Is Active Inference our way to superintelligence?

MIT scholars and AI experts back this theory. Here’s a breakdown of their research on how Active Inference will drive AI forward.

Before we dive in, a quick word:
Hey everyone, it’s been a while since I posted on Medium. This piece kicks off a new series of insights I’ll be sharing in the coming weeks. Join me on this journey! And if you haven’t yet, please subscribe to the Lighthouse newsletter, where I share inspiring achievements, the latest in AI and tech, and personal growth hacks designed to empower you and your teams to thrive in the Age of AI.

There is no denying that our world is increasingly being dominated by AI. From generative models that craft everything from artworks to new medicines, tools like ChatGPT, DALL-E, and others have propelled us into an era of rapid technological progress and innovation. Yet, despite these advances, these AI systems often fall short of reaching deeper levels of understanding and capability.

Are we merely scratching the surface with AI?

What does the next frontier of AI look like?

Active Inference and the Free Energy Principle answer these questions.

In the broad field of artificial intelligence research, these concepts are mainly discussed in academic settings and are typically presented in complex, scholarly language. Moving beyond the limitations of conventional AI systems that rely on static datasets and predetermined outputs, these principles focus on a revolutionary approach: an AI that learns and adapts continuously, in real-time.

MIT’s Thomas Parr, Giovanni Pezzulo, and Karl J. Friston are just some of the scholars who believe Active Inference is the key to fast-tracking the development of a genuine and super-intelligent AI. Let’s break down this concept and explore how it could transform the AI landscape and enable us to leap our technological capabilities into new realms of possibility.

What is Active Inference?

Active Inference is an approach in AI that models the brain’s way of minimizing surprises from sensory input. It’s based on the Free Energy Principle, introduced by neuroscientist Karl Friston, which argues that the brain is continuously predicting sensory inputs and updating its predictions based on what it actually perceives. This minimizes the free energy, or the difference between what the system expects and what it experiences, allowing for a more accurate understanding of the environment.

This method stands out because it integrates perception, cognition, and action into a unified framework. Instead of reacting to the world, Active Inference enables systems to anticipate and shape their interactions with the environment.

What sets Active Inference apart from traditional AI?

While typical AI systems rely on large sets of data and often need to be updated or retrained with new information to stay current, Active Inference AI is dynamic. It actively seeks out information to continuously improve its understanding of the world.

Let’s compare this to Large Language Models (LLMs) like ChatGPT.

LLMs produce responses based on data they’ve been fed during training. Once they’re trained, they don’t learn or adapt unless they’re retrained with new data, which can be slow and costly.

Here is a visualization from Giovanni Pezzulo’s article on the difference between Generative AI and Active Inference that compares two ways to predict a travel destination. Gen AI uses transformer networks that focus on important parts of data to make predictions. Active Inference uses a network that continuously updates its predictions based on new information and interactions, similar to how our brain works when we plan a route or decide where to go.

Active Inference operates like a curious learner, always looking for new information and using it to make better predictions about what might happen next. This means it doesn’t just wait for new data but goes out and finds it. This approach allows Active Inference to adapt to new situations much more like a human would, continuously learning and evolving.

Denise Holt explains this difference more deeply in her article “Unlocking the Future of AI: Active Inference vs. LLMs.”

She points out that while LLMs are good at creating text based on what they’ve already seen, they don’t truly understand or adapt to new situations unless they are specifically programmed to do so. Active Inference, however, integrates learning and adapting as a continuous process, much like how humans interact with the world.

This shows the potential of Active Inference in developing AI systems that need to operate in unpredictable environments or learn new tasks without being explicitly retrained.

Real-world innovative solutions

Situational awareness, adaptability, and autonomy.

These are the core advantages that Active Inference brings to AI systems.

The practical applications of Active Inference AI are shaping up to revolutionize our interaction with technology across multiple domains. By enabling AI systems to learn and adapt continuously, this approach is particularly suited to creating tools that can respond to complex, dynamic environments in real-time.

Consider an advanced property management system empowered by Active Inference.

This system could:

• Monitor Energy Consumption: Analyze patterns across a network of buildings, adjusting HVAC systems dynamically to optimize energy use while maintaining tenant comfort.
• Respond to Peak Demands: Predict potential spikes or drops in energy needs during peak times or unexpected weather conditions, managing resources proactively.
• Enhance Maintenance: Predict maintenance issues before they occur, autonomously scheduling repairs to prevent disruptions.

The shift towards systems that can self-adapt and learn from interactive experiences without needing explicit reprogramming represents a significant leap forward. Tools developed with Active Inference AI will not only be more efficient but also more robust, capable of handling the unpredictable nuances of real-world applications.

Achieving superintelligence

Moving from today’s AI to systems that operate like the human brain involves new technology and a complete reimagining of AI’s role in our society. Everything I’ve learned about Active Inference converge on a singular point:

Active Inference is a pathway to achieving genuine artificial intelligence.

It promises to catalyze innovations that forge deeper connections between humans and machines, paving the way for AI that supports and enhances human endeavors. Embracing Active Inference could well define the future landscape of technology and innovation, setting the stage for a new era where AI and human intelligence amplify each other in unprecedented ways.