Valmiz™ AI at the AI Asia Expo: A Novel Multi-Agent Human-Centric Approach
By: Dawn Zoldi (Colonel USAF Ret.)
™, creators of the Valmiz™ Augmentive Artificial Intelligence (AAI) recently captivated audiences at the 4,000 attendee-strong in the Philippines, the location of one of the company’s two international headquarters (the other is in the U.S.). At this exclusive high-caliber event focused on the distinct needs and challenges faced by the Southeast Asian region, over 90 speakers from 15 countries engaged in strategic discussions around AI and its responsible integration across diverse sectors. Among them, Rommel Martínez, ASTN Group’s™ Chief Technology Officer, an AI researcher with over 24 years of experience in the tech industry and the brains behind Valmiz™, a ground-breaking multi-agent human-centric AI, made his mark. This article provides highlights from his various presentations on the issues surrounding contemporary AI and his company’s cutting-edge AI solutions that use novel approaches.
The Flaws in Modern AI
Martínez outlined the limitations of popular AI models such as typical machine learning (ML) and neural networks, including GPT systems.
“Modern AI systems have unpredictable behavior,” Martínez explained. “They are known to hallucinate. There have been numerous cases of accidents with self-driving cars, namely, Tesla and Cruise. There was also a case of a military drone that attacked its operator during a simulation.”
They are “black boxes,” he said, which are not good at handling “black swans,” events that are highly improbable but still occur. Such black box systems also do not allow users to inspect data while using them.
Modern AI systems also cannot stand alone reliably. Most, if not all, of them have centralized operations. That means that if the key servers become unavailable, then the key AI functionality becomes disabled or impaired.
They are inefficient, too. It takes the “energy of a small city to train them,” Martínez said. He noted that OpenAI enlisted Kenyan workers, at a pay of less than $2 USD per hour, to actively comb, sift, and filter data for its popular generative AI.
Finally, these systems are not environmentally sound. The carbon footprint of ML systems in 2022, he said, reached 2020 metric tons.
As an alternative, Martínez presented Valmiz™, an Augmentive AI, a term he coined, which has been more than 20+ years in the making. Augmentive AI presents toolkits that augment a company’s existing ideas, workflows, and pipelines, using knowledge and expertise from different knowledge domains, while putting a human at the center to supervise operations.
It combines features from knowledge bases, traditional databases, and symbolic AI.
The Conceptual Model
Martínez explained the concepts behind his new approach to AI. (Nerd alert – this gets technical!)
When a piece of information is connected to other pieces of information, he said, they form a network. Each of these connecting nodes of information are, in turn, connected to more pieces of information. There’s a point, a threshold, where an information branch has very few connecting nodes relative to the starting node. When you collect this information together, it forms a compound object, a collective network that has both direct and indirect paths to the parent node.
Martínez refers to the amount of information that can be accessed from the center of this network, all the way to the edge, as the “information radius.” This radius sets a perimeter around what can be considered within the context of the central idea.
“When we are able to compute the information radius of any idea, we are able to effectively contain and aggregate information into a single globular unit,” he said. “This unit can then interact with other such units to form super networks.”
In principle, every idea, every object, is connected to each other. Martínez used the example of a mango and a truck. A mango is connected to a truck insofar as a truck has the capability to transport mangoes. Computing the information between those two items is what Martínez calls the “information distance.” The smaller the information distance from a mango to a truck, the less contextual information they need to share. The bigger the information distance, however, the more contextual information they will both need to share. This can be derived both actively and passively.
“By being able to compute information distances, we are able to determine the amount of information traversal needed to properly contextualize them. This also provides information between two points which may be of significant interest to a user,” according to Martínez.
Having the knowledge necessary to perform a task is the key to doing them effectively. Having this kind of accessible and connected knowledge at one’s disposal can enable doing a task in a month instead of days. Normally, acquiring that kind of knowledge would be difficult and time consuming. Now there’s Augmentive AI to do this across a wide range of use cases.
Introducing Augmentive AI
“We named it Augmentive AI,” Martínez said, “because to augment means to enhance, to increase and to support,” Martínez explained. “Valmiz™ is used to enhance an organization’s existing process without changing the workflow.”
ASTN Group™ uses the same kind of AI technology as NASA’s Remote Agent on Deep Space 1. That mission, a flyby of an asteroid and comet 100 million miles away from Earth, required NASA engineers to develop AI that enabled remote code updates on the spacecraft, in order to make mission corrections.
“Building on NASA’s legacy, we created true distributed AI,” Martínez noted. “We removed the traditional heavy reliance on dedicated servers and took a non-monolithic approach.”
Valmiz™ employs multiple agent redundancy. By decoupling AI agents, they can act independently performing specific tasks or they can be used for tighter integration. Such redundant AI agents can receive and execute instructions and still have the ability to converge, to form a “hive mind.”
This further allows the Valmiz™ program and data flow to be examined and patched while operations are being executed. It enables users to perform preemptive manipulation and task adjustments, literally on the fly.
Each agent in Valmiz™—Vera, Veda, Vega, Vela, Vix—has their specific roles.
