A.I.
Tennessee institutions partner to develop dependable AI for national security applications
DOE/OAK RIDGE NATIONAL LABORATORY
IMAGE:
AT THE TENNESSEE VALLEY CORRIDOR SUMMIT 2024 IN NASHVILLE, TENN., ON WEDNESDAY, VANDERBILT UNIVERSITY AND OAK RIDGE NATIONAL LABORATORY ANNOUNCED A PARTNERSHIP TO DEVELOP TRAINING, TESTING AND EVALUATION METHODS THAT WILL ACCELERATE THE DEPARTMENT OF DEFENSE’S ADOPTION OF AI-BASED SYSTEMS IN OPERATIONAL ENVIRONMENTS.
view moreCREDIT: ORNL, U.S. DEPT. OF ENERGY
Artificial intelligence is rapidly becoming one of the most important assets in global competition, including AI-assisted autonomy and decision-making in battlefield applications. However, today’s AI models are vulnerable to novel cyberattacks and could be exploited by adversaries. Moreover, the models are not sufficiently robust and dependable to orchestrate and execute inherently human-centric, mission-critical decisions.
“AI and autonomous vehicles have great potential to let our military operate in contested environments without having to needlessly put our brave men and women in harm’s way — as long as we can trust the AI,” said U.S. Rep. Chuck Fleischmann. “ORNL and Vanderbilt University have the infrastructure and expertise to develop solutions that will give national security leaders the confidence that these AI systems are secure, reliable and dependable.”
Under a new partnership announced during the Tennessee Valley Corridor 2024 National Summit in Nashville this week, Vanderbilt and ORNL will build on complementary research and development capabilities and create science-based AI assurance methods to:
- Ensure AI-enabled systems deployed for national security missions are able to function in the most challenging and contested environments.
- Test and evaluate the resilience and performance of AI tools at large scales in mission-relevant environments.
- Provide decision-makers with the confidence to rapidly adopt and deploy AI-enabled technologies to maintain U.S. competitive advantage.
Vanderbilt’s basic and applied research in the science and engineering of learning-enabled cyber-physical systems, particularly through the renowned Vanderbilt Institute for Software Integrated Systems, provides a foundation for AI assurance research.
“We are excited to partner with Oak Ridge National Laboratory to ensure AI-enabled programs are safe, accurate and reliable at a time when it has never been more imperative to do so,” said Vanderbilt Chancellor Daniel Diermeier. “This radical collaboration among our best researchers and one of the nation’s premier national laboratories will address these crucial challenges head-on. We look forward to the great work we will do together.”
Building on expertise in high-performance computing, data sciences and national security sciences, ORNL recently established the Center for Artificial Intelligence Security Research, or CAISER, to address emerging AI threats. CAISER leads AI security research and AI evaluation at scale, capable of training and testing the largest AI models.
“With ORNL’s unique expertise and capabilities in computing and AI security, we can train, test, analyze and harden AI models using massive datasets,” said ORNL Director Stephen Streiffer. “Working in close cooperation with Vanderbilt, I look forward to advancing the Defense Department’s deployment of AI-based systems for national defense.”
The partnership will initially focus on enabling the U.S. Air Force to fully utilize autonomous vehicles, such as the AI-enabled X-62A VISTA that recently took Air Force Secretary Frank Kendall for a flight featuring simulated threats and combat maneuvers without human intervention. Together, Vanderbilt and ORNL will provide evidence-based assurance that enables Air Force systems to meet DoD’s requirements for Continuous Authorization to Operate in vital national security roles.
“The growth in AI applications is breathtaking, most notably in the commercial marketplace but increasingly in the national defense space as well. While all users of AI are concerned about security and trust of these systems, none is more concerned than the DoD, which is actively developing processes to ensure their appropriate use,” stated Mark Linderman, chief scientist at Air Force Research Laboratory’s Information Directorate. “This partnership will advance the science to enable the U.S. Air Force to confidently field autonomous vehicles, such as the AI-enabled X-62A VISTA, improve situation awareness, and accelerate human decision making.”
Autonomous vehicles operating in a truly independent fashion could be a game-changer for the U.S. military.
