AI - The Maker vs. The Operator

Comparing LLM vs. Active Inference AI

ByDenise Holt

May 9, 2023

VERSES AI is a cognitive computing company that has overcome the limitations of machine learning AI, by building a new kind of artificial intelligence based on a real-time world model described as:

AI that grows in knowledge, perception and awareness through the evolving real-time data produced by all things, at all times, throughout our living, breathing, and ever-changing world.”

This is Active Inference AI, an entirely new type of networked AI that is designed for Real-Time Operations, including things like managing the daily operations of a hospital, an airport, global supply chains, an entire smart city, global drone and autonomous vehicle traffic, and other real-world activities. This introduces a new era of artificial intelligence which enables us to streamline, optimize and automate every mission-critical, real-life function that we interact with every day.

The days of training AI on big data with probabilistic results will give way to an interconnected “Internet of Everything” that deploys Active Inference AI throughout the network; inherently secure and accurate because it takes any point of real-time data and makes it smart through an empowered ecosystem of interconnected AI Apps called Intelligent Agents. This networked system operates in tandem with humans, growing its intelligence in sync with humans and the world.

What makes a predictive machine learning model different from the Active Inference AI of the Spatial Web?

Deep Fakes of Intelligence — stochastic parrots.

A Machine Model displays superficial intelligence. It can only give answers with a guess that is accurate to the point of the answer having been included in its training data, otherwise it will make up an answer. It is all based on probability of accuracy, but never 100%. And even at that, it can still be wrong with a tendency to make things up, attempting to make it sound appropriate.

An appropriate output is very different from an accurate output.

Active Inference AI with the Spatial Web Protocol is accurate because it’s based on real-time, real life data that is continually updating as the changing details and context around any entity in any space unfolds over time.

Predictive machines are great as content generators, idea enhancers, and task execution. Active inference can make decisions incorporating real-time data. Active Inference can run the planet. These are two very different animals.

Generative and Transformer AI models have no ability for comprehension, awareness, or perception. They possess no capacity to reason or plan. They have no understanding of how the world works. Predictive pattern matching machines can fake it, given enough data that it models from, but they are merely interpolating and regurgitating the patterns they have recognized that are already in use and familiar to us.

A GPT will tell you a great story about your city, but it can’t control the traffic within your city in real-time because it does not look outward at the world around us. Instead, it looks inward to its historical memory which consists of billions of parameters and words from the past that were downloaded into it. If it tells you a story about New York, that narrative is based on the words and references that it has been exposed to, not what is happening in New York right this instant. And because it has no window into the actual real-time activities within New York itself, it can make tremendous errors which would be unacceptable for mission-critical applications like Smart Cities, or Banking, or Policing, or the myriad other real-world applications in which we need AI

Fluent vs. Factual

LLMs are well-versed because the statistics of fluency are easily obtained from the training data. Logic and accuracy are a much harder ask.

GPT-3, ChatGPT, LLaMA, Bing Chat, and GPT-4 are all autoregressive LLMs. Autoregressive generative model structure operates in a way that when structuring a language sequence, it produces one token after another, reflecting on the result of the previous token to determine what to produce next. (A token is a word or sub word.)

At each step of token production, there is a point of probability taking place for steering toward accuracy or inaccuracy. There will always be probability that the next token produced can take the model down an incorrect path.

Each token produced provides an opportunity to branch off into a unique response, while also producing a new opportunity to stray from the truth. This deviation plays out exponentially with each new fork that is taking cues from the previous token, analyzing it to determine its next move. Each prediction impacts every prediction thereafter. Without reasoning and understanding, these errors are likely to compound as the process feeds on those errors.

At no point is the machine able to ask itself if this string of text entries makes intellectual and reasonable sense.

This problem cannot be fixed with LLMs. There will always be an element of possibility for the model results to go haywire. This is where the AI hallucination factor comes in.

A model based on probability is unmanageable in a continuous environment because if/when it goes off track, there’s no way to rein it back in.

For AI to become factual, it would have to accurately predict every detail of the world.

Enter the Long-Awaited World Model of AI

Spatial Web Protocol HSTP and HSML for the World Model and model predictive control.

VERSES AI is expanding the internet for networked intelligence.

There is a fundamental power within a connected world: collective intelligence, a hive mind.

This computable and distributed intelligence is what powers the AI of the future.

By building a new protocol as a public standard, and then giving it away free to the public, VERSES AI has enabled a frictionless network for scalability.

HSML is the programming language that computes context, enabling the AI’s perception to understand the real-time changing state of anything in the world.

HSTP, the Spatial Web Protocol, will recognize any person place or thing as a unique entity within this network, providing it with a unique ID, and acting as a gatekeeper for access, interaction, and transaction by and between all things.

Any thing inside of any space becomes a nested domain uniquely identifiable and programmable within a digital twin of the earth, producing a model for data normalization.

