When Openai introduced ChatGpt to the world in 2022, it brought generative artificial intelligence into the mainstream, launched the snowman effect, leading to rapid integration into the daily lives of people using industry, scientific research, healthcare and technology.
What’s next for this powerful but incomplete tool?
With that question in mind, hundreds of researchers, business leaders, educators and students gathered at MIT’s Kresge Auditorium to gather the first MIT Generic AI Impact Consortium (MGAIC) symposium on September 17th to share insights and discuss the potential future of generator AI.
“This is a crucial moment. Generation AI is moving rapidly. As technology continues to advance, it is our job to ensure that our collective wisdom maintains pace.”
Emphasizing the key needs of this collaboration, MIT President Sally Kornbluth said the world relies on faculty, researchers and business leaders like MGAIC to tackle the technical and ethical challenges of generative AI as technology advances.
“Part of MIT’s responsibility is to bring these advances to the world. …How can we manage the magic of (general AI) so that we can all rely on ourselves with confidence for critical real-world applications?” Kornbluth said.
Meta’s chief AI scientist Yann Lecun said that the most exciting and important advances in generating AI are likely not likely to come from the continuous improvement or extension of large-scale language models such as Llama, GPT, and Claude. Through training, these giant generative models learn patterns on huge datasets to generate new outputs.
Instead, Lucun and others are working on developing a “world model” that infants learn as they do by looking and interacting with the world around them through sensory input.
“A 4-year-old has seen a lot of data through vision, just like the biggest LLM… The world model will be a key component of future AI systems,” he said.
Robots with this type of world model can learn to complete new tasks on their own without training. Lecun believes that the world model is the best approach for businesses to make robots smart enough to be useful in the real world.
However, even if future generation AI systems become smarter and more human-like through the incorporation of world models, Lecun is not worried that robots will escape human control.
Scientists and engineers need to design guardrails to keep future AI systems on track, but as a society, we have already done this for thousands of years by designing rules to coordinate human behavior with the common interest, he said.
“We need to design these guardrails, but with construction, the system cannot escape those guardrails,” says Lecun.
Amazon Robotics Chief Technology Tye Brady also discussed how generator AI will affect the future of robotics.
For example, Amazon already incorporates generative AI technology into many warehouses, optimizing how robots move, moving materials and streamlining order processing.
He hopes that many future innovations will focus on the use of generator AI in collaborative robotics by building machines that allow humans to become more efficient.
“Genai is probably the most influential technology I’ve seen throughout my entire robotics career,” he said.
Other presenters and panelists discussed the impact of generated AI in companies, from large companies such as Coca-Cola and analog devices to startups such as healthcare AI firm Abridge.
Several MIT faculty members also spoke about their latest research projects, including the use of AI to reduce noise in ecological image data, the design of new AI systems to alleviate bias and hallucinations, and the ability for LLM to learn more about the visual world.
A day after exploring new generative AI technologies and discussing their impact on the future, Professor Patrick J. McGovern, co-leading MGAIC Faculty, Vivek Farias, MIT Sloan School of Management, said, “We hope that attendees will leave to make the possibilities and their possibilities come true.
