Scientists Move Closer to Lifelong Learning AI, Improving Safety and Adaptation
Researchers at Ohio State University are one step closer to developing artificial intelligence (AI) that can mimic human learning. At a recent machine learning conference in Honolulu, they presented their findings on continual learning, a process that enables computers to continuously acquire new skills while retaining previous knowledge. This capability is crucial as society becomes increasingly reliant on AI systems, such as self-driving cars.
The team discovered that artificial neural networks often suffer from catastrophic forgetting, meaning they lose information from earlier training as they take on new tasks. To address this issue, the researchers found that AI networks retain information better when trained on diverse and dissimilar tasks rather than those that share similarities. Just like humans, teaching AI algorithms a wide range of tasks early on allows them to better absorb new information.
Ness Shroff, an Ohio Eminent Scholar and professor of computer science and engineering, who led the study, emphasized the importance of AI not forgetting previous lessons: As automated driving applications or other robotic systems are taught new things, it’s important that they don’t forget the lessons they’ve already learned for our safety and theirs.
The implications of this research are significant. Developing AI capable of lifelong and human-like learning could expedite the scaling up of algorithms and enable their adaptation to evolving environments. This advancement holds promise for various applications and industries.
Meanwhile, another team at MIT presented a technique aimed at disrupting the creation of deepfake images. By injecting tiny disruptive bits of code into source images, they hope to combat the spread of manipulated content.
Major tech giant Google also had a significant presence at the conference, with their AI and machine learning research featuring in over 80 scientific papers. Among their notable contributions were demonstrations of 3-D protein modeler AlphaFold and new models for robotics and video generation.
Google DeepMind, a major sponsor of the event, emphasized the potential of machine learning to address significant challenges. Shakir Mohamed, the director for science, technology, and society at Google DeepMind, stated, From healthcare to climate change, machine learning has huge potential to tackle major challenges and advance society.
The machine learning conference showcased groundbreaking research from various institutions, all aimed at pushing the boundaries of AI and its practical applications. These advancements indicate a promising future for AI, where intelligent machines can learn and adapt like their human counterparts.
As AI technology continues to advance, it is essential to consider the ethical and social implications. While there are significant benefits, there are also concerns about issues like privacy, bias, and the impact on jobs. It is crucial to ensure that the development of AI is guided by social purpose and benefits all people.
As researchers and scientists delve deeper into the possibilities of AI, the world eagerly awaits the advancements that will shape our future. Lifelong learning AI can revolutionize various industries, improve safety, and enhance the quality of life for people worldwide.