AI already has a massive footprint in the digital landscape.

Modern AI models have access to vast amounts of information, from books and websites to images and videos, along with social media.

However, one big dataset is difficult to capture.

The physical world.

10 a.m. in a warehouse is not the same as 4 p.m. in a warehouse. The city bustles with life around the corner every few minutes. A supermarket makes changes to its products. Construction changes buildings. New roads are constructed and old ones are removed.

These changes are a normal part of life for humans.

For machines, it's quite difficult.

This could be one of the biggest data challenges of the physical AI era.

Robots Need More Than Internet Knowledge

Suppose you were to feed a robot a million images of chairs.

Can get pretty good at identifying a chair.

However, recognition isn't the only thing.

The robot should also be able to know the position of the chair, distance, if something is obstructing the chair movement, and how the environment has changed.

Spatial data is very useful in this scenario.

For Physical AI systems, they require information about locations, distance, movement, objects and environments.

Not everything can be downloaded from the Internet.

It is something that must be picked up.

Your Phone Is Already a Sensor

So, that's where the fun begins.

In today's day and age, smartphones are equipped with cameras, motion sensors, GPS and incredibly powerful processors.

We primarily use those features to take photos, scroll through social media and watch videos.

However, the same hardware could be the component of a much greater sensing network.

A limited number of high-cost mapping vehicles or teams of data collection personnel could no longer be the only sources of environmental data, but thousands or millions of devices could provide data about the environments around them.

A single phone doesn't make a worldwide map.

Millions of devices might be able to.

This is the larger concept of physical-world data networks that are decentralized.

How to transform photos into spatial intelligence.How to convert photos to spatial intelligence.

From Photos to Spatial Intelligence

The difference between collecting photos without useful spatial information and collecting photos with useful spatial information is important.

An image can provide an AI with information about a building.

It can be potentially used to describe where that building is, how its environment is organized and how the environment evolves over time, using spatial data.

In the case of robotics it is a huge factor.

Imagine that an independent robot goes into a warehouse.

It must be made aware of the existence of shelves.

It must know where they are.

It must move from one to the other.

It should find blockages.

Ideally, it should have information on what the environment is like today, not months ago.

New environmental data might thus become a key component of physical AI.

The Physical World Is a Moving Dataset

The internet is constantly being updated.

The physical world is also too.

Traffic changes.

Buildings are renovated.

Stores move products.

Manufacturers restructure their facilities.

A big change occurs when construction projects reshuffle whole streets.

Environment can change when the weather changes.

A simple thing such as a parked car can alter the scene a robot faces.

This means that if the data is static it can become stale over time.

But a distributed network that can continuously gather the environmental information can take a different approach to the problem.

Instead of asking:

What was it like here?

AI systems may be more often asking themselves:

What is it like around here?

That's a much more helpful question for autonomous machines.

Why Decentralization Could Matter

There are clear benefits to having centralized data collection.

A company is able to manage the equipment, collection process and quality.

This could be costly and time-intensive to scale globally.

A distributed approach alters the equation.

A network of infrastructure could also be made up of many people's sensors, coordinated together.

What will emerge is similar to a "crowdsourced sensor layer for the physical world".

And that's where the ideas of Web3 enter the picture.

Blockchain infrastructure can be leveraged to offer solutions for tracking how contributions are made, establish possible provenance and coordinate incentives among participants.

The technology is not the answer to the data issue.

The hard part is actually to create useful, accurate and privacy-friendly information.

However, it is worth considering another organization of that network when thinking of decentralized infrastructure.

The Bigger Opportunity Isn't Mapping

It can be used for mapping, which is just one application.

Environmental information is required by robotics companies.

Surroundings of autonomous vehicles must be constantly updated.

Spatial understanding is needed to support AR systems.

Information about the physical environment is required for industrial automation.

The idea of a continuously updated spatial data set is suitable for smart cities.

AI developers creating world models could also require vast amounts of data pertaining to changes in real environments.

That opens up a potential huge market to a traditionally hard-to-collect item:

Information that is available in machine readable format in relation to the physical world.

The Data Layer Behind Physical AI

The first generation of AI was focused mainly on the digital data.

The next generation can require something new.

Machines should learn about space, moving and physical context.

This could extend past servers and datasets to the infrastructure that surrounds AI.

It could eventually become a network with millions of cameras, sensors and everyday devices fed into it from the real world.

The smartphone could be one small part of that infrastructure.

What's interesting is that the hardware is already in place.

What's more important is whether decentralized networks can transform those disorganized devices into a dependable, valuable and scalable repository of spatial intelligence.

If they can, then the next big AI dataset may not be located on the internet.

It might be all around us.

Alpha stuff:

Claim your Zerion XP!

Content: Publish0x & Hive

Play2Earn: Splinterlands & Holozing