Yelena Yesha, University of Miami: Gen-AI ChatGPT

The speaker, a computer scientist and chief innovation officer, opens the final talk of a conference by addressing her extensive experience in both industry and academia. She introduces the topic of generative AI, emphasizing its transformative impact on science and innovation. The speaker outlines her goals for the talk: to provide an overview of the current state of AI, explain what generative AI is, and discuss its relation to the discovery process. She also mentions her interest in applying AI to solve complex problems with national and international significance, particularly in the context of the University of Miami's Proost Institute, where she works on translational value and AI revolution.

The speaker delves into the definition of generative AI, explaining that it enables the creation of new content by learning and understanding underlying patterns and structures in data. She contrasts this with traditional machine learning, which was primarily analytical and predictive. Generative AI goes beyond analysis to actual content creation, which can revolutionize various fields such as art, entertainment, engineering, and science. The speaker also touches on the ability of generative AI to augment data sets, enabling scientific discovery where data is scarce. She emphasizes the excitement around generative AI, particularly with the emergence of large language models like GPT, while acknowledging their limitations, such as the lack of critical thinking and feelings.

The speaker discusses the importance of critical thinking in scientific discovery and how AI, despite its vast knowledge corpus, lacks this capability. She clarifies misconceptions about AI reaching a singularity where it surpasses human intelligence, emphasizing that AI is not yet at that stage. The speaker provides a brief history of AI development, highlighting key milestones such as the founding of OpenAI and the release of GPT models. She also explains the architecture of AI, focusing on the Transformer model, which resembles neural networks in the human brain and excels at understanding sequential data. The speaker advocates for the involvement of the scientific community in AI development, given its profound implications.

The speaker explores the wide-ranging impact of generative AI on science and the economy. She mentions its potential to overcome major scientific challenges, accelerate drug discovery, and improve clinical trials. Generative AI can also repurpose failed clinical trial molecules for different uses, enhancing efficiency. The speaker provides examples of how generative AI can revolutionize industries like cosmetics and cement by enabling real-time product formulation with minimal environmental impact. She also discusses the horizontal nature of generative AI, highlighting its applicability across various fields, from medical imaging to network security.

The speaker identifies several challenges and limitations in the field of generative AI. She emphasizes the need to address biases in models and data, ensuring that AI is both generalizable and specialized where necessary. The speaker also discusses the importance of privacy, security, and ethical considerations in AI development. She mentions the resource intensity of AI, its environmental concerns, and the need for improved model efficiency and scalability. The speaker calls for a multimodal approach to generative models and stresses the importance of ethical AI, which is an evolving field with significant government interest due to its implications for security and society.

The speaker shares insights into the Institute's innovations in generative AI, which have the potential to revolutionize fields like healthcare, legal, and personalized shopping. She mentions the submission of patents and the commercialization of generative AI through spinoff companies. The speaker highlights specific use cases, such as improving medical diagnosis in radiology and dermatology, and accelerating medical research. She also discusses the development of multimodal language models and partnerships that provide access to medical imaging data. The speaker emphasizes the need to understand the value and functionality of generative AI and how to integrate it into existing ecosystems for positive disruption.

The speaker envisions a future where small generative AI models are integrated onto chips, enabling edge computing and automating various environments like IoT. She discusses the potential for real-time AI processing without the need for training, which could revolutionize how devices operate. The speaker also mentions the Institute's collaboration with the FDA to create a platform for evaluating software as a device, ensuring compliance with regulatory requirements and understanding the performance and fidelity of AI-based devices. She emphasizes the importance of this neutral ground in academia for evaluating the medical soundness and regulatory compliance of these novel technologies.

The speaker discusses the use of AI in university administration, including curriculum development, delivery of classes, and operational efficiency. She highlights the rapid change in technology and the need for forward-looking academic institutions to embrace it fully. The speaker mentions the use of AI for AI, where technology is leveraged to enhance the teaching of AI courses and deliver the best information in a timely manner. She also touches on the business side of universities, where AI is used to manage supply chains, maximize profits, and minimize expenses. The speaker suggests that AI can help make academic institutions more efficient and revolutionize the donation process to universities.

The speaker addresses the trend of universities creating their own proprietary AI models for various purposes, including curriculum and operational efficiency. She discusses the lack of public knowledge about the effectiveness of these tools but notes that universities claim positive results. The speaker also touches on the issue of data poisoning and the responsibility of using AI responsibly, including data content privacy, understanding how to manage the information, and avoiding negative exploitation. She emphasizes the importance of clean data aggregation and the challenges of training models without introducing bias, suggesting that AI can be used to improve the selection of models and algorithms.

The speaker notes the increasing trend of global corporations developing their own AI models, controlling their data sources, and the potential implications for data poisoning. She discusses the fine lines between factual information and data corruption, emphasizing the importance of using AI responsibly and the challenges of maintaining data integrity. The speaker also mentions the potential for customized transformations based on larger training sets to still carry connectivity issues, leading to concerns about data poisoning and the need for responsible AI use.

The speaker outlines the future of generative AI, emphasizing the need for advanced language understanding, enhanced context complexity, and multimodal capabilities. She discusses the importance of personalization, adaptability, scalability, and real-time learning and adaptation. The speaker also highlights the need for ethical and responsible AI, improved interactivity, and textual awareness. She concludes by charting the future impact of generative AI as a catalyst for efficiency and creativity, while acknowledging the challenges of navigating ethical dilemmas and embracing the technology responsibly.

In her closing remarks, the speaker reiterates the transformative potential of AI, emphasizing the need to push conversations beyond AI boundaries and predict what is doable with current infrastructure. She acknowledges the challenges of embracing AI and the fears that it may take over, but encourages a responsible and innovative approach to using AI. The speaker concludes by inviting questions from the audience, highlighting the interactive nature of her talk and her willingness to engage in dialogue about the future of AI.