AI learns better when it talks to itself
Inner speech combined with working memory helps AI to learn and generalize across tasks
Okinawa Institute of Science and Technology (OIST) Graduate University
image:
Inner speech and working memory architecture boost AI performance when multitasking and completing complex pattern generation challenges.
view moreCredit: Kaori Serakaki/OIST
Talking to oneself is a trait which feels inherently human. Our inner monologues help us organize our thoughts, make decisions, and understand our emotions. But it’s not just humans who can reap the benefits of such self-talk. Published in Neural Computation, scientists from the Okinawa Institute of Science and Technology (OIST) have demonstrated the potential of inner speech to improve AI learning, showing how AI models can generalize across different tasks more easily when supported by both inner speech and short-term memory.
“This study highlights the importance of self-interactions in how we learn. By structuring training data in a way that teaches our system to talk to itself, we show that learning is shaped not only by the architecture of our AI systems, but by the interaction dynamics embedded within our training procedures,” says first author Dr. Jeffrey Queißer, Staff Scientist within OIST’s Cognitive Neurorobotics Research Unit.
By combining self-directed ‘mumbling’ with a unique working memory architecture, the researchers improved how their AI models learned, adapted to new situations, and multitasked.
Brain-inspired modeling for AI learning
The team has long been interested in content agnostic information processing; the ability to perform tasks beyond the exact situations that we’ve previously encountered, by learning general methods and operations.
“Rapid task switching and solving unfamiliar problems is something we humans do easily every day. But for AI, it’s much more challenging,” notes Dr. Queißer. “That’s why we take an interdisciplinary approach, blending developmental neuroscience and psychology with machine learning and robotics amongst other fields, to find new ways to think about learning and inform the future of AI.”
The researchers initially focused on the AI models’ memory architecture, examining the importance of working memory for task generalization. From remembering instructions to quick mental math, working memory is the short-term ability of a system to retain and use information. By simulating tasks of varying difficulty, they examined the effectiveness of different memory structures, demonstrating that systems which included multiple working memory slots (temporary containers for pieces of information) could generalize better on the tricky tasks of reversing the order of and regenerating patterns.
Upon adding self-mumbling targets—telling the system to talk to itself a certain number of times—the researchers gained better performance, particularly when multitasking or completing tasks with many steps.
“Our combined system is particularly exciting because it can work with sparse data instead of the extensive data sets usually required to train such models for generalization. It provides a complementary, lightweight alternative,” emphasizes Dr. Queißer.
Learning to learn better
Looking forward, the researchers plan to make things ‘messier’. Dr Queißer says, “In the real world, we’re making decisions and solving problems in complex, noisy, dynamic environments. To better mirror human developmental learning, we need to account for these external factors.”
This ties in with the team’s overarching goal to understand the neural basis of human learning. “By exploring phenomena like inner speech, and understanding the mechanisms of such processes, we gain fundamental new insights into human biology and behavior,” concludes Dr. Queißer. “We can also apply this knowledge, for example in developing household or agricultural robots which can function in our complex, dynamic worlds.”
Journal
Neural Computation
Method of Research
Computational simulation/modeling
Article Title
Working Memory and Self-Directed Inner Speech Enhance Multitask Generalization in Active Inference
DRI awarded grant to advance AI and computer science education for K-12 preservice and in-service educators
The $2.7 million Department of Education award is a Fund for the Improvement of Postsecondary Education grant
image:
This FIPSE grant allows DRI to provide educators with professional development training and to create enduring resources that educators across the country can access.
view moreCredit: DRI
“To be competitive in the future, students need comprehensive instruction on AI and computer science,” said Emily McDonald-Williams, Project Director, and Director of DRI’s STEM Education Program. “This grant allows us to provide educators with professional development training and to create enduring resources that educators across the country can access.”
DRI and the University of Nevada, Reno (UNR) will collaborate to create an undergraduate internship program that provides preservice and inservice teachers access to professional learning and applied learning opportunities with newly developed curricula and associated resources. In addition, these institutes will partner to develop virtual learning pathways that lead to a State of Nevada Grades K–12 Introductory Computer Science Education License Endorsement through UNR’s web campus.
DRI will also develop an industry-recognized certification in AI and Computer Science Integrated Instruction, creating additional opportunities to integrate AI and computer science competencies into preservice preparation and inservice professional development. Educators will have access to comprehensive teaching and learning resources on the topics, including standards-aligned Green Boxes - self-contained teaching kits that provide educators with two or more weeks of lesson plans along with all of the supplies necessary to conduct each activity, and associated training modules, to provide teacher preparation programs with sustained support.
