Welcome and Gratitude:
Diane Lipscomb, the Society for Neuroscience’s president, began her speech by expressing gratitude to individuals and organizations that made the 49th annual meeting possible, including staff, volunteers, program committee members, and speakers.

Record Attendance:
As of the first day of the meeting, there were 26,815 registered attendees, making it the most exciting neuroscience annual meeting in the world.

NeuroSpace Art Installation:
A giant cube called NeuroSpace, located in the Grand Concourse, is an art installation that merges technical databases and AI methodologies to create a beautiful rendition of the human brain.

Global Community and Science Knows No Borders:
The Society for Neuroscience welcomes colleagues from over 74 countries, recognizing the importance of a global community and the exchange of ideas. Some members were denied US travel visas, so their presentations are being delivered through the Science Knows No Borders program.

Mutual Respect and Scientific Discourse:
SFN promotes an environment of mutual respect and shared values, emphasizing that a person’s appearance, diction, accent, or personal choices have no bearing on their potential contribution to neuroscience. Scientific discourse, questioning scientific premises, and asking for clarification are core values of being a scientist.

Presidential Special Lectures and Future Discoveries:
Attendees will hear from prominent scientists who have made significant discoveries and changed the world through their research. Many attendees will present their own research, potentially leading to innovations, discoveries, and advancements in the field of neuroscience.

Dialogues Between Neuroscience and Society:
The meeting features dialogues exploring the intersections between neuroscience and the global community, supported by Elsevier.

Speaker Introduces Dr. Fei-Fei Li:
Dr. Fei-Fei Li is the keynote speaker at the Society for Neuroscience’s 50th-anniversary annual meeting. Dr. Li is a professor of computer science and director of Stanford University’s Human-Centered Artificial Intelligence Institute. She has a distinguished academic background and is recognized as a global thinker in the field of AI.

Dr. Li’s Accomplishments and Contributions:
Dr. Li is a pioneer in AI research and the inventor of ImageNet, which revolutionized the field and brought AI from the lab into the real world. She is an advocate for using technology to improve humanity and was named the 2016 Global Thinker by Foreign Policy. Dr. Li brings ethical discussions and considerations to the field of AI and promotes technologies that augment rather than replace human experiences.

Dr. Li’s Perspective on AI and Machine Learning:
Dr. Li believes that AI and machine learning have the potential to transform society positively. She emphasizes the importance of developing AI technologies that align with human values and contribute to human well-being. Dr. Li advocates for responsible and ethical development and use of AI to ensure that it benefits society rather than causing harm.

Dr. Li’s Early Academic Conference Experience:
Dr. Li attended her first academic conference as a first-year graduate student at the Society for Neuroscience annual meeting. She expresses her gratitude for being invited back to SFN as a keynote speaker, considering it a homecoming.

Dr. Li’s Upcoming Presentation:
Dr. Li will share her research in artificial intelligence and discuss her vision for the future direction of AI. She will emphasize the importance of ethical considerations and responsible development of AI technologies. Dr. Li aims to inspire attendees to think about the potential of AI to enhance human lives and address global challenges.

Early History of AI:
The quest for creating intelligent machines has been ongoing for over 60 years, starting with Alan Turing’s famous question about whether machines could think. AI pioneers like John McCarthy and Marvin Minsky played significant roles in shaping the field and coining the term “artificial intelligence.” Stanford’s AI lab, founded by John McCarthy, has been a major hub for AI research for decades.

Computer Vision:
The field of computer vision focuses on developing machines that can understand visual information. Early attempts in computer vision involved recognizing lines and edges, inspired by neuroscientific studies on human visual processing. As the field progressed, researchers began exploring more complex visual features and objects.

Challenges in Computer Vision:
A major challenge in computer vision is dealing with the large amount of data and variability inherent in visual information. Machines need to be able to recognize objects despite changes in lighting, perspective, and occlusion. Researchers are also working on developing machines that can interpret and understand the context of visual scenes.

