By Tom Edwards
Artificial intelligence is changing fast. We’ve gone from AI that understands and writes human-like text to AI that can make decisions and take actions in the real world. This shift is more than just an upgrade—it’s a big step toward creating AI that understands cause and effect, known as Causal AI. Let’s explore what this means and whether moving from language-focused AI to action-oriented AI is leading us down that path.
In a recent AI Keynote, I talk about the coming shifts in AI and how the next 18-24 months will introduce seismic shifts as we move from LLM driven experiences towards action-oriented systems that span multiple modalities. From agentic architectures that will unify digital tasks to the coming wave of physical robots powered by deep learning systems that inch us closer to truly causal AI based systems.
What Are Large Language Models (LLMs)?
LLMs are AI systems trained on huge amounts of text data. They can understand context and generate human-like sentences. Examples include models like GPT-4. They excel at:
• Understanding Language: Grasping the meaning behind words and phrases.
• Generating Text: Writing essays, answering questions, and more.
• Recognizing Patterns: Finding relationships in language data.
Limitations:
• No True Understanding: They predict words based on patterns, not real comprehension.
• Lack of Causality: They don’t understand cause and effect—just correlations.
Introducing Large Action Models
These models take AI a step further. Instead of just processing text, they can interact with their environment.
• Interacting with the World: Making decisions based on input and feedback.
• Learning from Outcomes: Adjusting actions based on results, using techniques like reinforcement learning.
• Handling Multiple Data Types: Processing visuals, sounds, and text together.
Why It’s Important:
• Autonomy: They can perform tasks without constant human guidance.
• Adaptability: They learn and improve from experiences.
• Real-World Applications: Useful in robotics, self-driving cars, and more.
What Is Causal AI?
Causal AI understands why things happen, not just that they happen.
• Cause and Effect: Knows how different factors influence outcomes.
• “What If” Scenarios: Can predict what might happen if circumstances change.
• Explainable Decisions: Can tell us why it made a certain choice.
Benefits:
• Better Decisions: More accurate predictions and actions.
• Reliability: Performs well even in new or changing situations.
• Ethical AI: Helps prevent biases and unintended consequences.
Connecting the Dots: Are We Moving Toward Causal AI?
From Understanding to Acting
LLMs are good at processing language but are passive. Large Action Models actively engage with their surroundings, which is key to understanding cause and effect.
Learning by Doing
By interacting with the environment, these models can:
• See Results: Observe what happens after they act.
• Adjust Accordingly: Learn from successes and mistakes.
• Build Understanding: Start to grasp which actions lead to which outcomes.
Challenges Ahead
• Complexity: Real-world situations are complicated.
• Data Needs: Requires lots of varied and high-quality data.
• Technical Hurdles: Current AI needs improvements to truly understand causality.
Looking Forward
So, is moving from LLMs to Large Action Models leading us to Causal AI? The short answer is, potentially yes.
• Integrating Causality: By combining action with learning, AI can begin to understand cause and effect.
• Research Progress: Scientists are working on methods to help AI models grasp causal relationships.
• Ethical Considerations: As AI becomes more advanced, we need to ensure it’s used responsibly.
Final Thoughts
We’re on an exciting journey in AI development. Transitioning from language-focused models to action-oriented ones is a significant step toward smarter, more understanding AI. While there’s still work to be done, especially in teaching AI about causality, the progress so far is promising.
By continuing to innovate and addressing the challenges head-on, we can move toward AI that not only interacts with the world but truly understands it.
Follow Tom Edwards across social media @BlackFin360