The artificial intelligence sector is currently dealing with a severe training data bottleneck, especially as centralized technology monopolies are locking out early stage developers from high-quality information pipelines. Decentralized data infrastructure platform Perceptron is trying to address this structural bottleneck by deploying a decentralized infrastructure layer that crowdsources web information through everyday user devices.
Summary
- Perceptron is using idle consumer bandwidth to collect publicly available web data and provide lower cost AI training datasets.
- The platform says its network spans more than 150 countries and rewards contributors while verifying data quality before it is supplied to enterprise clients.
- Perceptron has launched a $10 million AI Data Fund to help developers access data infrastructure and accelerate the development of AI models.
Modern day media is entirely focused on highlighting how leading names in the artificial intelligence space are constantly deploying next-generation hardware systems to buff up their raw computing power. But one of the least talked about operational constraints is the quality of the training data that makes up the core foundation of any functional AI model.
The problem is that with the vast majority of open-web content already thoroughly harvested, aggressive corporate control over public application programming interfaces has locked the remaining foundations of dataset collection behind exorbitant multi-million dollar paywalls. It has essentially become a prohibitively expensive exclusive privilege for a handful of massive tech monopolies.
For the tech giants that are currently leading the AI race, securing these high cost information pipelines aren’t much of a financial challenge, but what about the underfunded innovators? Without the necessary budgets, early-stage startups are left struggling to build competitive products.
“OpenAI pays approximately $60 million to $100 million per year to companies like Reddit and Twitter in order to be able to access data through APIs,” Perceptron Co-Founder & CEO, Peter Anthony told crypto.news during a recent interview.
“Many new AI projects out there don’t have budgets to be able to spend $60 million to $100 million to be able to access data. If you build the best model in the world, it’s pretty useless if it doesn’t have access to good quality data. You could be the smartest kid at school, but if you’re not able to access any books, you don’t really have very much information to present.”
Anthony realized that this market asymmetry leaves room for alternative infrastructure that would serve the independent market segment, which eventually led him to co-found Perceptron, a platform which plans on using idle consumer bandwidth to solve “the data bottleneck problem” AI is suffering from right now.
“The majority of the world’s data has already been accessed and scraped, but there’s a lot of data that’s kind of hidden behind different places that are not yet accessible, so we’re gathering data and positioning ourselves to be able to provide data for AI companies at a reduced cost,” Anthony explained.
Harvesting the idle bandwidth
But what is this idle bandwidth that Perceptron plans to leverage? Anthony explained that this is the unrecognized economic asset that everyday users constantly produce through routine digital browsing, only to watch major corporations extract and profit from it.
“Right now, every time you and I use the internet on our phones, our computers, we’re generating data. That data gets collected, packaged into massive datasets by companies like Google, and sold for millions, sometimes billions of dollars. Yet you and I never see a cent of that value.”
What Perceptron has done is to completely flip this extractive model on its head. They have built a network spanning more than 150 countries comprising roughly 800,000 nodes, and these nodes are powered by individual users who are simply running a browser extension on Chrome or an application on their Android devices.
While these endpoint installations don’t scrape private digital files or provide the firm with sensitive personal telemetry, it instead secures localized geographic perspectives, which Anthony described as “different vantage points” on the open web, which can then be extracted in small pieces and combined into one meaningful dataset.
“It’s very important that we focus on the fact that it’s not using individuals’ data, it’s not tapping into your own personal data and information, but let’s say right now you’re in Malawi. When you’re looking at a particular website, I could go and look at the same website, but chances are, because I’m in Dubai, we’re going to see a different kind of set of results. All we’re gaining from this situation is being able to use your computer to look at something like a normal web page, or whatever it might be.”
To illustrate, Anthony noted that if a corporate client requires a dataset of healthcare-related social media posts from the US, Perceptron can coordinate across its global node mesh to extract individual public posts without interfacing with restrictive enterprise APIs.
Because this data is already freely accessible to the public via any standard web browser, routing the collection through individual terminal nodes legally sidesteps commercial paywalls. Once these minor data packets are retrieved, the network transfers the unrefined data back to a centralized server where specialized artificial intelligence models scrub and audit the information for quality control.
“By doing this, we can cut down the cost significantly that is currently being charged by a lot of the big centralized companies like Google.”
Powered by an economic loop that incentivizes quality network participants
The next question is why would anyone volunteer their hardware to a network like this, and the answer is straightforward, a shared value loop ensuring that these nodes earn points for their passive connectivity that are scheduled to convert into native crypto tokens down the line.
According to Anthony, this distributed model ”will enable them to earn points” that act as a direct metric of their network contribution, and therefore “whenever there’s revenue generated by the company, tokens will get fed back into the ecosystem” to sustain a cyclic economic loop.
“There will also be tokens set aside that are used for buying back tokens,” he added.
However, not everyone running a node essentially qualifies for consistent rewards, as there’s the ever-present challenge of quality control, which can compromise dataset integrity if left unchecked.
Perceptron addresses this by routing gathered packets back to a centralized server, where automated algorithms systematically evaluate the inputs against target benchmarks before releasing any compensation.
Further, Anthony said that the startup recently acquired a company specializing in transaction and payment verification software to structurally automate this validation process.
To further engage network participants while also driving the creation of data sets, Perceptron also plans to launch a structured Data Questing platform, which will allow contributors to turn active human effort into unique training inputs.
“We aim to effectively be able to build datasets and create datasets that are currently not available through centralized processes,” Anthony added.
The end goal
Over the long haul, Anthony said he would like to see the network transition to a business intelligence-focused model that is able to provide deep-layer analytics for enterprise clients.
“The difference is that traditional datasets are static, they’re collected once and quickly become outdated. But there’s an enormous amount of data being generated every time you interact with anything online, and right now, most of it is simply going to waste,” Anthony said.
“One single server trying to monitor all these different users can’t really gather meaningful intelligence at that scale. What we need is a shift toward distributed business intelligence, so we can actually improve services across things like e-commerce, trading, and much more.”
Perceptron has also launched a $10 million AI Data Fund, through which the platform expects to fund independent developers and support the deployment of “actual projects that are providing real services.” Under the terms of the program, selected engineering teams receive five weeks of dedicated data infrastructure assistance and up to 5 TB of real-world data free of charge to accelerate the optimization of early-stage AI models.
“The goal is to support projects as they grow and their data requirements increase. We can become one of their go-to providers, it’s both an investment in the broader ecosystem and a way for us to build consistent, long-term revenue,” Anthony noted.
As of publication time, Anthony said Perceptron is already actively supplying diverse data products to a variety of commercial enterprises. The network provides extensive image datasets to text-to-video generative platforms, including a company called Everlyn AI, to train models to accurately synthesize visual content.
Beyond that, the project is also moving past standard image compilation, as the platform has entered the sentiment analysis sector by tracking public discourse across Twitter, YouTube, and digital asset markets. Analyzing this public sentiment helps crypto firms and exchanges build tracking tools that give early signals to preempt sudden price swings.
