OptimalAI
RESEARCH
Pushing boundaries
We bring together advances in AI, physics, engineering and neuroscience to push the boundaries of AI and AGI research.
Research Areas
Human-Centered AI
Across all research areas we believe AI systems should enhance human intelligence, not replace it. Our human-centered AI (HCAI) approach to research pursues user-driven design principles to ensure that the technology serves human needs, is user-friendly, and adaptable to user preferences. We believe that the future of AGI is a collaboration between humans and AI, in which both automation and human expertise are indispensable. AGI decision-making should be transparent and explainable so that humans can update, change or disagree with any AGI outcome.
Grounded in user behaviour understanding and real-world use, OptimalAI is engaged in a variety of HCAI disciplines such as predictive and intelligent user interface technologies, mobile and ubiquitous computing, social and collaborative computing, interactive visualization and visual analytics. Our researchers develop frameworks for evaluating human-AI interaction models, including novel visualization techniques that augment a human's ability to glean meaningful insights from large, disparate data sets and then interact with the system to maximise the most efficient outcome.
We partner with industry-specific experts to apply our scientific advances to real-world challenges and we regularly publish our breakthroughs in Nature, Science, and other scientific journals. We often release projects as open source, and apply research to products to benefit users at scale.
Across all research areas our teams aspire to make scientific discoveries that positively impact industry and humanity. Core to this is our human-centered approach to research and applications.
Our team co-leads the world's largest human-centered AI research program which brings together some of the world's leading AI research organizations and Global 2000 partners including SAP, Volkswagen, ING, Telefonica, Thales and Airbus Defence. The program invents and deploys novel HCAI systems and creates intergovernmental guidelines on creating human-centred AI systems.
AGI decision-making should be transparent and explainable so that humans can update, change or disagree with any AGI outcome.
AGI Techniques
Our AI research focuses on developing specialized AI systems that excel at specific tasks or domains. These systems are designed to solve particular problems, such as image recognition or natural language understanding. However, they typically lack the ability to transfer their knowledge to different tasks or domains. For example, a machine learning model trained to translate languages may not perform well in a medical diagnosis task.
Our Artificial General Intelligence (AGI) research, on the other hand, aims to create machines that possess general intelligence comparable to human intelligence; human-like cognitive abilities, such as understanding context, reasoning, learning from limited data, and exhibiting common-sense understanding. These systems will eventually be capable of learning and understanding a wide range of tasks and domains, exhibiting adaptability, common-sense reasoning, and problem-solving across various domains.
The eventual form of AGI algorithms will depend on a variety of factors, including theoretical breakthroughs, computational resources and practical considerations. Our AGI research explores a wide range of techniques, and our researchers are open to various approaches. Various techniques employed include:
Neural Networks
Deep learning and neural networks have shown great promise in handling complex tasks and capturing patterns in data. Researchers continue to investigate architectures, training methods, and scaling techniques to push the boundaries of what neural networks can achieve.
Reinforcement Learning
Reinforcement learning is another key approach, especially in the context of agents that learn through trial and error in environments. Combining reinforcement learning with neural networks has led to impressive results in areas like game playing and robotics.
Symbolic AI
Symbolic AI, which involves representing knowledge using symbols and performing logical reasoning, has a long history in AI research. Some AGI approaches combine symbolic reasoning with neural networks to leverage the strengths of both paradigms.
Evolutionary Algorithms
Evolutionary algorithms mimic the process of natural selection to evolve solutions to problems. Some researchers explore the use of evolutionary algorithms to evolve AGI systems.
Hybrid Approaches
Many AGI research efforts involve hybrid approaches that combine elements of different AI techniques. For example, a system might use neural networks for perception and learning from data, while relying on symbolic reasoning for higher-level cognition.
Collaboration
While we conduct significant in-house, post-doctoral research, we also collaborate with other research institutions to solve challenges in fundamental research and to pursue innovation in core areas relevant to our products and services. Our scientists work with these organizations to research and develop novel methodologies, models, and algorithms to enhance the capabilities of AI systems and expand their application domains.
In particular, our scientists have worked periodically with Google DeepMind to help unravel the underlying principles governing the behavior and performance of deep neural networks. This includes investigating novel architectures and mechanisms to improve the interpretability, explainability, and reasoning abilities of deep learning systems. The objective being to develop models that not only achieve exceptional performance but also provide insights and justifications for their decisions, enabling users to understand and trust the outputs of these models. These endeavors have aimed to provide deeper insights into the fundamental properties of deep learning systems and lay the groundwork for further advancements.