⚡ “They’ll raise my electricity bill.”

Only grid-connected ones can. Behind-the-meter facilities build their own power, completely disconnected from the public grid. Zero ratepayer impact. This distinction is important.

💧 “They consume massive water.”

Older evaporative cooling did. Modern closed-loop liquid cooling recirculates coolant with minimal consumption. The outrage targets technology most new builds no longer use.

🗳️ “It’s one political side.”

Both extremes. The Freedom Caucus and AOC are aligned: far right calls it overreach, far left calls it exploitation. We have 833 opposition groups across 49 states. 75 projects worth $130B blocked in Q1 2026. New York just imposed the first state moratorium.

🇨🇳 “It’s all grassroots.”

Not entirely. OpenAI’s June 2026 report documented Chinese accounts generating anti-data center content targeting US communities. The motive: slow American compute capacity. Legitimate local concerns and manufactured opposition both exist; conflating them hurts everyone.

📍 “Build them anywhere.”

AI inference needs low-latency fiber routes. Training needs massive reliable power. Existing infrastructure is a hard constraint.

💰 “Economic benefits are overstated.”

One facility can add $70M annual revenue to a $100M county budget. Thousands of construction jobs over 5-10 years plus multiplier effects. But only if we train the local workforce first.

🔒 “The industry has been transparent.”

In Virginia alone, 80% of localities with data centers have NDAs with developers that shut the public out of decisions shaping their electricity rates for decades (Public Citizen, 2026). Dominion projects residential bills could double by 2035 to subsidize data center infrastructure. Community benefit agreements covering hiring, traffic, noise, and emissions are the minimum.

🌍 “It’s just a tech issue.”

AI leadership is a compute race. Whoever builds the infrastructure controls the capability. 12,000 drones operated in a single area in the Russia-Ukraine conflict. In-Q-Tel (US intelligence venture fund) just took a board seat at a major developer. This is national security.

Brookings this week: “Americans don’t know how to fight AI. So they’re fighting data centers.” Data centers have become a proxy for broader public anxiety about AI’s effect on jobs and daily life. They’re the only physical thing people can point at and say no to.

Transparency and workforce training are very good, but until we address what’s really driving this debate, what AI means for people’s livelihoods and futures, no amount of good-neighbor agreements will be enough.

#DataCenters #AI #Infrastructure #NationalSecurity #EnergyPolicy

Coordinated cyberattacks hit water utilities across at least 7 US states. 30+ communities were disrupted — one plant went offline, others disconnected equipment and switched to manual operations. The FBI confirmed the attacks are spreading.

The likely path: internet-exposed PLCs connected via cellular modems, accessed using legitimate vendor engineering tools. Attackers modified controller logic, altered what operators saw on SCADA displays, and changed passwords to lock them out. No need to breach the corporate network — the field devices were directly reachable.

An AI-powered digital twin addresses this at multiple layers. It maintains a live knowledge graph of every asset, connection, and trust boundary — so an internet-exposed PLC isn’t an unknown risk, it’s a tagged, flagged, governed node. It detects anomalous telemetry patterns before operators are locked out. It computes blast radius: if this controller is compromised, what’s downstream? And it gives operators a simulation-backed response playbook rather than improvisation under pressure.

This is what our current work models— AI-powered digital twins for water system cybersecurity.

AI-Powered Digital Twins (Wiley, 2026)

🤝 Our CRM is a digital twin of customer relationships — pipeline, interactions, service history, account health.

☁️ Our SaaS stack — Jira, Slack, HubSpot, NetSuite, QuickBooks — each one is a digital twin of a specific business function.

🏭 Our SCADA/DCS is a digital twin of physical processes. Our BMS is a digital twin of the building. Our MES is a digital twin of the production floor. Our EMR is a digital twin of a patient.

They all do the same thing: digitized ONE domain ✅

And they all move the bottleneck to the same place ⬇️

From “we don’t have the data” to:

🔴 Humans must enter, maintain, and reconcile data manually

🔴 Rigid schemas that take years to implement

🔴 14 tools that don’t talk to each other

🔴 Data locked in proprietary systems

🔴 Systems that track what happened but can’t reason about what should happen next

The EHR is the most painful example — clinicians spend more time entering data than caring for patients.

What changed with AI?

💡 We were never meant to sit on laptops speaking machine. Entering data into forms. Writing SQL. Navigating rigid interfaces built for databases, not humans.

With AI, machine speaks human. This inverts the human-machine interface.

🏗️ An AI-powered digital twin eliminates these bottlenecks:

→ Semantic spine unifies meaning across all our systems — a Customer in Salesforce, an Account in SAP, and a Contact in HubSpot map to ONE canonical entity

→ MCP servers per platform expose standardized tools that agents discover at runtime

→ Event-native core preserves lineage across every system

→ GenAI removes the human data-entry constraint entirely — machines observe work and populate themselves

→ Natural language replaces SQL, forms, and rigid interfaces

Adding a 15th SaaS tool = writing one adapter. Not re-engineering the twin.