Veda is the core unit that fuses knowledge graphs and knowledge bases. It is the component of Valmiz™ responsible for converting raw data into indexable knowledge stores. When Vega ingests data sources, it creates a semantic network of all the available data points from various sources.
“The true power of Veda,” Martínez said, “comes from creating worlds inside worlds.” Users can collect heterogeneous information banks into a single block of information. “This is what I call ‘fusing,’” he continued.
The data can be pictures, it can be logistics data, etc. A user can combine them together and they will aggregate into a single block of information. The information backs up once it is collected together.
“Inside of Veda, you can combine different kinds of bindings to correlate and connect information together. These are highly malleable. Registries are the top-level building blocks of Veda. You can manipulate information inside Veda across time. You can have a time series layer traversal and you can set data snapshots—meaning to say, at any point in the computation, you can rollback,” Martínez explained.
Every computation performed inside Valmiz™ is captured with no loss of information. With traditional systems, once you make the computation that is lost in the future. You cannot go back to it.
Vera is the reflective and reflexive key-value database that allows full backward and forward references. In Vera, the input data are called “declarations.” When computing a single object, they contain an identifier, a primary value and arbitrary amount of metadata. All changes that happen with declarations are tracked linearly. This allows users to execute those rollbacks at any given point in time.
Vega is the dynamic storage manager that allows for instantaneous restoration of compound information. With Vega, users can store and restore highly sophisticated types of computation at ease. Unlike modern AI, in the event of a full power shutdown, using Valmiz™’ specially designed state-of-the-art algorithms, users can easily restore terabytes of data in a matter of seconds. In precision operations, seconds count.
Valmiz™ is also fault-tolerant by design. It has a repair mode that allows operators to perform surgical operations and recover from any anomaly.
Vela is the data collector. It compiles data from local and external sources to facilitate information augmentation. Vela essentially acts as a scout that continually scans data regions to extend the information distance of any stored piece of data.
Vix is the human-to-machine and machine-to-machine interface that receives and processes text and voice commands, input and compounds and processes them as they are being made, in real time. When users make a request to Vix, as the user speaks to it, it is already computing. Computations are done on-the-fly as they come through the computation-communication channels, giving users a stream of query-answer pairs.
Humans, Machines, & The Future
With Valmiz™, humans are also the final arbiters, not machines. When in doubt, the better AI systems default to human control. For Martínez, the absence of morals, values and ethics in machines require that humans be the final decision-makers in AI. He designed Valmiz™ to supplement, not supplant, humans in various operations.
Take the drone industry as an example. Valmiz™ can direct certain actions or provide maintenance updates to a single drone or an entire fleet. It can be used to monitor and provide temperature regulation for autonomous and remote medical package deliveries. In the event of a temperature discrepancy, Valmiz™ could direct that dry ice be dispensed inside a transport box automatically. These are just a few of the use cases in one industry that could benefit from this technology.
But there are markets and use cases out there in the future that we haven’t even thought about yet. For this reason. Martínez designed Valmiz™ to be fully integrated into other systems, and future proof.
It also has full modularity. It can be used as a single compound system or as select parts. The source code of Valmiz™ is platform independent and guaranteed to work with definite hardware architectures. In the event that a new computer architecture comes out, users can still be able to build with it.
Martínez purposefully built the system to be reliable. Data that the customer owns becomes the authoritative source for its pre-validated data. Valmiz™ turns that data into a company’s own knowledge base.
Martínez says his tech is currently in its “alpha phase.” He anticipates it to reach beta status in the end of the second quarter 2024, with an initial public release to follow shortly after.
In the meantime, you can watch Martínez on the newscast on Tuesday January 23, 2024 at 9am MT | 11am ET. He will also plug into an all-star AI panel at ’s 3rd Annual at AUVSI on April 22nd in San Diego, CA. And the folks in charge of the AI Asia Expo have already invited ASTN Group™ back for a reprise at their event in Thailand, scheduled for next August.
Dawn M.K. Zoldi (Colonel, USAF, Retired) is a licensed attorney with 28 years of combined active duty military and federal civil service to the U.S. Air Force. She is the CEO & Founder of P3 Tech Consulting and an internationally recognized expert on uncrewed aircraft system law and policy. Zoldi contributes to several magazines and hosts popular tech podcasts. Zoldi is also an Adjunct Professor for two universities, at the undergraduate and graduate levels. In 2022, she received the Airwards People’s Choice Industry Impactor Award, was recognized as one of the Top Women to Follow on LinkedIn and listed in the eVTOL Insights 2022 PowerBook. For more information, follow her on social media and visit her website at: https://www.p3techconsulting.com.
Miriam McNabb is the Editor-in-Chief of DRONELIFE and CEO of JobForDrones, a professional drone services marketplace, and a fascinated observer of the emerging drone industry and the regulatory environment for drones. Miriam has penned over 3,000 articles focused on the commercial drone space and is an international speaker and recognized figure in the industry. Miriam has a degree from the University of Chicago and over 20 years of experience in high tech sales and marketing for new technologies.
For drone industry consulting or writing, Email Miriam.
Subscribe to DroneLife here.