Applications of Computer Vision:
Computer vision has a wide range of applications, including: Object recognition and tracking Image classification Medical imaging and diagnosis Robotics and autonomous vehicles Augmented reality and virtual reality

A Brief History of Computer Vision and Deep Learning:
Computer vision researchers have been working for decades to develop algorithms that allow computers to understand and interpret visual information. In the early days, progress was slow due to the limited capabilities of hardware and algorithms. However, in the 2010s, a combination of advances in hardware, algorithms, and the availability of big data led to a breakthrough in computer vision known as the deep learning revolution.

ImageNet Challenge and the Deep Learning Revolution:
The ImageNet Challenge was an international competition launched in 2010 to develop computer vision algorithms that could recognize 1,000 different everyday objects. In 2012, Professor Geoff Hinton and his students won the ImageNet Challenge using a type of neural network called a convolutional neural network. This result showed that deep learning algorithms could achieve state-of-the-art performance on computer vision tasks.

Factors Contributing to the Deep Learning Revolution:
Advances in hardware, particularly the development of GPUs, enabled the parallel processing required for training large neural networks. The rise of statistical machine learning and deep learning algorithms provided powerful methods for learning from data. The availability of big data, including large datasets of images and labels, allowed deep learning algorithms to learn from vast amounts of information.

Impact of Deep Learning on Society:
Deep learning has had a profound impact on society, revolutionizing industries and driving the fourth industrial revolution. Businesses are using deep learning to automate tasks, improve efficiency, and create new products and services. There has been a surge in startups and entrepreneurial opportunities in the field of AI and machine learning. The global AI market is projected to continue growing rapidly in the coming years.

Conclusion:
Deep learning has transformed the field of computer vision and has had a significant impact on society. Deep learning algorithms are now used in a wide range of applications, from image recognition to self-driving cars. The impact of deep learning is expected to continue to grow in the years to come, leading to new and innovative applications that will revolutionize the way we live and work.

Introduction:
Fei-Fei Li presents a human-centered approach to AI, emphasizing the need to consider human values and societal impact in the development and deployment of AI technology.

Rethinking AI as an Interdisciplinary Field:
AI should no longer be confined to computer science but should become interdisciplinary, involving other sciences, social sciences, and humanities. This interdisciplinary approach is crucial for understanding, anticipating, and guiding the human and societal impact of AI.

Case Study: Machine Learning Bias:
Machine learning bias is a significant concern, leading to dire human consequences in decision-making processes. Fei-Fei Li presents ongoing research at Stanford University aimed at turning machine learning bias into machine learning fairness studies.

Approaches to Machine Learning Fairness:
Research efforts focus on dataset fairness, algorithmic fairness, and mathematical definitions of fairness. Collaborations between computer scientists, mathematicians, statisticians, linguists, and other experts are essential in addressing machine learning bias.

Examples of Interdisciplinary Research:
Rebalancing the ImageNet dataset to address gender, skin color, and age distribution. Algorithmic balancing of natural language embeddings to reduce bias in language models. Mathematical definitions of fairness and algorithmic fairness in collaboration with cardiologists to address long-tail distributions in medical data. Causal reasoning and counterfactual robustness using computational fairness. Collaboration with law enforcement to improve the fairness of AI systems in decision-making processes.

Conclusion:
Human-centered AI is a new approach that brings human values and awareness into the development and deployment of AI technology. Interdisciplinary research is crucial for understanding and guiding the human and societal impact of AI. Ongoing research at Stanford University and other institutions is actively addressing machine learning bias and promoting fairness in AI systems.

Interdisciplinary AI Research for Societal and Human Impact:
Stanford researchers are exploring the impact of AI on the future of work, autonomous robots and vehicles, and refugee policy.

Augmenting Humanity with AI:
AI should strive to enhance human capabilities, not replace them. Automation of 50% of current work activities poses challenges that can be addressed through economic understanding, policy, and technology. AI-assisted healthcare delivery can enhance clinicians’ human care and reduce medical errors.

AI-Assisted Healthcare Delivery:
A collaboration between computer science and Stanford Medical School aims to improve healthcare delivery. Intelligent systems can assist clinicians in providing better care, similar to how driving assistance packages aid human drivers. Medical errors cause a quarter of a million deaths annually in the United States, highlighting the need for improved healthcare practices.

AI in Healthcare:
Hospital-acquired infections are a major problem, with 100,000 deaths annually in the US. AI can be used to monitor hand hygiene compliance and identify areas for improvement. AI can also be used to change clinical behaviors through alarm systems and other interventions.