“Stewarding our national security and military is one of my greatest responsibilities as a Senator,” said U.S. Sen. Marsha Blackburn. “Tennessee is leading the way in developing the advanced technologies that will ensure our nation’s global leadership and protect the lives of our brave service members.”
The collaborative new research program at Vanderbilt and ORNL continues Tennessee’s tradition of helping the U.S. maintain global leadership.
“Tennessee is once again leading the way to keep Americans safe. This exciting partnership will leverage two world-class institutions and employ their renowned expertise and resources to make our military stronger and more effective,” said U.S. Sen. Bill Hagerty. “Technological dominance is a key pillar of national security, and this partnership will ensure that the Department of Defense can utilize this developing technology in a secure, robust, continuous and dependable fashion.”
About Vanderbilt University
Founded in 1873 as an institution that would “contribute to strengthening the ties that should exist between all sections of our common country,” Vanderbilt University is globally renowned for its transformative education and pathbreaking research. The university’s 10 schools reside on a parklike campus set in the heart of Nashville, Tennessee, contributing to a collaborative culture that empowers leaders of tomorrow and prizes free expression, open inquiry and civil discourse.
Top-ranked in both academics and financial aid, Vanderbilt offers an immersive residential undergraduate experience, with programs in the liberal arts and sciences, engineering, music, education and human development. The university also is home to nationally and internationally recognized graduate schools of law, education, business, medicine, nursing and divinity, and offers robust graduate-degree programs across a range of academic disciplines. Vanderbilt’s prominent alumni base includes Nobel Prize winners, members of Congress, governors, ambassadors, judges, admirals, CEOs, university presidents, physicians, attorneys, and professional sports figures.
Vanderbilt and the affiliated nonprofit Vanderbilt University Medical Center frequently engage in interdisciplinary collaborations to drive positive change across society at large. The two entities recently reached a combined total of more than $1 billion in external research funding in a single year. This landmark achievement reflects the university’s deep commitment to expanding the global impact of its innovation and research as it increases opportunities for faculty, students and staff to pursue bold new ideas and discoveries.
Oak Ridge National Laboratory is managed by UT-Battelle for the Department of Energy’s Office of Science, the single largest supporter of basic research in the physical sciences in the United States. The Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit energy.gov/science.
Tennessee institutions partner to develop dependable AI for national security applications
Moe Khaleel, center, associate laboratory director for national security sciences at ORNL, joined Padma Raghavan, left, vice provost for research and innovation at Vanderbilt University, and Ryan Luley, right, with the Information Warfare Division of the U.S. Air Force Research Laboratory, on Wednesday for a panel discussion titled, "Dependable AI for National Security" during the Tennessee Valley Corridor National Summit 2024 in Nashville, Tenn.
CREDIT
ORNL, U.S. Dept. of Energy
AI-controlled stations can charge electric cars at a personal price
As more and more people drive electric cars, congestion and queues can occur when many people need to charge at the same time. A new study from Chalmers University of Technology in Sweden shows how AI-controlled charging stations, through smart algorithms, can offer electric vehicle users personalised prices, and thus minimise both price and waiting time for customers. But the researchers point to the importance of taking the ethical issues seriously, as there is a risk that the artificial intelligence exploits information from motorists.
Today's commercial charging infrastructure can be a jungle. The market is dynamic and complex with a variety of subscriptions and free competition between providers. At some fast charging stations, congestion and long queues may even occur. In a new study, researchers at Chalmers have created a mathematical model to investigate how fast charging stations controlled by artificial intelligence, AI, can help by offering electric car drivers personalised prices, which the drivers can choose to accept or refuse. The AI uses algorithms that can adjust prices based on individual factors, such as battery level and the car's geographic location.
“The electric car drivers can choose to share information with the charging station providers and receive a personal price proposal from a smart charging station. In our study, we could show how rational and self-serving drivers react by only accepting offers that are beneficial to themselves. This leads to both price and waiting times being minimized”, says Balázs Kulcsár, professor at the department of electrical engineering at Chalmers.