The pattern is the same across ALL of these systems: each digitized one domain, moved the bottleneck from missing data to fragmented data, and created silos that can’t reason across boundaries. They are all digital twins that predate the language 🔮 The architecture that connects them — unified meaning, standardized interfaces, machines that observe rather than require entry, reasoning across the boundaries these systems created — that’s what I put in my book, and it’s what we’re building now.

🛠️ My book AI-Powered Digital Twins (Wiley, 2026) is available (pre-release price point 😉)

#AI #DigitalTwins #ERP #CRM #SaaS #EnterpriseAI #DataArchitecture #FutureOfWork

* Tokens burned. Money spent.*

The product underneath was built before AI existed — adding a chat layer on top doesn’t change that.

Here’s the question we need to be asking: where does the intelligence actually live?

🧠 In the parameters of a language model — perceiving, reasoning over unstructured data, learning? This is expensive, powerful, and necessary... sometimes.

⚙️ In the knowledge graph and deterministic processes — formal equations, structured reasoning, gap analysis? No tokens burned. Works on edge. Works offline.

🤝 In a collaboration of both? The LLM perceives, the graph computes, results are governed together.

Most systems treat everything like it needs a language model. That’s like flying a helicopter to the mailbox 🚁📬

When you separate these clearly, your costs stay flat as usage grows. You can deploy on edge, on-premises, or cloud. And governance is enforced by the architecture — every path, every time.

The test: does your system know when to reason, think, and speak (Language Model) vs. when to reason and compute (graph) vs. when to do all? 🤔

Build the architecture first. Then AI has something meaningful to reason about.

* Your AI Doesn’t Need to Think That Hard

* Not Everything Deserves a Language Model

* The Most Expensive Lookup in History 🎭

If you’ve read up to here, photo: it’s my birthday weekend ;)

#AI #DigitalTwins #EnterpriseAI #AIGovernance #Architecture

I’d rather be the person who designed the architecture, built the system, and can tell where it breaks. 💡

If our AI can act — and AI can — then someone is liable when it acts wrong. The question is whether we’ve engineered the answer to “who approved this and why” into the system itself, or whether we’re going to scramble for it after the fact.

I wrote the book on the engineering architecture for how AI should operate in safety-critical systems. And right now I’m building it — deploying governed AI agents on a critical infrastructure facility that runs with unreliable satellite and no room for an AI to make an untracked decision. 🛠️

I think of governance as engineering, not policy. That means everything we say about AI governance has to work on hardware, not just look good on a slide.

For CEOs and leaders trying to figure out AI adoption — three decisions you can make this week, at zero cost:

🔥 Every AI action needs a human owner. Not a team. Not a department. A person whose name is on the approval. If nobody owns it, nobody governs it.

🔥 Define what AI cannot touch before you define what it can. Know what your data means and who it belongs to before AI gets near it (ontology-first). Customer data, production systems, legal records — draw the line first, open the door second.

🔥 Every AI interaction should produce a record that answers: who asked, what model responded, what it said, and when (immutable event log). If something goes wrong and you can’t trace the chain from decision to outcome, you don’t have governance — you have hope. Most enterprise AI tools already support this. You just have to turn it on.

One person’s authority, one boundary, one setting. And one question your board is going to ask whether you’re ready or not. 🎯

Now here’s the part nobody talks about:

Architecting a system and building it are two different skills. The architecture — 12 layers, ontology-first design, governed agents that are powerful but never sovereign — took a year from my brain. What I’m learning right now — Git workflows, IDE setup, localhost ports, model pulls — is implementation mechanics. 🔧

Important? Absolutely.

Learnable? In days, not years.

Most people are learning the tools without understanding the architecture. I prefer other way around. And my way scales better ;) 🤷‍♀️ Tools change every quarter. Architecture principles don’t.

That’s what ontology-first buys you: the intelligence is in the structure, not in any single tool.

.

Before writing a single line of code, I wrote the architecture. Every tool justified. Every decision documented. Every AI agent governed.

Here's my philosophy:

🧠 Define what matters before we start collecting data. Ontology first. If we don't know what we're looking for, the data is noise.

🤖 AI agents should be powerful but never sovereign. No agent configures its own tools. No agent escalates its own privileges. Human authority is non-negotiable.

🔒 Design for the constraint, not around it. If the system needs to work offline, start with offline and add connectivity as a bonus — not the other way around.

📦 If a tool can't justify its seat at the table, it gets cut. 23 tools made the stack. Each one has a "why" next to it. If I can't write the "why," the tool doesn't belong.

🐳 Simplicity is the hardest engineering. The entire production stack runs on one command on one machine. Getting there took weeks of decisions, not weeks of code.

The guide covers everything from "install your IDE" to "when to upgrade your hardware at scale." Built it so anyone walking into this project cold knows exactly what to do, in what order, and why.

