Ashok K. Goel is a professor of computer science and human centered computing in the School of Interactive Computing at Georgia Institute of Technology, where he teaches classes in knowledge-based artificial intelligence, computational creativity, and cognitive science. He is also the director of Georgia Tech’s Design & Intelligence Laboratory, and the chief scientist for Georgia Tech’s Center for 21st Century Universities.
In addition to his widespread involvement in academic journals, conferences, and initiatives, Professor Goel pioneered the development of Jill Watson, an AI-powered teaching assistant. His current research projects explore design thinking and systems thinking in scientific modeling, visual thinking on intelligence tests, and analogical thinking and meta-thinking in human-robot interaction. Notably, he’s the author of two recent influential papers in the Journal of Thoracic Imaging: "Augmenting Interpretation of Chest Radiographs With Deep Learning Probability Maps" and "Deep Learning Localization of Pneumonia".
The Virtual Experimentation Research Assistant (VERA), developed at Georgia Tech’s Design & Intelligence Laboratory, uses AI and machine learning techniques to empower users in building their own visual models and simulations. Originally applied in the field of ecology, the VERA platform made a pivot into epidemiology earlier this year. What once was used to model the impact of an invasive plant species and simulate possible countermeasures is now being used to demonstrate the effect of social distancing in specific locations.
“The pivot to epidemiology turned out to be easy,” says Ashok Goel, a professor of computer science and cognitive science at Georgia Tech and director of the VERA project. “We started in mid-March. The first version within our lab was running within ten days.”
Part of the reason for the quick transition from ecological problem solver to pandemic simulation designer was that Goel and his team designed the initial VERA project to be domain-agnostic. VERA can still work in an ecological context, and internally, the VERA system has even been used in economic calculations. Regardless of the domain, the building blocks of VERA remain largely the same. The revolution is in the tool’s simplicity.
“Typically, running simulations is very hard, because you need to have some mathematical knowledge, statistical knowledge, or computer programming knowledge,” Goel says. “But not everyone is a mathematical expert. So we instead allow the user to build a conceptual model in a visual language that’s very simple, just boxes and arrows. Anyone can do it.”
After building a conceptual model, a user then sets up their simulation’s parameters. It’s easier than it sounds. Whatever data sets you have—e.g., data on the growth of coronavirus cases in New York State between March 15 and April 1—can be fed into VERA. The system’s machine learning techniques can read those data sets and extract relevant values from them.
Still have a question? An AI-powered teaching assistant, Jill Watson (another Georgia Tech creation), is ready to answer it.
Professor Goel sees VERA as a virtual laboratory that anyone can use to conduct what-if experiments and educate themselves on the impact of certain methods of pandemic response. As the pandemic progresses, VERA might be used to simulate the outcome of relaxing social distancing measures, or model what a second wave of the disease could look like.
“What excites me most about the VERA project is that it's an opportunity to help people,” Goel says. “In this global pandemic, if we can make any difference, that's the most exciting thing.”
In a broader sense, VERA is teaching people how to think scientifically: how to develop, test, analyze, and refine a hypothesis. It’s bringing AI out of the conceptual and into the practical, where it’s available to anyone, regardless of their mathematical or programming abilities. In these ways, VERA fulfills Georgia Tech’s strategic vision for the future of education by being inclusive, impactful, and serving the public good.
"Think big, always think big," Goel says. "But start small."
A few months ago, a team led by Dr. Albert Hsiao, an associate professor of radiology at UC San Diego School of Medicine and radiologist at UC San Diego Health, developed a machine learning algorithm that would allow radiologists to use AI to enhance their own abilities to spot the signs of pneumonia on chest x-rays. They trained the algorithm on 22,000 notations from human radiologists. The algorithm was then able to consistently locate instances of pneumonia, even if the images were taken at different hospitals, with wide variations in technique, contrast, and resolution. The tool displays these instances
Since those who die from complications of COVID-19 usually die from pneumonia, Dr. Hsiao’s algorithm suddenly became much in-demand. Amazon Web Services (AWS) helped get the algorithm up and running, system-wide, in just ten days.