In addition to professional development training, 60 Green Boxes covering six AI and computer science topics will be created with content for elementary, middle, and high school education. The Green Boxes provide curriculum, materials, and paired training modules to support teacher instruction and enhance teachers’ capacity to implement AI and computer science education. Undergraduate UNR students will test the Green Boxes and provide input prior to distribution to classrooms.
The grant will culminate in AI education summits in Las Vegas and Reno where educators and stakeholders will be invited to a day of workshops, panels, and hands-on learning.
Over the four-year grant, nearly 1,000 teachers and more than 42,000 K-12 students in Nevada will benefit from the curriculum, materials, and professional development training. In addition, these resources will have an annual post-grant impact of up to 22,000 K-12 students and 240 teachers, sustaining a pipeline of AI and computer science informed teachers and learners.
For more information, please email stemeducation@dri.edu.
About DRI
We are Nevada’s non-profit research institute, founded in 1959 to empower experts to focus on science that matters. We work with communities across the state — and the world — to address their most pressing scientific questions. We’re proud that our scientists continuously produce solutions that better human and environmental health.
Scientists at DRI are encouraged to follow their research interests across the traditional boundaries of scientific fields, collaborating across DRI and with scientists worldwide. All faculty support their own research through grants, bringing in nearly $5 to the Nevada economy for every $1 of state funds received. With more than 600 scientists, engineers, students, and staff across our Reno and Las Vegas campuses, we conducted more than $59 million in sponsored research focused on improving peoples’ lives in 2025 alone.
At DRI, science isn’t merely academic — it’s the key to future-proofing our communities and building a better world. For more information, please visit www.dri.edu.
MLEDGE project proves federated learning can support real-world AI services
The project, coordinated by IMDEA Networks, is a notable example of how advanced research can be translated into tangible benefits for society and industry
IMDEA Networks Institute
After two and a half years of work, the MLEDGE project (Cloud and Edge Machine Learning), led by Professor Nikolaos Laoutaris at IMDEA Networks, has demonstrated that it is possible to combine federated learning with cloud and edge computing infrastructures to develop artificial intelligence solutions that are more secure, efficient, and closer to end users. The project’s results have been translated into real-world applications in both the traditional and digital economy.
Applications that make a difference
The project has developed and tested concrete applications that clearly illustrate its impact. First, real-time COVID risk maps were produced in collaboration with Orange and Acuratio, enabling authorities to take fast and well-informed decisions in the event of in the event of future health crises.
In addition, together with Inmarepro and Acuratio, MLEDGE enabled the optimization of energy consumption in industry by connecting steam pumps from four industrial sites. This approach improves energy efficiency without compromising data privacy and contributes to environmental protection. Inmarepro is currently developing a commercial offering for its customers based on the results and technology transferred through MLEDGE.
Key innovations of the project
Among the project’s main advances are:
- FedQV, a new algorithm inspired by the concept of quadratic voting, which improves the way artificial intelligence models trained on different devices are combined.
- PriPrune, a system that improves their performance without compromising the privacy of users’ data.
- The integration of these innovations into a commercial federated learning platform, developed by the Spanish company Acuratio, as a clear example of technology transfer from the laboratory to Spanish industry.
“Thanks to MLEDGE, we have been able to bring the latest research results directly to companies and clients, demonstrating that it is possible to combine efficiency, security and privacy in real time,” highlights Nikolaos Laoutaris.
Impact on society and the data economy
MLEDGE contributes to a more secure data economy, in which sensitive information from users and companies can be reliably used to train artificial intelligence models, while optimizing the use of energy resources.
The project leaves a tangible legacy in the form of systems that can support the management of health crises, tools that optimize industrial processes, and methodologies that accelerate the adoption of federated learning in Spain. Moreover, “the collaborations established with companies such as Orange, Acuratio, and Inmarepro remain active and continue to explore new applications in areas such as smart cities, digital health, and urban mobility,” adds Laoutaris.
MLEDGE has been funded by the Spanish Ministry for Digital Transformation and the Civil Service, through the Recovery, Transformation and Resilience Plan (PRTR), with funds from the European Union – NextGenerationEU.
No comments:
Post a Comment