focuses on math, the main purpose is to teach the kids fundamental social skills through their interactions with the robot, such as turn-taking (is it my turn or Kiwi’s turn to talk?) and eye contact (should I look at Kiwi when I’m talking?). Kiwi also uses data such as dialogue and eye contact to predict whether children are engaged in a given activity. If it detects that the child is not engaged, it tries to re-engage them for an extended period of time. When tested, Kiwi managed to reach a 90% accuracy in predicting the child’s engagement.

Collecting Data from a Realistic Environment

The study is based on the information collected after leaving Kiwi with 17 autistic children for a month in their homes. Participants regularly played the games on Kiwi’s attached tablet. The robot would then give personalized feedback through a reinforcement-learning algorithm. This algorithm enables Kiwi to elicit the best possible feedback by modifying its response based on each child’s reaction in the same way as the study’s lead author describes:

“If you think of a real learning environment, the teacher is going to learn things about the child, and the child will learn things from them. It’s a bidirectional process and that doesn’t happen with current robotic systems. This study aims to make robots smarter by understanding the child’s behavior and responding to it in real-time.”

The following video shows the robot, its interaction with an autistic child, and the researchers’ insights about it:

YouTube Preview Image

Source: NSF|YouTube

Surprising Results

Assessments were conducted for each participant before and after the month-long interventions. The results surpassed the researchers’ expectations of participants’ improvement. At the end of the study, 100% of the participants demonstrated improved math skills; 92% of them also improved in social skills.

Despite having promising results, such interventions are typically inaccessible to most people due to their high costs. The hope is that in the future, such socially assistive robots become affordable and turn into personalized therapeutic companions for all autistic children to improve their development.

I’ll leave you with the following short film telling the story of an autistic boy improving his social skills with the help of Kiwi:

YouTube Preview Image

Source: USCViterbi|YouTube

-Samin Shadravan

Protein Folding: Solved

Just as the turmoil of 2020 was coming to a wrap, a scientific breakthrough came about. On November 30th AlphaFold, coming out of Google’s DeepMind, claimed to have solved the protein folding problem using artificial intelligence.

The Problem

From making our DNA to getting rid of waste, proteins are like small machines that perform the majority of work done in cells. In fact, within our bodies there are an estimated 80,000 to 400,000 unique proteins each playing their own role. And, just like the way a building is built determines its use, a protein’s structure decides what tasks it performs. Yet, although it is easy to distinguish an apartment from an office, according to UCONN Health, it can take scientists between a few weeks to a few months to piece together what a protein looks like.

The Game Changer

This is where AlphaFold sneaks in. Although, as seen in the video above, the task was not easy, AlphaFold chose a different approach to this problem: artificial intelligence. 

Nowadays, the word artificial intelligence pops-up everywhere from self-driving cars to artificial voices, but what is most important is how it works and how it can be applied to the protein folding problem.

General scheme for developing an artificial intelligence model.

 

For the computer it all starts with data. As seen in the diagram above, once given data the computer looks for patterns between points. These patterns can then be used to make predictions on new data. Before in their final structure, proteins begin as a simple string of amino acids, or the building blocks of proteins. Given a dataset with the original string paired with the protein in its final form, the computer looks for patterns between the two. Using these patterns it can then predict what a protein might look like from just its string.

The Importance

Just one of the many protein folding predictions generated by AlphaFold’s model.

To the left you can see one prediction Alpha Fold’s model created. In comparison to the time it takes in the lab, this model is able to make a prediction in a mere half an hour with 90% accuracy according to their statement. In fact, it has already helped a biologist named Andrei Lupis with piecing together a protein his team has been stuck on for a decade. In an interview with Nature, Lupis even said: 

This is a game changer, this will change medicine. It will change research. It will change bioengineering. It will change everything.

With this new break-through, not only will scientists save time and money by not having to experimentally determine a protein’s structure, but research will accelerate at a pace never seen before. 

Beyond AlphaFold

While AlphaFold may be a hot-topic, beyond protein folding AI has also been used for a variety of tasks including interpreting MRI images or even predicting climate change. The applications seem to be limitless so make sure to keep an eye out, the next breakthrough could be coming up just around the corner!

Jessica Petrochuk

 

Protein Folding: Solved

Just as the turmoil of 2020 was coming to a wrap, a scientific breakthrough came about. On November 30th, AlphaFold, coming out of DeepMind, claimed to have solved the protein folding problem using artificial intelligence.

The Problem

Proteins perform the majority of work done in our cells from synthesizing DNA to getting rid of waste. Of course, the way a protein functions is largely dependent on its structure. This can include characterizations such as what parts of the protein are exposed versus tucked away. The proteomics field is dedicated to studying these, what is currently estimated to be, 80,000 to 400,000 proteins in our bodies and use two main strategies to determine their structure in the lab: X-ray crystallography and NMR. And yet, even in the midst of these complex protocols and high-tech machinery, a structure can take between a week to a few months to piece together according to UCONN Health.

The Game Changer

This is where AlphaFold sneaks into the picture. AlphaFold chose to take a different approach to this nominal problem: artificial intelligence.

Artificial intelligence has taken the world by storm and has improved the accuracy and efficiency of processes in almost every industry. From self-driving cars to artificial voices the possibilities are endless. 

General scheme for developing an artificial intelligence model.

Put very simply by the diagram above, artificial intelligence, more specifically machine learning, trains a computer to look for patterns within a given dataset. Once trained, this program can use the patterns it learned to make predictions of its own. In the case of AlphaFold, their model was trained off of amino acid sequences and their predetermined structures.

Just one of the many protein folding predictions generated by AlphaFold’s model.

In comparison to the time it takes in the lab, AlphaFold’s model was able to predict protein structure in a mere half an hour with an accuracy of 90% according to their statement. In fact, it has already helped an evolutionary biologist named Andrei Lupis with piecing together a protein his team has been stuck on for a decade. In an interview for nature, Lupis even said: 

“This is a game changer, this will change medicine. It will change research. It will change bioengineering. It will change everything”

Beyond AlphaFold

Of course, while AlphaFold may be a hot-topic, beyond protein folding, AI has also been used for a variety of tasks including interpreting MRI images, predicting climate change, or even sifting through astronomical data. The applications seem to be limitless so make sure to keep an eye out, the next breakthrough could be coming up just around the corner!

Jessica Petrochuk