Contingencies and changing details and circumstances for all objects and situations can now be measured and computed, providing a basis for AI perception, affecting all entities and their interrelationships to each other.

Connecting the planet is a massive undertaking that can only be enacted by putting it in the hands of individuals, developers, and builders. For AI, this is the critical piece for scaling AI to AGI.

This is what VERSE AI has built.

So What Makes Active Inference So Accurate?

Active Inference AI is based on biological design — Embodied AI — with the ability to take action. This the core engine.

Active Inference is so accurate because it continually looks outward into the world, measuring the world, in real-time through a global network of sensors — IoT devices, cameras, robots, drones — anything that is connected within the Spatial Web — the Digital Twin network of the world. This mimics the way humans and animals make decisions. As we use our senses to SEE the world “as it is” — moment by moment — we can then more accurately estimate what might happen if we perform an action.

As a human, if we are walking down the street and see a reckless driver coming toward us, we recognize it, notice the erratic actions, infer what would happen if we stood in the way of the car, and we can make the decision to jump out of the way, and save ourselves. We do this because we are able to take in all of the information about the situation and the context involved. The car, the driver, the speed, the pattern of driving, where we stand in relation to the trajectory of the moving vehicle. Through our senses, we obtain awareness to the situation, and therefore can make decisions based on the perception feedback loop that is updating moment by moment as the car gets closer — to the point of understanding exactly how to take action — when to jump and in which direction to go.

In the Spatial Web, Active Inference AI can reach out and interact with the world through all the networked IoT sensors, machines, and the context of HSML informing the constantly changing details and characteristics of the interrelationships between all things. This creates a cybernetic feedback loop of Perception of the World, updating its model of the world, with belief of what it knows to be true; gaining understanding of the intricacies and inner workings of world so that it can make decisions, and take Action. The more this feedback loop plays out, the AI learns more about the world and the results of actions taken, (just as a child learns about its world as it grows and interacts with it), and the more accurate this AI becomes by further updating its understanding of the World.

What are Intelligent Agents?

AIs vs. IAs

AIs are large, siloed machines with independent functions. IAs are intelligent agents with agency for interoperable behavior.

The Spatial Web creates an ecosystem of distributed intelligence with computational context equipping Intelligent Agents with the right kind of data: data that specifies the details, characteristics, and attributes about any thing in any space and how they interact and relate to each other.

This enables the Intelligent Agents to have agency to function through self-organization using belief updating (what it knows to be true at any given moment). It is a system of distributed cognition throughout a unified network containing multiple entities interacting intelligently with each other, communicating with each other, and sharing what they know to be true. This sort of data facilitated agency empowers these IAs to develop perception, awareness, and curiosity about their environment — the data surrounding all agents within the environment.

This enables curiosity about me and you.

According to VERSES AI Chief Scientist, Dr. Karl Friston, curiosity when referring to the Free Energy Principle is described as the resolution of uncertainty.

Dr. Friston describes curiosity this way, “to be curious you have to imagine what would happen if I did that, and what would I know if I did that, but you’ll have to imagine it before it’s actually happened, which is the big bright line between the anthropomorphic kind of intelligence and the intelligence you find in the thermostat.”

A New Era of Networked Intelligence

VERSES AI has laid the foundation for a universal mind that is grounded in reality, as a digital twin of earth and anything within it.

While Deep Learning/Machine Learning AI creates powerful content generating tools and task execution, it can’t run a city. These predictive machine model neural networks are incredible, but they won’t get to AGI without being incorporated into a world model of everything.

The AI of the Spatial Web can..

Dan Mapes, President and Founder of VERSES AI, says this, “I think that ChatGPT is a miracle and is introducing everyone into the potential benefits of AI in their lives. All the variations of ChatGPT and LLMs in general are total miracles and are saving humanity millions of hours doing things — especially anything mundane. BUT — it’s more of a MAKER. Make me this, Code me this, Tell me this. What it is not good for is being an OPERATOR. Run my hospital, Run my airport, Run my City. For that, you need REAL-TIME data and you need to understand how the mind learns. For that, you need Active Inference (Active = Real Time Decision Making based on Real Time Data and a constantly updated World Model).”

Active Inference AI is not a Language Model “generating words” about the world that are based on a historical encyclopedia of knowledge it has been fed regarding the world.

Active Inference is more like a biological organism that perceives and acts on our World by generating more accurate models, understandings, and beliefs about our World.

These ever more accurate World Models enable better decisions — a Smarter World. This is the true measure of intelligence.

You can learn more about the Spatial Web Protocol and Active Inference AI, by visiting the VERSES AI website: https://www.verses.ai and the Spatial Web Foundation: https://spatialwebfoundation.org.

All content for Spatial Web AI is independently created by me, Denise Holt.

Empower me to continue producing the content you love, as we expand our shared knowledge together. Become part of this movement, and join my Substack Community for early and behind the scenes access to the most cutting edge AI news and information.