In the study, the drivers always had the option to refuse the personal price, and choose a conventional charging station with a fixed price instead. The personal prices received by the drivers could differ significantly from each other, but were almost always lower than the market prices. For the providers of charging stations, the iterative AI algorithm can find out which individual prices are accepted by the buyer, and under which conditions. However, during the course of the study, the researchers noted that on some occasions the algorithm raised the price significantly when the electric car's batteries were almost completely empty, and the driver consequently had no choice but to accept the offer.
“Smart charging stations can solve complex pricing in a competitive market, but our study shows that they need to be developed and introduced with privacy protection for consumers, well in line with responsible-ethical AI paradigms”, says Balázs Kulcsár.
More about the study
The researchers created a mathematical model of the interaction between profit-maximising fast charging stations and electric car users. The "charging stations" could offer public market prices or AI-driven profit-maximising personal prices, which the "electric car users" could then accept or reject based on their own conditions and needs. In most cases, the results were promising, as the AI-generated prices were lower than the market prices.
The research is presented in the paper: Personalized dynamic pricing policy for electric vehicles: Reinforcement learning approach published in the journal Transportation Research, Part C: Emerging Technologies
The researchers involved in the study are Balázs Kulcsár, Sangjun Bae and Sebastian Gros, and they are active at Chalmers University of Technology, Sweden; Seyong Cyber University, China, and Norwegian University of Science and Technology.
The research has been financed by the Swedish Electromobility Center and partially by the EU project E-Laas.
For more information, please contact
Balázs Kulcsár, Professor, Department of Electrical Engineering, Chalmers University of Technology, +46 31-772 17 85, kulcsar@chalmers.se
The contact person speaks English and is available for live and pre-recorded interviews. At Chalmers, we have podcast studios and broadcast filming equipment on site and would be able to assist a request for a television, radio or podcast interview.
JOURNAL
Transportation Research Part C Emerging Technologies
METHOD OF RESEARCH
Computational simulation/modeling
SUBJECT OF RESEARCH
Not applicable
ARTICLE TITLE
Personalized dynamic pricing policy for electric vehicles: Reinforcement learning approach
Detecting machine-written content in scientific articles
The recent surge in popularity of AI tools such as ChatGPT is forcing the science community to reckon with its place in scientific literature. Prestigious journals such as Science and Nature have attempted to restrict or prohibit AI use in submissions, but are finding it difficult to enforce because of how challenging it is becoming to detect machine-generated language.
Because AI is getting more advanced at mimicking human language, researchers at the University of Chicago were interested in learning how frequently authors are using AI and how well it can produce convincing scientific articles. In a study published in the Journal of Clinical Oncology Clinical Cancer Informatics, Saturday, June 1, Frederick Howard, MD, and colleagues evaluated text from over 15,000 abstracts from the American Society for Clinical Oncology (ASCO) Annual Meeting from 2021 to 2023 using several commercial AI content detectors. They found that there were approximately twice as many abstracts characterized as containing AI content in 2023 as compared to 2021 and 2022 – indicating a clear signal that researchers are utilizing AI tools in scientific writing. Interestingly, the content detectors were much better at distinguishing text generated by older versions of AI chatbots from human-written text, but were less accurate in identifying text from the newer, more accurate AI models or mixtures of human-written and AI-generated text.
As the use of AI in scientific writing will likely increase with the development of more effective AI language models in the coming years, Howard and colleagues warn that it is important that safeguards are instituted to ensure only factually accurate information is included in scientific work given the propensity of AI models to write plausible but incorrect statements. They also concluded that although AI content detectors will never reach perfect accuracy, they could be used as a screening tool to indicate that the presented content requires additional scrutiny from reviewers, but should not be used as the sole means to assess AI content on scientific writing.