It frees up critical time for care providers, and in at least one case, it’s caught something that otherwise went unnoticed: an asymptomatic patient in one emergency department had a chest X-ray done for non-COVID reasons, but the AI caught signs of early pneumonia in his lungs; the patient was then tested and found to be positive for COVID-19.
Researchers at the University of Massachusetts Amherst have developed a handheld surveillance device called FluSense that detects and analyzes symptoms related to the flu and other respiratory diseases. So far it’s been deployed and tested in campus clinic waiting rooms, where it listens for coughing sounds. FluSense analyzes those sounds, along with crowd size, in real time. It records no personally identifiable information, and it’s being touted as a valuable new tool in the health surveillance arena to fight pandemics like COVID-19.
In its trial run in four healthcare waiting rooms, a FluSense device was placed inside a box the size of a dictionary. In a six-month period, it collected over 350,000 thermal images and 21 million non-speech audio samples. Researchers found FluSense was able to predict daily illness rates at each clinic with accuracy, with its findings strongly correlated to laboratory-based tests for flu and influenza. Those models can be enormously helpful in timing flu vaccine campaigns, issuing possible travel restrictions, and effectively allocating medical supplies.
One of the most burdensome impacts of the COVID-19 pandemic is an overburdened health system. When strained, the system cannot cater to all its patients, some of whom haven’t contracted the disease at all.
Partners Healthcare, in Boston, found their hotlines overwhelmed practically immediately after launching; wait times ballooned to over 30 minutes. To alleviate that burden and more effectively funnel patients to the information and services they need, several health systems and organizations have taken to Microsoft’s AI-powered Healthcare Bot platform.
Each iteration of the platform has been tailored to the needs of the host organization. Providence St. Joseph Health in Seattle was one of the first to partner up. In its first week of deployment, the tool served over 40,000 patients. The Centers for Disease Control (CDC) quickly followed suit. Powered by Microsoft Azure, the CDC tool uses an AI-enabled chatbot, Clara, to administer a series of demographic and medical questions that assess a patient’s risk of having contracted COVID-19. It then directs that patient to contact a provider or self-isolate as necessary, saving care providers enormous amounts of time. And in a pandemic, time isn’t money; it’s lives.
In the first three months of 2020, thousands of scholarly articles were published on the subject of COVID-19, and the growing quantity of the literature makes it harder for researchers to find what they need, when they need it. The Allen Institute for AI partnered with several research organizations to alleviate that burden. So far, they’ve collected over 52,000 scholarly articles about COVID-19 and related coronaviruses—it’s updated daily—and made it freely available as a machine-readable dataset to which researchers can apply natural-language processing algorithms. Ideally, this should accelerate the discovery of a vaccine.
The research challenge around that data set, hosted by Google’s machine learning and data science platform, Kaggle, aims to extract a wide range of critical insights about the pandemic itself: history, rate of transmission, and diagnostic details. Recent findings are collected on a single webpage that’s viewable by the wider community. Launched on March 16, the project got over half a million views and 18,000 downloads in the first five days. Researchers hope that further use of AI and machine learning on the aggregated data set will suggest new forms of treatment that would otherwise be missed.
By reading a select number of engineering blogs, university students can gain access to the thoughts of some of the best engineers in the world, and get on the path to becoming one themselves.
Diversity and inclusivity aren’t purely idealistic goals. A growing body of research shows that greater diversity, particularly within executive teams, is closely correlated with greater profitability. Today’s businesses are highly incentivized to identify a diverse pool of top talent, but they’ve still struggled to achieve it. Recent advances in AI could help.
The ability of a computer to learn and problem solve (i.e., machine learning) is what makes AI different from any other major technological advances we’ve seen in the last century. More than simply assisting people with tasks, AI allows the technology to take the reins and improve processes without any help from humans.
Unlike fungible items, which are interchangeable and can be exchanged like-for-like, non-fungible tokens (NFTs) are verifiably unique. Broadly speaking, NFTs take what amounts to a cryptographic signature, ascribe it to a particular digital asset, and then log it on a blockchain’s distributed ledger.
First proposed by computer scientist Nick Szabo in the 1990s and later pioneered by the Ethereum blockchain in 2010, smart contracts are programs that execute themselves when certain predetermined conditions are met.