Eren Kurshan
Dr. Eren Kurshan is an AI researcher and technology executive focused on building AI systems for large-scale industrial use cases. Kurshan received her Ph.D. in Computer Science from the University of California, Los Angeles, as well as a Master's in Computer Science and a Bachelor of Science in Electrical Engineering. She has been leading AI, machine learning and innovation programs at Morgan Stanley, J.P. Morgan, Bank of America and IBM T.J. Watson Research Labs. She was a Visiting Fellow at Princeton's Center for Information Technology Policy (2015-2016) and served as an Adjunct Professor at Columbia University between 2014-2020. Dr. Kurshan published over 80 peer reviewed technical publications and holds ~265 patents, with approximately 125 granted. She has served as an associate editor of several IEEE and ACM journals and transactions including the Transactions on Emerging Technology, Transactions on Computers and the Journal of Emerging Technologies in Computing. She was the recipient of 2 Best Technical Paper Awards from IEEE and ACM conferences, as well as top inventor and licensing awards from Bank of America and IBM. She received 2 Outstanding Research and Corporate Accomplishment Awards from IBM for her work on system design and optimization and emerging technology development respectively. Dr. Kurshan received the "Inventor of the Year Award" from New York Intellectual Property and Law Association for her contributions in financial crime detection computer systems.
less
Related Authors
Uploads
Papers by Eren Kurshan
The leap from AI to AGI requires multiple functional subsystems operating in a balanced manner, which requires a system architec- ture. However, the current approach to artificial intelligence lacks system design; even though system characteristics play a key role in the human brain; from the way it processes information to how it makes decisions. System design is the key to alignment, one of the most challenging goals in AI. This difficulty stems from the fact that the complexity of human moral system requires a similarly so- phisticated system for alignment. Without accurately reflecting the complexity of these core moral subsystems and systems, aligning AI with human values becomes significantly more challenging.
In this paper, we posit that system design is the missing piece in overcoming the grand challenges. We present a Systematic Ap- proach to AGI that utilizes system design principles to AGI, while providing ways to overcome the energy wall and the alignment challenges. This paper asserts that artificial intelligence can be real- ized through a multiplicity of design-specific pathways, rather than a singular, overarching AGI architecture. AGI systems may exhibit diverse architectural configurations and capabilities, contingent upon their intended use cases. It advocates for a focus on employing system design principles as a guiding framework, rather than solely concentrating on a universal AGI architecture.
The leap from AI to AGI requires multiple functional subsystems operating in a balanced manner, which requires a system architec- ture. However, the current approach to artificial intelligence lacks system design; even though system characteristics play a key role in the human brain; from the way it processes information to how it makes decisions. System design is the key to alignment, one of the most challenging goals in AI. This difficulty stems from the fact that the complexity of human moral system requires a similarly so- phisticated system for alignment. Without accurately reflecting the complexity of these core moral subsystems and systems, aligning AI with human values becomes significantly more challenging.
In this paper, we posit that system design is the missing piece in overcoming the grand challenges. We present a Systematic Ap- proach to AGI that utilizes system design principles to AGI, while providing ways to overcome the energy wall and the alignment challenges. This paper asserts that artificial intelligence can be real- ized through a multiplicity of design-specific pathways, rather than a singular, overarching AGI architecture. AGI systems may exhibit diverse architectural configurations and capabilities, contingent upon their intended use cases. It advocates for a focus on employing system design principles as a guiding framework, rather than solely concentrating on a universal AGI architecture.
Today’s AI solutions face a trifecta of challenges: The Great AGI Leap, The Energy Wall and The Alignment Problem. Recent AI solutions have been quite energy inefficient. They consume unprecedented amounts of energy during training and unsustainable peak power during run-time. Making things worse, the amount of compute used for training doubles every 3.5 months. Current approach to AI lacks system design; even though system-level characteristics play a critical role in the human brain; from the way it processes information to how it makes decisions. For the AGI Leap, the required integration and balanced operation of multiple functional subsystems is impossible to achieve without system-design. Lastly, for the alignment problem, AI lacks the capacity to employ multiple subsystems (such as System 1 and 2, model-free and model-based learning) in a balanced way for moral decisioning. In this talk, we investigate the importance of system design for next generation AI solutions and argue that system-design is the missing piece without which the three grand challenges may never be solved.
