AI: Intriguing or extremely threatening?

It has been called the “genie out of the bottle” and a sign that an intellectual revolution is underway. Artificial intelligence (AI) is the simulation of human intelligence processes by computer systems that have access to massive amounts of data.

Professors and other instructors are trying to cope with the use by students of OpenAI’s ChatGPT to write their essays, a new form of cheating. ChatGPT (Generative Pre-Training Transformer) is a computer-based natural language model that has been “trained” with huge amounts of textual data. It generates easily understandable text of a requested length in response to a user’s question and commands.

ChatGPT was one of the topics of a recent talk by Georgia "Gina" Tourassi, director of the National Center for Computational Sciences and the Oak Ridge Leadership Computing Facility at Oak Ridge National Laboratory to the Friends of ORNL group.

“When given a prompt, ChatGPT can write essays, jokes, software code and presentation outlines. It is truly impressive," she said. "A major concern in the scientific community is that when scientists submit a paper to a journal, how do we know if the authors truly wrote the paper or if they had assistance from ChatGPT?

“This particular wave of AI, which is intriguing to me, is extremely threatening to some people. Now, scientists are bombarded with so much information that it has become more and more challenging to cross domain boundaries, which is usually where most innovation happens. We could throw various scientific concepts into ChatGPT and see if it generates what could be a creative insight. There have been papers published about that possible approach where ChatGPT was listed as a co-author," she said.

How AI began: Going back to the '50s

Tourassi summarized the history of AI, which started in the 1950s. She had been involved in the innovative use of machine learning, a form of AI, for diagnosing breast cancers from medical images.

“We trained computer systems on a large number of breast tumor images, and they often outperformed the ability of the typical radiologist to recognize early breast cancer,” she said.

But, she added, “The true revolution in artificial intelligence happened exactly 10 years ago, around 2012, with the development of deep learning.”

Deep learning is a type of machine learning based on an artificial “neural network” – a computer system that mimics the human brain’s behavior and “learns” from large amounts of data through multiple layers of processing that extract progressively higher-level features from the data. Neural networks consist of algorithms that recognize patterns in datasets of text, images or sounds.

She showed recent examples of deep learning’s ability to generate images from text, such as whimsical depictions of “a corgi playing a flame throwing trumpet.”

Tourassi speculated that new AI features may lead to fewer jobs for graphic designers, just as AI assistants in Google Maps and GPS navigator systems have created competition for taxi drivers in big cities by enabling new industries such as Uber and Lyft. She added that AI applications will likely threaten different job sectors.

While deep learning has been highly successful in some areas – for example, it has been a game changer in biology’s core problem of determining protein structures (protein folding) – it can be hacked, and it has had catastrophic failures, she said.

As director of the Oak Ridge Leadership Computing Facility (OLCF), which has almost 1,700 users from around the world, Tourassi noted that OLCF’s Frontier, the world’s fastest supercomputer, is a leader in the use of AI for solving scientific problems.

ORNL achievements that use AI

“AI is pervasive in ORNL mission applications,” she said, noting that it is being used for research to advance materials design, transportation technology, medicine, climate science, electrical grid technologies, cybersecurity and nonproliferation, as well as to make instruments smarter and facility operations more efficient. She gave examples of recent ORNL achievements that use AI.

AI is using data from atomic models to predict the properties of materials and design new ones. In advanced manufacturing research using 3D printing, ORNL engineers have developed an AI approach for determining the positions and shapes of defects formed to help guide improvements in the additive manufacturing process.

ORNL technology developments may hasten the use of autonomous vehicles - self-driving, long-haul trucks already carry cargo on California and Texas highways - and better AI-assisted ways to make driving safer.

According to Tourassi, ORNL researchers earned an R&D 100 Award for an AI technology called Gremlin that identifies weaknesses in neural networks, possibly exposing safety issues such as a glitch in an AI model that prevents a type of autonomous vehicle from avoiding all obstacles.

ORNL’s first AI technology to be commercially licensed may help its licensee, General Motors, improve its vehicle technology and design. It may also accelerate advanced driver assistance technology by determining how cars with onboard cameras and other sensors use AI to quickly and accurately perceive and navigate safely through their surroundings.

In health care, Tourassi mentioned ORNL innovations using AI that will help identify potential drug molecules that could block a virus like COVID-19 from binding to a receptor of a healthy human cell, keeping the virus from entering the cell, producing copies of itself and making the host sick.

She cited ORNL’s TwoFold, a new AI methodology that promises to accelerate drug discovery for novel viruses like COVID-19. TwoFold’s unprecedented speed and accuracy in predicting the behavior of proposed three-dimensional protein structures to aid in the design of new anti-viral molecules could help make supercomputers like Frontier an integral tool in drug discovery, thereby greatly accelerating the process. The study earned the team a 2022 finalist nomination for the Association for Computer Machinery’s Gordon Bell Special Prize.

Another ORNL team that was a Gordon Bell Prize finalist nominee employed an algorithm on two OLCF supercomputers to scan hundreds of thousands of biomedical concepts from millions of scientific publications and the National Library of Medicine’s PubMed database in search of potential connections among symptoms, diseases, conditions and treatments. The goal: streamline the exploration for promising leads to generate new scientific hypotheses and accelerate drug discovery – what Tourassi calls “computer-assisted serendipity.”

She noted that this work is important because of the possibility of a future pandemic. She added that AI can also be useful in determining whether potential drugs found to be effective are toxic.

She has pioneered the use of AI with huge amounts of reported health data to help the National Cancer Institute (NCI) determine the incidence and distribution of various cancers. As the principal investigator on the cancer epidemiology project for NCI, Tourassi said the project, which long had been relying on two-year-old data, has found that certain AI models used with U.S. cancer registries have enabled almost real-time cancer surveillance on data that are only 30 days old.

“Now we can use AI to capture and evaluate quickly the broader effectiveness of new diagnostic systems or therapeutics for cancer being deployed in different communities,” she asserted.

'Digital twin' could help people improve their health

Just as there are well-known AI assistants such as Alexa and Siri, Tourassi said she dreams that all Americans will someday have on their smartphones or laptops a personal “digital twin” that incorporates data on their early health, changes in their health conditions and updated information from new medical studies that are relevant to their heath conditions, including any disease they are genetically determined to get. She suggested you might be able to “play” with your digital twin by asking it how your life trajectory will be affected by potential changes in your diet and exercise routine or relocation to an area that has known pollutants.

AI is already being used in the health care field for scheduling patient visits and interpreting medical images, she said, adding that it will be needed even more in the United States with an aging population that is growing and likely to develop geriatric diseases while the number of doctors and other health care workers shrink.

This article originally appeared on Oakridger: AI: Intriguing or extremely threatening?