AI in Aging Homes:
AI can be used to monitor activities of daily living and detect early signs of dementia and other health conditions. AI can also be used to understand gaze changes, which can be indicative of various conditions.

AI in Education:
AI can be used as a teaching aid, providing personalized tutoring and lifelong learning opportunities.

AI in Rescue Operations:
AI-powered robots can be used to go to dangerous places, such as underwater or disaster zones, to save lives.

AI Inspired by Human Intelligence:
AI should be inspired by human intelligence to enable collaboration between humans and machines. AI needs to understand human intelligence and possess the necessary skills to work effectively with humans.

Motivations and Challenges of AI Research:
Current AI systems, despite their impressive performance in object recognition, lack the dynamic, multisensory, complex, uncertain, and interactive nature of human intelligence. Fei-Fei Li’s research focuses on developing AI with human-like capabilities, inspired by the interactive nature of natural intelligence.

Interactive AI Agents:
Collaboration with computational neuroscientist Dan Yamans to create AI agents that interact like babies. Inspired by the phenomenal study of developmental neuroscience and cognitive science.

Interactive Learning:
The AI agent learns by interacting with the environment, receiving feedback, and adapting its actions accordingly. This enables the agent to learn about cause-and-effect relationships and develop a more comprehensive understanding of its surroundings.

Long-Term Goals:
To develop AI agents that can communicate and learn from humans, adapt to new situations, and develop a sense of self. These agents would be able to assist humans in various tasks, such as healthcare, education, and scientific research.

An AI System Intrinsically Motivated by Curiosity:
Fei-Fei Li emphasizes that human intelligence does not develop through labeled data alone. Babies learn through exploration, breaking things, and driven by curiosity. The AI system is designed with intrinsic motivation driven by curiosity.

World Model and Self-Model Networks:
A world model neural network predicts how the world will look after the AI agent’s action. A self-model network predicts the error between the actual world and the AI’s prediction. Reinforcement learning policies aim to maximize curiosity.

Early Results and Behavior:
The AI baby agent (blue curve) shows behavioral patterns similar to human infants. It goes through a stage of understanding self-motion and then shifts focus to objects. The AI agent starts to interact with one and two objects without supervision.

Beyond Supervised Learning:
The AI learning approach is inspired by developmental data and studies of infants. It moves away from typical supervised labeled learning methods.

Collaboration with Robotics:
The project collaborates with robotics researchers to explore AI-powered tool use. Humans are creative tool users, and this research aims to replicate that creativity in AI.

Planning in Complex Environments:
Fei-Fei Li highlights the need to shift robotics from planned, narrow scenarios to complex reasoning and planning in uncertain environments. The goal is to enable robots to navigate and interact with the real world effectively.

Tool Manipulation:
Researchers are exploring how machines can learn to use tools in different ways and in various scenarios. An example is using a hammer to put in a peg by banging or pushing away objects. This involves understanding tool affordances and devising algorithms for robots to use tools effectively.

Integrating Vision and Robotics:
Computer vision plays a crucial role in enabling robots to understand the real world and its complexity. Vision techniques help robots perceive and interpret the environment, including recognizing objects, their affordances, and their relationships.

Composing New Tools:
Researchers are exploring how robots can compose new tools by recognizing different grasping points and affordances. This enables the creation of tools tailored for specific tasks, combining different building blocks.

Multi-Stage Planning:
Multi-stage planning allows robots to break down complex tasks into smaller, more manageable steps. This enables them to plan and execute a sequence of actions to achieve a desired goal, such as clearing a desktop or inserting objects into holes.

Real-World Challenges:
Robotics research involves not only planning and learning but also recognizing real-world scenarios and states. Computer vision plays a significant role in this by providing robots with the ability to perceive and interpret the physical world accurately.

Future Directions:
Researchers aim to combine tool use understanding with multi-stage planning to create real-world multistage tool invention and tool affordance. This involves integrating computer vision, machine learning, and robotics to enable robots to interact effectively with the messy and complex physical world.

AI Learning Through Human Interaction:
AI agents can learn new knowledge by interacting with humans, expanding their knowledge of the visual world. Example: Instagram users post pictures and talk about them, but this type of dialogue is not useful for AI learning.