JOURNAL
JCO Clinical Cancer Informatics
ARTICLE TITLE
Characterizing the Increase in Artificial Intelligence Content Detection in Oncology Scientific Abstracts from 2021 to 2023
ARTICLE PUBLICATION DATE
1-Jun-2024
Innovating learning with ChatGPT-based Prompt Tutor
SINGAPORE MANAGEMENT UNIVERSITY
IMAGE:
SMU ASSOCIATE PROFESSOR OUH ENG LIEH IS DEVELOPING SINGAPORE’S FIRST CHATGPT-BASED PROMPT TUTOR TO PROVIDE REAL-TIME FEEDBACK TO ENHANCE STUDENT LEARNING.
view moreCREDIT: SINGAPORE MANAGEMENT UNIVERSITY
By Jovina Ang
SMU Office of Research – “Giving students immediate and frequent feedback makes online learning more effective,” Associate Professor Ouh Eng Lieh told the Office of Research.
However, based on how most online lessons are designed, questions could not be answered nor doubts clarified until students meet their instructor in the following face-to-face class.
The time delay of a few days to a few weeks can impede student learning as it might make it difficult for students to catch up and understand the subsequent topics in the course.
Learning also does not occur until the knowledge gained is stored in long‐term memory.
The Cognitive Load Theory (CLT) states that for short-term knowledge to be committed to long-term memory, there must be effective cognitive load management, repeated reinforcements and clarification of doubts.
“That’s why providing timely feedback is vital for maintaining a seamless and effective student learning experience. It also helps the student to manage the cognitive load and learn,” Professor Ouh explained to the Office of Research.
He added: “It has been found that reflection prompts, which are essentially reminders to get students to reflect and journal their learnings, can significantly improve learning by getting students to retain and apply what they have learned. Additionally, by leveraging ChatGPT, we can evaluate the degree of understanding by comparing what the student has written to the context or the topic of the lesson.”
“With these in mind, we wanted to develop a learning tool that can provide personalised prompts so that students could either have their questions answered or doubts clarified as they go through their online lesson. This tool, which we have fondly named ‘Prompt Tutor’, will also be programmed to assign additional exercises or generate a quiz for students to hone their learning,” Professor Ouh elaborated.
“And to the best of our knowledge, we believe that this will be the very first Prompt Tutor to be developed in Singapore for teaching computing courses in tertiary institutions,” Professor Ouh went on.
The research
Professor Ouh’s project is funded by an MOE Tertiary Education Research Fund (TRF) grant. An expected three years is needed to develop a fully functioning Prompt Tutor and conduct the experimental research.
He is collaborating with Associate Professor Tan Kar Way and Assistant Professor Lo Siaw Ling, both of whom are from Singapore Management University (SMU), as well as Dr. Lin Feng from Singapore University of Social Sciences (SUSS).
The research builds on Professor Ouh’s previous work where he has successfully developed a Doubt Identification Machine Learning Model to single out doubts from written reflections.
The integration of Prompt Tutor to SMU’s ITSS (Interactive Tutorial Software System) makes it possible for the research team to engineer real time feedback by providing the context to the machine learning algorithms.
The initial part of the research project entails developing a sufficiently accurate Prompt Tutor – essentially a Prompt Tutor that can provide at least 90 percent accurate responses to the student queries or detect doubts in the written reflections.
A total of 80 SMU students enrolled in introductory programming courses will be recruited to participate in the research project. 40 students will be exposed to the Prompt Tutor while the remainder 40 will not be.
There are four steps in the research methodology.
Step 1: Students will be required to watch a five-minute video and write a reflection on what they have learnt.
Step 2: With the use of the Doubt Identification Machine Learning Model, the Prompt Tutor will be programmed to detect doubts or inaccuracies in the student reflections.
Step 3: The Prompt Tutor will then evaluate the students’ written work by checking against the context and accuracy of the topic of the lesson. If any doubt or inaccuracy is detected, the Prompt Tutor will notify the student by sending a personalised prompt such as asking the student a follow-up question for him/her to further reflect and learn.
Step 4: If no inaccuracy is detected, the Prompt Tutor will either prompt the student with more exercises or generate a quiz to test understanding.
Implications of the research
This research has wide ranging applications beyond the teaching of Computer Science or Information Systems courses.
Given that the development of the Prompt Tutor is based on CLT, and therefore, discipline agnostic, its use can be easily extended to the teaching in fields such as Social Science, Medicine, and even languages.
Other than benefiting the students, the Prompt Tutor can benefit instructors – by reducing the time spent on student consultation.
No comments:
Post a Comment