Google defines Artificial General Intelligence, or AGI, as
“The hypothetical intelligence of a machine that possesses the ability to understand or learn any intellectual task that a human being can.”
AGI can adapt the skills learned in one domain and apply them to another, to seamlessly align with any unforeseen situations. It holds a diverse library of knowledge regarding social relationships, norms, and facts that help it operate on a common understanding. Just like humans.
Google has taken its “last” step towards turning AGI into a reality.
It has developed a new world model, Genie 3, that allows AI systems to interact with a “simulation of the real world.” It will help with training warehouse machines and autonomous vehicles, as it has successfully done in recreated environments.
For AGI to come to fruition, there is a need for a world model, i.e., to gauge, predict, and solve real-world complex problems (such as healthcare or climate change), it requires an earnest, unbiased insight into how the world operates.
It could easily do someone’s job.
And will play a significant role in AI development, especially those that are intended to function autonomously.
Google claims that its world model will also help humans to recreate simulations using simple text prompts, whether it’s for training or exploration. And any expectations in the simulation would become a reality by adding further text prompts.
“While AIs are trained on vast quantities of internet data, allowing an AI to explore the world physically will add an important dimension to the creation of more powerful and intelligent AIs.”
-says Andrew Rogoyski from the Institute for People-Centred AI.
AGI has largely remained a theory that researchers and engineers are still working towards.
Even though magnificent strides have been made in an attempt to bridge this gap. The closest we have come to developing AGI is narrow AI, one that focuses on specialized tasks.
AGI previously seemed like a far-off concept, at least one that entailed boundless potential.
But Google’s latest is a crucial stride, even with Genie 3’s undisclosed limitations.
And the most fascinating facet is Genie 3’s emergent memory. It allows the model to maintain physical consistency and remember the generated world over time. And teaches itself the physics of how objects move, interact, and fall, all without a built-in engine.
Maybe Genie 3 is a step forward. But with significant limitations, to what extent will this model do what we hope for it to?
The primary aspect is to use the world model to train current AI agents in a dynamic and virtual sandbox.
Here’s the truth.
Given today’s AI models that lack real-world understanding, artificial general intelligence requires significant architectural breakthroughs and a deeper dive into human intelligence. AGI is incompatible with the current AI models- mathematically and fundamentally.
But Google has unearthed a way ahead toward AGI. And unlocked the potential with Genie 3. It has established a foundation, even though it remains in the testing phase.
While its initial impact could be felt across the creative industry and robotics, Genie 3’s potential echoes the advent of general-purpose agents.