Elia: An AI Agent for Human Conversation:
Elia is an AI agent designed to engage in human conversations to expand its knowledge of the visual world. It has two goals: engaging with humans by asking questions and expanding its knowledge. Uses a reinforcement learning algorithm to optimize engagement and knowledge objectives.

Elia’s Engaging Questions:
Elia asks interesting and specific questions about images, such as “Is the animal standing on snow?” These questions are more engaging and lead to more knowledge acquisition compared to generic questions like “What animal is that?”

Expanding Visual Knowledge:
Through human interaction, Elia expands its visual knowledge, recognizing more objects and concepts in the world. Examples include recognizing Magpie, Dahlias, and Feta cheese, which it didn’t initially know.

Human-Centered AI at Stanford:
Stanford is committed to research, education, and outreach in human-centered AI. Engaging students, artists, policymakers, journalists, lawyers, business leaders, and civil society to promote human-centered AI thinking.

Opening Discussion with Audience:
The speaker opens the floor to audience questions and discussion for approximately 35-40 minutes.

Moderators’ Introduction:
Matteo Carandini, Isabel Heyman, and Tricia Janik are introduced as the moderators for the audience discussion. The speaker begins introducing each moderator.

Matteo Carandini:
Vice president of research and professor at University College London. Conducts research on brain processing of sensory signals, their integration with internal signals, and how these processes guide decision-making and actions.

Isabel Heyman:
Clinical child and adolescent psychiatrist. Honorary Professor at Great Ormond Street Institute of Child Health at University College London. Expertise in obsessive compulsive disorders in children. Awarded UK Royal College of Psychiatrists Psychiatrist of the Year in 2015-16.

Tricia Janik:
Bloomberg Distinguished Professor in the Department of Psychology and Brain Sciences at Johns Hopkins University.

Empathy in Machines:
Fei-Fei Li highlights the broad spectrum of empathy, ranging from recognizing another’s situation to internalizing emotions and feelings. She suggests that machine development will initially focus on shallower forms of empathy, such as understanding expressions and states.

Potential of Empathy in Machines:
Li believes that advances in neuroscience, brain science, and cognitive science may enable machines to develop deeper levels of empathy. She envisions research assessing patient experiences and emotions through AI-powered systems.

Contextualization of Empathy:
Li emphasizes the importance of contextualizing empathy in various ways.

AI and Ethical Concerns:
Fei-Fei Li emphasizes the importance of considering ethical guidelines and societal collaboration when developing AI technologies.

AI and Neuroscience:
Li’s journey in AI was inspired by neuroscience, particularly cognitive neuroscience. She believes that understanding the brain’s computational principles is a fundamental aspect of AI research.

Engineering and Inspiration from Nature:
Li draws an analogy between AI and airplanes, explaining that while AI aims to achieve engineering goals, it also borrows inspiration from neuroscience, much like airplanes are influenced by the study of birds and aerodynamics.

Li’s Background in Neuroscience:
Li’s undergraduate studies in physics led her to explore the work of physicists like Einstein and Schrodinger, who later turned their focus to life and brain research. She delved into neuroscience during her PhD, conducting research in cognitive neuroscience, fMRI, and replicating Hubel and Wiesel studies.

Neuroscience as a Guiding Force:
Li’s understanding of neuroscience, particularly cognitive neuroscience, has significantly influenced her approach to AI research. She emphasizes the importance of examining neural correlates, computational findings, and cognitive goals of the brain as a computing machine.

Public Perception of AI:
Li acknowledges the need to educate the non-scientific public about AI to alleviate fears and promote trust.

Public Engagement:
Fei-Fei Li believes that public engagement is essential for shaping the future of AI. The public should not just passively accept AI but should actively participate in shaping its development and use. Engagement can take many forms, including writing, art, design, and policymaking.

Communication:
There is a need for better communication about AI to the public. This can be done through writing, art, and design to make AI more accessible and understandable. Policymakers also need to engage with AI experts to develop regulations and policies that will guide the development and use of AI.

Education:
Education is critical for preparing people for the future of AI. This includes both K-12 education and continued skilling for workers. AI education should be inclusive and accessible to all, regardless of background or income.

