Or, a robot does the dishes

For a while now we’ve been talking about transformers, boundary neural network logic models, as a transformative technology, no pun intended.

But now these attention mechanisms have other competing methodologies that can push AI’s ability to function even further, toward what people now call AGI, or artificial general intelligence, where AI becomes better than humans at routine tasks. .

To illustrate one of these new alternatives. I’ll be drawing heavily from a presentation by Konstantin Rusch that happened recently at TEDxMIT, and I’ll also get some help from ChatGPT along the way to put some of the more technical material into plain English.

Using oscillatory state-space models

“The new models are all powered by transformers or liquid base models,” said Rusch, “but what if I told you there’s a whole new radical approach, a new AI system that’s more efficient, more expressive and more capable?”

Then Rusch explained oscillatory state-space models, where you take a mass-source system, plot its position in time, and review what he called “beautiful harmonic oscillations.”

This sort of thing, he said, abounds in biological neuroscience.

“If we look at a single biological neuron, the firing of the action potential is actually nothing more than a kind of relaxing oscillator,” he noted. “But even if you look at entire networks, we find that those oscillatory dynamics emerge.”

He cited the example of the EEG invented by Hans Berger in 1924.

“Can we somehow abstract the essence, the core idea of ​​oscillators and build an AI system based on that?” he asked.

After facing this, I still wasn’t completely clear about the oscillatory state space models, so I asked ChatGPT to explain it in a simple way, in two paragraphs.

Here’s what I got:

“Imagine watching a spring bounce up and down or tracking temperatures that rise and fall with the seasons. A state space oscillatory model is a mathematical tool that helps us describe and predict these types of back and forth patterns. It breaks down the movement or changes in “states”—hidden parts of a system that evolve step by step—and shows how they cycle in an orderly, repetitive fashion. Instead of constantly going up or down, these states swing like a pendulum, capturing the idea that something is repeating itself over time. Using state space models, we can construct equations that keep track of what’s going on inside a system (such as how stretched a spring is or how high or low the temperature might be), as well as what we can see or measure from the outside. . When these equations describe repeated ups and downs, we call them oscillatory. This approach helps us predict future behavior – like figuring out when a spring will be at its lowest point or a climate cycle will peak – so we can plan or control things effectively, be it in engineering, economics or even brain research.

So this methodology helps to keep track of what’s going on inside a system – not just something like a resource, but perhaps a neural network…

Neurological models in development

Rusch went on to talk about how these systems mimic the human brain.

“We would expect the system to be very, very expressive, but then it would further bridge (and bridge) the gap between artificial and biological intelligence,” he said.

Contrasting state-space oscillatory models with recurrent neural networks, he noted how the relationship between weighted connections works and how there are “dense connections” in the system. It also addresses the principle of sparse representation. But all this adds up to one thing: faster and more powerful systems.

“If you have a standard repeating pattern, the computational time is proportional to the length of the sequence,” he explained. “For our approach, we can show that the computation time is actually proportional to log 2 of N. That’s an incredible speedup.”

Using linked parallel scans

In the middle of the explanation, Rusch mentioned a term that is useful in understanding how these models do things much faster.

Again, if you go and ask ChatGPT, you’ll find that a linked parallel scan is a method in computation that applies a certain shift to each part of an array or matrix. The model tells us this in a nutshell:

“Imagine you have a list of numbers, like the distances your car has traveled every hour on a road trip. A parallel scan can quickly figure out the total execution at each hour—hour 1, hour 2, hour 3, and so on—by breaking the list into chunks and having each chunk do its own cumulative work. It then joins those pieces together into a final list that shows the total distance you’ve traveled at each step along the way. Because this happens in parallel, it can be much faster than having a single processor do each step one at a time.”

This helps to explain some of the math behind these concepts, but at the end of his presentation, Rusch moves into very different territory.

Universality and Task Robots

Universality is essentially the idea that a Turing machine can complete the work of other Turing machines in a unified field theory of ability.

Rusch mentions this idea in the transition from experimental things to actual laboratory experiments.

“We trained a humanoid robot in our lab to do some kitchen work,” he explained. “The oscillating dynamics appear …meaningful physical representations, representations that were remarkably close to human trajectories.”

This in itself points to a solution to some of the biggest challenges people notice about modern robotics.

The argument often goes like this – yes, AI is incredibly intelligent at processing information and creating things, but what about physical dexterity?

Many people have misconceptions about what AI would need to put into a robot to be able to mimic human movements. This begins to define how this would work in detail, and so you would have robots capable of doing the physical human work that we all take for granted is our exclusive domain – washing dishes, taking out the trash, or caring for a man, with hygiene and other personal care. All this, or cooking your favorite meal with whatever you have in the fridge.

In a way, it’s the last leap we haven’t seen yet. Our computers are super smart, but they don’t have physical bodies. I would submit to you that this will all change soon.