Thumbnail
Use Cases and Applications

BUILD.5, Confidential Computing, and the data privacy revolution. Part 3: AI

Song Choi & Mike Isaia
#build5#confidential_computing#ai

In Part 2 of our series on Confidential Computing we further explored the innovation of CC in secure private transactions on the edge, and the role BUILD.5 will play in delivering trust throughout to even the outer fringes of compute through the Red Hat Openshift Platform. If you have not read that article, we highly suggest you get up-to-date as we will be expanding on the concepts and features introduced in both Part 1 and 2 here.

In Part 3 we will be building on the foundations of cloud and edge, to bridge to the most consequential technology in modern history, artificial intelligence, or more specifically federated AI:

The Limits of Centralized AI

Tools like ChatGPT and other LLMs require the collection and processing of enormous volumes of data in a centralized cloud. The not so hidden secret about this centralized architecture, is it simply does not scale in its current form. ChatGPT currently has over one trillion parameters, and the latest GPT-4 model is already reaching its practical limits. OpenAI’s CEO Sam Altman said it himself “I think we’re at the end of the era where it’s going to be these…giant, giant models,” however, he went on to say that “We’ll make them better in other ways.”

Whether it’s exponentially rising data transfer costs, incompatibility with privacy focused industries like healthcare and government, or performance issues like latency and security, centralized AI is speeding towards a wall as quickly as it experienced its remarkable ascendancy into the zeitgeist of modern society. The need to make AI models “better in other ways” has never been more vital.

Federated Learning and the Decentralization of AI

Federated AI, also referred to as Federated Learning (FL) offers an alternative approach to AI training. FL is a decentralized style of machine learning that uses a global model to collaboratively train itself across a multitude of devices and/or servers. The data used is private and localized, while the learnings and computational workload are shared.

From: https://www.ibm.com/docs/en/cloud-paks/cp-data/4.7.x?topic=models-federated-learning

IBM, NVIDIA, Intel, Microsoft, Google and a growing army of AI startups are aggressively investing in new initiatives in recognition of the role that FL will play in addressing the many shortcomings of centralized AI.

A Confluence of Technological Challenges

As we’ve previously outlined in our CC Showcase, Part 1 and 2, bandwidth constraints of current cloud environments are quickly reaching levels where a reallocation of computational power closer to the device/user will be required. Especially when supporting the quickly maturing, emerging (and converging) technological trends of smart cities, digital tokenized assets, AR/VR, autonomous mobility, and IoT. These are strikingly similar headwinds being experienced by centralized AI today.

In both instances the optimal solution to the problem is a decentralized version of the technology. However, as in any movement to decentralization, new models of security and privacy will be required, something operators in the web3 space, like BUILD.5 know all too well. As we’ve highlighted throughout this series we believe the emergence of confidential computing addresses many if not all of these concerns, however, there is one major “blindspot” that needs to be addressed to properly weave these technologies into a singular, open, self-sustaining operating ecosystem. This is of course verifiable trust at every level throughout the technological stack.

How Modern Applications Are Built

BUILD.5 radically simplifies the consumption of web3 technology by designing integrations based on how modern applications are built. This means designing natural compatibility with Kubernetes-based platforms like IBM’s Supercomputer Vela or seamless interaction with federated learning tools in Watsonx studio; or interconnectivity with confidential computing libraries in the Linux Foundation’s OpenFL, which was recently adopted by Intel to develop privacy focused open source AI using both federated learning and confidential computing.

To put the scale of the impact our integration with modern applications creates into clearer focus, let’s talk about a single integration, one that we explored in our previous article, RedHat’s OpenShift. OpenShift has more than 17,000 corporate clients, 26% of which are large enterprises with revenues in excess of 1 billion USD. More than 90% of all Fortune 500 companies leverage RedHat enterprise products. Which explains why 4.5 million developers use OpenShift to deploy and scale their applications. Now let’s compare that to an estimate by VC firm Electric Capital, that counted a total of 23,334 total developers in the entire web3 industry at the end of last year. The entire industry.

By creating natural compatibility with Kubernetes-based platforms like OpenShift, and interoperability with adjacent technologies like Watsonx, Vela, OpenFL, millions of developers can access BUILD.5 with the click of a button through the applications they use to build and solve problems each day. That is orders, upon orders of magnitude more developers and productivity than the entire web3 industry can hope to develop over the next decade. Imagine the innovations we will discover together. Now, this is a lot to process, so let’s explore this in a more specific scenario to get you a cleaner view of the implications of it all.

Use Case: Data Deserts

An example of a broad application of this tech stack can be seen in what is often termed as “data deserts”. Data deserts are essentially industries, regions, or just situations where data collection is restricted due to privacy concerns (military or healthcare) or insufficient due to practical barriers such as lack of technology (small businesses, rural areas). This is actually a major impediment to the progress of AI models as large datasets develop inherent biases as they are unable to train in some areas when developing their models.

Let’s make a simple outline of how our stack solves this problem:

Conclusion

BUILD.5 radically simplifies the consumption of web3 technology by designing integrations with how modern applications are built. That is vastly different from other web3 projects. Almost everything in web3 is designed to work with how “their” applications are built, treating centralization as an enemy, rather than just another type of tech to interact with. The reality web3 often fails to process, is that the first billion blockchain users will come from web2 building into web3, not the other way around.

So, like TCP/IP, HTTP, and FTP together created the modern internet, the stack of FL, CC, and Blockchain through BUILD.5 has the potential to revolutionize data utilization by decentralizing the foundational aspects of Cloud, AI, and the trust layers it relies on to authenticate truth, sovereignty, and provenance.

Related Articles