AI for All:
Fei-Fei Li is passionate about inclusion and diversity in AI. She believes that AI should be used to benefit all of humanity, not just a select few. To achieve this, it is essential to have a diverse workforce in AI, including women and underrepresented minorities.

AI Summer Camp:
In 2015, Fei-Fei Li and Olga Rusakofsky started a pilot program at Stanford University, inviting high school girls to learn about AI. The program was a success and expanded to two campuses in 2016. In 2019, the program was offered on 11 university campuses across the United States. The goal of the program is to create the next generation of AI leaders who are more inclusive and diverse.

AI as a Powerful Tool:
Machine learning offers advanced mathematical optimization and computation capabilities. Neuroscience research leverages AI tools to analyze terabytes of complex data.

Neuroimaging:
Machine learning algorithms aid in understanding neural patterns and multivariate patterns. AI-powered analysis enhances neuroimaging techniques, providing nuanced insights.

Theoretical Models:
Researchers like Surya Ganguli use deep learning algorithms to corroborate models of neuronal learning.

Discovery and Inspiration:
AI serves as a tool for discovery and inspiration in neuroscience research. Models of brain computation and cognitive models benefit from AI’s insights.

Collaborations:
Cross-disciplinary collaborations between AI and neuroscience experts drive innovation in the field.

Societal Acceptance of AI Decision-Making:
Fei-Fei Li believes that society is already becoming comfortable with AI making decisions, as evident in the widespread use of AI-powered navigation systems. However, she emphasizes the need for a nuanced and complex approach to addressing the moral and ethical implications of AI decision-making. She calls for multi-stakeholder discussions involving policymakers, engineers, computer scientists, and the general public to establish common guidelines and institutionalize decision-making processes.

AI’s Role in Addressing Climate Change:
Li expresses hope that AI can contribute to addressing climate change and global warming. She highlights the potential of AI and machine learning to analyze data and provide insights on climate patterns, environmental conditions, and human impacts. She mentions specific research projects using satellite imagery and computer vision to assess sustainability and support climate policies.

Ethical Considerations for Neuroimplantation and Brain-Machine Interfaces:
Li emphasizes the importance of recognizing the ethical and societal implications of neuroimplantation and brain-machine interface technologies as they advance. She draws parallels to the journey of AI scientists, who initially focused on scientific advancements and later encountered ethical issues as AI technologies became deployed in society. She encourages researchers in this field to engage in multi-stakeholder conversations early on to ensure that these technologies benefit humanity and minimize potential harms.

AI’s Accessibility and Equity in Rich and Poor Countries:
Li acknowledges the concern that AI might exacerbate inequalities between rich and poor countries or socioeconomic classes. She stresses the need to prevent AI from becoming a tool that polarizes wealth distribution and incites geopolitical conflicts. She sees the potential for AI to democratize access to healthcare, education, and other essential services in underserved regions through telemedicine, AI-enabled education platforms, and other technological advancements.

Government Regulation and Public Engagement:
Li recognizes the need for government regulation and legislation to govern the development and deployment of AI technologies. She also emphasizes the importance of public engagement and involvement in shaping AI policies and guidelines. She believes that a collaborative effort between governments, industry, academia, and the public is crucial for ensuring responsible and beneficial use of AI.

AI in Law:
Fei-Fei Li and her colleagues at Stanford Law School are exploring the use of AI in law and legislation. From a technical perspective, AI may eventually be capable of writing documents in logical and causal ways. However, the bigger questions lie in the philosophical and ethical implications of AI-generated laws, requiring human involvement.

AI as a Profession:
AI is a young and rapidly growing field with immense opportunities for interdisciplinary collaboration. Students should not limit themselves to majoring in computer science but can explore AI through minors, courses, and interdisciplinary programs. There is a significant opportunity for collaboration between neuroscience and AI, making it an exciting field for young professionals to consider.

Inspiration for Young Scientists:
Fei-Fei Li’s journey and struggles in developing AI serve as an inspiration for young scientists to persevere in the face of challenges. Her openness in sharing resources like ImageNet has contributed to the field’s advancement. Fei-Fei Li’s willingness to confront the potential future implications of AI demonstrates her commitment to addressing complex societal issues.