Alongside the many technological changes occurring in this era, robotics continues to evolve with new capabilities in artificial intelligence. Of those more recent advances and updates, generative AI took the lead, but there’s a new tech in town and it’s going “fully autonomous.”
In my Cloudspotter’s Journal columns, I have shown by example that the basics of next-generation cloud computing now include an advanced “cloud” infrastructure, shifting toward AI integration, edge computing and cross-cloud federation (coming up in October). Crucial elements include “AI-first architectures,” “agentic data systems” and “distributed cloud-to-edge” connections and their associated networking.
The importance of connecting the cloud to the world of AI cannot be underestimated. At first, it seems to be a bit far-fetched and beyond the reach of everyday use, but hold on—artificial intelligence is now “everywhere,” with a wider definition and an expanded dimension touching all walks of life. Most evident today is the explosion of data center activity popping up everywhere. Only a few see this as an important element of our tech future, but the data centers of today and tomorrow are “the cloud,” with compute power, storage and interconnection across the globe.
In the not-too-distant future, these new data centers will be self-managed and, in many cases, autonomous in nature. They will become the backbone for AI activities as their integration grows.
In this installment, we take a broad overview perspective of some new AI-related terms and how they generally apply to workflows, data centers and AI:
- Artificial Intelligence Integration: Based upon custom AI agents, robust LLM integration and AI-data engineering. As stated in Google Search, “AI integration is the process of embedding artificial intelligence into existing systems, workflows and applications to automate processes, generate insights and optimize performance.”
- AI Systems Integration: The fundamental elements associated with this widespread migration, modernization and optimization of services using a leaner stack are being accomplished with zero disruption. An “always-on digital workforce” is a key element, constructed as a unified platform built on at least three integrated layers: Data Quality, Agentic AI and Plain-Language Interfaces.
Data Quality (DQ)
Data Quality (DQ) encompasses elements of trust and Autonomous Data Engineering—the automation of jobs for data analysts, data engineers and operations (see workflows depicted in Fig. 1). DQ aims to improve the accuracy, completeness, consistency and reliability of the data used to train, test or run machine-learning models and AI systems. DQ generally includes model accuracy, which aims to ensure all AI outputs are trustworthy and to prevent error factors where poor data leads to biased or incorrect results.
DQ concepts include machine learning models that depend on “clean inputs”—free of duplicates, errors or misrepresentations, aka missing values or skewed anomalies.
Modern platforms use other AI tools to automate DQ checks and build validation rules that scale. Validation rules are those surrounded by an AI simulation of human intelligence in machines specifically programmed to think, learn and make decisions—i.e., utilizing building blocks typical to AI systems that include data (i.e., numbers, characters, media images—audio, video, etc.) and operations which are performed by using compute processes.
- Algorithm(s): Basically a sequence of calculations or rules used to solve a problem employing data that is “optimized in terms of time and space.” Such relative data should fit the time, place and application, and must provide suitable results that resolve the problem constructed of appropriate prompts with sufficient depth to properly answer the question “posed” of the associated content.
- Model: Sometimes referred to as an “agent,” a model is a combination of data and algorithms used to generate the response. Once you have a model, you can constantly provide it with new data and algorithms and continuously refine it. Fundamentally, the goal is to perform these actions autonomously.
- Response: Outputs generated as responses from models, otherwise known as the results.
- Ethics: The “moral principles” and “guidelines” from responses. These and related fundamental principles ensure that the responses (replies, reactions and/or outputs) from the AI systems “contribute to positive social, economic, and environmental impacts of the organization and the community.”
One significant application of AI is automating tasks that do not necessarily require human intervention during routine operations. This brings on some relatively new definitions for workflow and actions—e.g., idempotent, which means an operation that can be applied multiple times without changing the final result beyond the first time, and Customer Lifetime Value (CLV). When combined, idempotency and CLV can match analytics or calculations without double-counting revenue or corrupting historical cohorts—i.e. the use of past records to identify groups of people with or without a specific exposure.
The latest evolution in AI-driven automation is known as Agentic AI.
Agentic AI for RPA
Agentic AI is the next step in automation. Autonomous or semiautonomous artificial intelligence systems can independently plan, make decisions, use external tools and execute multistep workflows to achieve a specific goal with minimal human supervision. Robotic Process Automation (RPA) is rule-based, using software “bots” to mimic human actions. RPA automates repeated tasks (invoicing, inputting data, extraction and validation) which have since evolved with new AI capabilities. In principle, that evolution has stepped into various levels of agentic automation used to aid in determining the level of agents your organization may need; agentic workflow automation in action, or how those AI-powered agents perform; and how to scale agentic automation—responsibly and securely—while autonomously evaluating foundational skills aimed at end-to-end task completion.
The main types of automation (in robotics) include:
- Attended RPA: Bots that run on a user’s computer to help with live tasks like customer calls.
- Unattended RPA: Bots run on servers in the background to complete large batches of work automatically.
- Hybrid RPA: Blends both attended and unattended approaches so humans and bots can work together on complex jobs.
In addition to those RPA/human modes, those rational and easily manipulated sets of actions or instructions—including reporting and outputs—are orchestrated in plain language instead of software-driven expressions in specific forms of new or complex terms that must first be thoroughly learned and trained.
Plain Language
For an AI interaction to be “conversational,” the prompts and stages should repeatedly master the following steps in its sequences:
- Removes Jargon: Finds hard-to-read technical words, acronyms, or legalese and swaps them for everyday terms, with “audience identification” obviously being a prime focus on both output and delivery. For example, a response for a lawyer would be output differently for an administrator without a legal background.
- Shortens Sentences: Breaks down long, tangled sentences into short, direct thoughts aimed for a C-level executive set, TV/radio news reporter or professional-level manager.
- Active Voice: A sentence structure where the subject performs the verb’s action, following a clear pattern: actor, verb—targeting its searches for passive phrasing and changing it to clearly show who is doing what.
- Layout Improvement: Outputting clear headings and bullet points so text is easy to scan and reading for understanding is elevated smoothly and rapidly.
Assuring Data Quality
A part of maintaining and assuring Data Quality is documentation (especially if auto-generated). Docs are a foundational principle in assuring Data Quality and are extremely important when assembling any AI platform. Having a consistent and accurate reference set that includes the stages and steps discussed is essential to the AI system’s intelligence for long-term maintenance of the large model learning system, for both immediate and future or long-term support of the system overall. Instructions on documentation integrated with the “checks and values” portions of a system are essential.
Elements crucial to AI-developed outputs (Fig. 2), based on the prompts submitted and the audience you are addressing, include accuracy, where data matches real-world values and facts; completeness, with no essential fields or values missing; consistency, where all information matches across different systems, tables or equations; that data is fresh and timely, appropriate to the topics and audience and easily available when needed; and validity, where data follows defined (business) rules, formats and constraints, is unique and does not contain duplicate records.
Appropriate expectations are part of the primary requirements in qualifying the validity of an AI system and integration. These are some of the guidelines most solutions or outputs will produce with a properly utilized agentic AI or general practices as modeled across many levels of industry—providing a good checklist and process for relatively effective and useful autonomous applications and solutions.
