Exploring the Anatomy of a Digital Edge-Based AI/ML Platform

While we can cite numerous examples of how artificial intelligence/machine learning (AI/ML) has impacted so many facets of our daily lives, a simple example that should resonate for most comes to mind. A handful of years ago many of us audiophiles made the leap from static music media (LPs, cassettes, CD-ROMS, etc.) to digital media, which includes the MP3 format and the many others that followed. On-demand music consumption eventually led to music streaming, which is essentially the accepted norm today.

Over time, popular music “streamers” like Spotify, Amazon, Apple and others have somehow seemingly amassed a digital universe of nearly every piece of music ever recorded and made it available to consumers across the globe with an intuitive internet streaming service enjoyed by millions. This consumption model essentially set the table for AI/ML. Streamers made a reasonable assumption that consumers couldn’t possibly recall all the music they grew to appreciate during their lifetimes (or new, similar music they had yet to hear) and decided to lend some assistance to the consumer. As streaming memberships mature over time, the AI/ML feature within the offering begins to “learn” what we like and from that learning process, begins to suggestively draw conclusions on other titles that we might like based on these derived insights.

An AI/ML-enabled cybersecurity system can, in a similar fashion through an ongoing learning process, discern normal or acceptable user behavior on a given information system and detect anomalous behaviors that deviate from acceptable or permissible user practices. Whether an access attempt is merely to join a network and/or to access specific resources within it, AI/ML can be trained to be wholly cognizant of access attempts based on a broader set of metrics and interdict unauthorized access in real-time.  AI/ML is an integral component to a broader Zero Trust framework where access by personas, devices, locations, etc. is strictly governed to ensure access is granularly limited.

To pull that thread a bit further, there’s something else that might not always be readily apparent. A user may have unscrupulously captured credentials from an authorized user. Other Zero Trust criteria – geography, IP address, date and/or time of the access request – will render the access attempt suspect to a trained AI/ML-enabled system, resulting, in a series of access challenges to validate the suspect user/access attempt.

Similar rules will apply for authorized users who may have nefarious intent in accessing proprietary corporate assets. Many of us have likely encountered a similar scenario when attempting to withdraw funds from our bank while on vacation abroad. The AI/ML-enabled system can conduct this security analysis continuously with millions of valid and suspect access attempts. Clearly this level of protection would be difficult at best with even the most robust cybersecurity team. AI/ML serves as an extension of our cognizant capacity with the flow of live and/or post-egress data.

Evolution of AI/MI

Developing an AI/ML capability is very much akin to raising a child. In its infancy, the parent does much if not all the thinking (Human Decisions) but as development continues, a collective mindshare emerges (Augmented Decisions). Then as the child continues to learn and develop, varied algorithms subjected to learned and established rules, policies, etc. enable it to exponentially extend analytical capacity in autonomous fashion (Machine Decisions). We now recognize that AI/ML is effectively in its third generation of development.

Data/Compute Convergence at the Digital Edge

For an AI/ML platform to truly deliver on its potential, a few requisite considerations must be factored into the equation, particularly as we consider the distributed and varied construct of today’s digital landscape. For example, in the cybersecurity example mentioned above, it is assumed that the network in question is global in scale with perhaps millions of users. If we’re to effectively employ an AI/ML platform to cull and analyze transactional data from this environment, it stands to reason that the platform must also be regionally distributed – strategically adjacent to the sources of data ingested for analysis.

As we’ve become more reliant on timely, actionable insights from AI/ML platforms, a low latency, regionally distributed, edge-based/data source adjacent approach becomes an imperative. Centralized core architectures of the past where streams of data from varied global sources would be fed for processing to a central location is essentially untenable at this point. Positioning this capability at the digital edge assures localized/adjacent symmetry with data sources as well as any peripheral services that come into play with the platform.

Equinix provides the on-ramps to hundreds of CSPs, along with a growing number of vendor ecosystem participants who operate and peer globally on the world’s largest carrier-neutral interconnection platform. Paired with a seamless and frictionless mode of global WAN transport via Equinix Fabric – the industry’s first SDN-enabled Network as a Service (NaaS) – Platform Equinix provides on demand, private WAN transport between any of nearly 60 Equinix Metro locations across the globe.

Traditional centralized-core based AI/ML environments were often instantiated in remote repository locations, creating challenges with delivering data for analysis within a reasonable time frame. Extending and/or regionally distributing an AI/ML capability like AI Anywhere to the digital edge (proximal to original data sources) can greatly enhance functionality and scale by defeating excessive latency while also providing proximal adjacent access to an ecosystem of thousands of clouds, networks and peripheral digital services. Ecosystem participants such as AWS, Google, Microsoft, Verizon, and others leverage Platform Equinix as a digital infrastructure meeting place.

Here are insights on the three trust models employed in next-gen, edge-based AI/ML solutions.

Anatomy of the next generation digital edge-based AI/ML platform – “AI Anywhere”

Next generation AI/ML will require the full utilization of data resources in a secure, controlled environment. As data continues to proliferate from data lakes to data oceans coupled with the increased risk of transporting data to centralized locations for processing it is no longer scalable and does not meet the needs of the mission. The critical component to the solution is establishing a true digital edge where data (IoT devices) is collected and aggregated – think ’edge-in‘ – keeping the data localized and available for analytics. The digital edge unlocks the next generation of AI/ML with three key pillars – distributed environment, data governance and operational simplicity within a highly secure ‘digital edge’ environment.

Equinix provides the secure confines and private interconnection services to thousands of globally distributed digital and cloud service providers to enable this regionally distributed AI Anywhere platform. Single or collaborative multi-tenants can leverage this platform as a service to conduct a myriad of analytics use cases. Instantiated with US Government consumers in mind, the platform is hosted within an Equinix ICD-certified space to ensure requisite security, governance and data custodianship.

AI Anywhere represents a natural progression on Platform Equinix where we believe the following three key technology trends with continue to drive future innovation and digital transformation.

Future infrastructure will be:

• distributed at the edge
• developer led and software defined
• comprised of interconnected partner ecosystems

To learn more, contact us to schedule a Digital Edge Strategy Briefing to help establish your digital roadmap for the future:

For details about the AI Anywhere platform, read the AI Data Marketplace at Equinix white paper