Meaning and AI by Dr. Timothy Smith

Photo Source: Unsplash

A few days ago, I participated in a cool conference—X-Bio Convergence in Boston’s Seaport district. The theme of the conference focused on the question of making biology computable. In other words, how can computers and AI begin to make biology and healthcare better through artificial intelligence? The concept focuses on the strength of AI in finding patterns in complex data that humans may not detect. For example, AI’s analysis of electrocardiograms or EKGs now surpasses even the best cardiologists in the world. In the case of EKGs, machines can better and more consistently decipher the subtle differences in the EKG waveforms, providing a safer, more consistent diagnosis for patients. At the conference, one of the panelists from MIT advanced a provocative thought that advances in foundation models have ushered in a new level of cognition for artificial intelligence. The panelist suggested that the latest versions of AI can grasp the meaning of words. For a machine to understand  “meaning” would represent a massive advancement in computer science. However, machine understanding of “meaning” in a human-like way remains a subject of ongoing debate.

Foundation models trained on multiple data types, including text and images, can recognize patterns and generate convincing, human-like text and pictures. As the models become more powerful, the context of words becomes more extended. In other words, foundation models like GPT4 can predict the placement of a word in a sentence and even a paragraph based on probability learned from millions of documents. The human-like text responses and chats give the impression that the computer understands and even feels how people feel. Furthermore, contextual awareness and persistent memory allow new AI agents to perform complex reasoning that builds on previous results and decisions.

Agentic AI is another way of saying groups of AI systems work together to solve problems. Agentic systems can break down a complex problem into smaller parts, delegate the smaller parts to other AIs in the group, and collaborate to find the solution. The different elements in the AI agent communicate with each other through APIs or direct communication. Through inter-agent communication, computer scientists have set up systems that allow the different AIs to create new symbols or words to signify an experience. For example, if a robot arm controlled by an AI can feel the resistance of pushing through water, it could make up a word for that to share with the other AIs. This process, called symbol emergence, represents a new capability for machines to communicate and categorize their experiences.

In light of symbol emergence and human-like responses to questions, the latest generation of foundation models appears more intelligent than ever. Because these systems rely on human explanations of the inter-human experiences fed to them through words, images, and mathematics, they do not experience or “feel” what they are saying. The AI systems rely on statistical patterns, not direct experience. The AI agents do not feel loneliness or wetness as a person does, but the bigger question remains. The AI agents will likely evolve to behave symbiotically with people to enhance human capabilities. However, it may not matter whether machines feel or think like us. Rather, how much control people hand over to machines will dictate how we live and enjoy our world.

Dr. Smith’s career in scientific and information research spans the areas of bioinformatics, artificial intelligence, toxicology, and chemistry. He has published a number of peer-reviewed scientific papers. He has worked over the past seventeen years developing advanced analytics, machine learning, and knowledge management tools to enable research and support high-level decision making. Tim completed his Ph.D. in Toxicology at Cornell University and a Bachelor of Science in chemistry from the University of Washington.

You can buy his book on Amazon in paperback and in kindle format here.