To enhance energy efficiency, it is essential to leave the current restrictive view of AI as a software only solution and embrace fully integrated system design and novel hardware technologies, such as neuromorphic computing.
Addressing alignment challenges involves recognizing the pivotal role of system architecture in moral decision-making, echoing the human brain's reliance on signal comparators, feedback mechanisms, and control functions, without which it will be nearly impossible to achieve alignment.
System design also proves essential for advancing AGI solutions from multiple narrow AI models to integrated co-processing and high-level AGI functions.
In this talk, we argue that system design is the key to finding solutions to all 3 seemingly unrelated grand challenges. Energy efficiency requires a paradigm shift from
the way we built AI as a software-level tool running on general purpose hardware to achieve orders of magnitude improvement needed. Similarly, one of the fundamental challenges in alignment is the fact that the human brain relies heavily on its system architecture for moral decisions. Without a system and its specialized components, feedback mechanisms, control and regulatory functions it may never be possible to achieve alignment. System design is essential in building AGI solutions from the current state of narrow AI capabilities. However, our current vision of AI lacks system-level thinking. Fixing this not only requires a paradigm shift in AI, but also a cultural shift to incorporate the learning from other computer science disciplines into the narrow AI community.
I argue that effective tackling these challenges relies on system design. To enhance energy efficiency, it is essential to leave the current restrictive view of AI as a software only solution and embrace fully integrated system design and novel hardware technologies, such as neuromorphic computing.
Addressing alignment challenges involves recognizing the pivotal role of system architecture in moral decision-making, echoing the human brain's reliance on signal comparators, feedback mechanisms, and control functions, without which it will be nearly impossible to achieve alignment. System design also proves essential for advancing AGI solutions from multiple narrow AI models to integrated co-processing and high-level AGI functions.
Hardware and Systems Approach to Nextgen AI and AGI
Artificial Intelligence encounters three grand challenges: The Energy Challenge, characterized by a troubling and unsustainable rise in training energy consumption; The Alignment Challenge, where jail-broken and misaligned AI pose significant safety and societal threats; and The AGI Challenge, involving the transition to Artificial General Intelligence, of fully integrated, coherently functioning modalities and higher level functions. We argue that effectively tackling these challenges relies on system design and leveraging substrate capabilities. To enhance energy efficiency, it is essential to leave the current restrictive view of AI as a software only solution and embrace fully integrated system design and novel hardware technologies, such as neuromorphic computing.
Addressing alignment challenges involves recognizing the pivotal role of system architecture in moral decision-making, echoing the human brain's reliance on signal comparators, feedback mechanisms, and control functions, without which it will be nearly impossible to achieve alignment. System design also proves essential for advancing AGI solutions from multiple narrow AI models to integrated co-processing and high-level AGI functions.
Enterprise knowledge graphs have become increasingly popular in the past decade. They are actively used in recommender systems, drug discovery, image understanding and semantic search. Knowledge graphs are also actively used in AI models for knowledge representation and reasoning capabilities.
Contemporary knowledge graphs are challenged by the size and complexity of real-world applications and face significant limitations in representing relationships, which are often highly temporal, evolutionary, probabilistic, uncertain and mathematically complex. Recent attempts of representing this edge complexity has caused challenges in the embedding processes.
In this talk, we explore the current issues knowledge graphs face. We argue that constraining knowledge graphs to pure data storage systems is a restrictive assumption. Integration of computing, memory and communication capabilities provides unique opportunities in advancing graph-based system capabilities.
The proposed Integrated Knowledge Graph Computing framework maps the relationships on neural networks, which then enables arbitrarily complex relationship representations, embedded learning and multi-modal semantic co-processing capabilities, which are highly critical for AI. We look at architectural and system design paradigms to explore potential solution paths towards integrated semantic computing.