Sometimes the most important engineering artifact is the document that makes the code inevitable, then the code, in that order. 💡

Retraining strategies vary by model type and operational constraints. Some models retrain continuously on streaming data, adapting to evolving patterns in near real-time. Others retrain on schedules—nightly, weekly, monthly—balancing freshness against computational costs. Critical models may maintain shadow deployments where new versions process production data without affecting outputs, enabling performance comparison before cutover. A/B testing exposes portions of traffic to candidate models, measuring improvement before full deployment.

Model tuning optimizes hyper-parameters and architectures for changing requirements. As computational budgets evolve, models can be scaled up for accuracy or down for efficiency. As new data modalities become available, models can incorporate additional inputs. As understanding deepens about what matters in predictions, model architectures can be refined to emphasize relevant features. This tuning represents continuous optimization along accuracy-efficiency-interpretability tradeoffs, adapting to shifting organizational priorities and technological capabilities.

Model replacement becomes necessary when architectural limitations prevent further optimization. A classical machine learning model may hit accuracy ceilings that deep learning surpasses. A deep learning model may lack the interpretability required for regulatory compliance, necessitating replacement with inherently transparent alternatives. Foundation model advances may make custom training obsolete for applications adequately served by prompted or fine-tuned general models. Replacement strategies must account for validation requirements—proving new models match or exceed predecessors—and rollback plans enabling quick reversion if replacements underperform or introduce unforeseen issues.

Photo today: Then and Now:

* Then- I knew only AI theory, foundational and super important.

* Now- I know integration, deployment, and business.

‘Then’ alone always felt like something big was missing... Not anymore ;)

#digitaltwins #ai #models #upgrades #newbook

Photo My BIG book cover REVEAL 🎉!

About the cover: I love sunset colors... 🌆 I love the twins... and the copper ;)

Wiley 2026. So soon...

In digital twins, this shift becomes concrete in several ways. Models increasingly require hardware-aware co-design, aligning algorithms with GPU parallelism, vectorized execution, and memory locality. Hybrid computational stacks combine physics-based solvers, optimization routines, and learned neural surrogates (usually neural networks that approximate the behavior of more expensive, physics-based or algorithmic models, but run much faster), each mapped to different hardware tiers.

Compute-aware modeling treats fidelity and resolution as adaptive choices rather than fixed parameters, while approximate inference replaces exact solutions with bounded, tractable approximations suitable for real-time operation. As a result, mathematical modeling is now inseparable from computational feasibility. Exhaustive computation was never truly practical; decades of hardware improvement merely postponed its consequences by masking poor algorithmic scaling. As gains from added compute diminish and architectural constraints dominate, digital twins must rely on representations that scale gracefully—preserving useful insight even when hardware can no longer compensate for computational inefficiency and poor design. The illusion of hardware improvements that Moore’s law provided for decades is now gone.

Come the era of quantum compute and model-system-compute co-design.

#newbook #math #computation #digitaltwins #enterprisemodeling #mooreslaw

Today’s Video: Summer and warmth must come back--H.

Fresh water systems are humanity’s most vital asset. As stress on these systems grows, AI-powered digital twins offer a valid solution providing real-time, data-driven insight for sustainable management. An AI-powered digital twin continuously integrates data from sensors, satellites, weather systems, usage logs, and public databases via APIs. This dynamic model reflects the current state of freshwater reserves and infrastructure, enabling accurate forecasting and rapid response. These twins blend traditional simulation models from operations research with AI models and agents that predict shortages, optimize allocations, and detect anomalies. Compliance rules are built in, ensuring water quality and usage align with regulations. Security measures, including encryption and threat detection, protect against tampering, while human oversight ensures responsible decision-making.
Most public data already enables highly accurate modeling of global water systems. When paired with localized consumption data—such as from data centers, agriculture, or municipalities—AI twins can recommend reallocation strategies and long-term preservation plans.
The strategic value is clear: if China offers hundreds of millions to acquire firms behind global AI-powered water twins, concerns over national security quickly follow. Control over freshwater intelligence is now a geopolitical asset.
AI-powered digital twins of fresh water systems offer visibility, resilience, and foresight. These are essential for securing the world’s most precious resource.

Photos today (Aug 2025): Jet lagged in Athens- Greece- 20 years and I still struggle with jet lag in this direction. It's brutal

To drive efficiencies, to make automated and streamlined decision making trustworthy and explainable, to leverage multiple integrated technologies, to advance AI, math, and digital twins technologies all at the same time.

You want to help? Let me know ;)
Photo: Ann Arbor- MI. At a best friend’s wedding

• I had a massive culture shock when I first moved to New York City (NYC) to pursue my PhD at New York University (NYU).

• I was extremely mathematically underprepared and found myself in one of the most prestigious schools in the country among the smartest students in the world. I wanted to live in New York City, so I applied to NYU. I had no idea at the time what being at the Courant Institute of Mathematical Sciences really meant.

• I had left a war and two engagements to fiancés I did not know in Lebanon, and I had a heavy religious baggage to shed, alongside an incredibly demanding PhD program.

• I got married during my third year of graduate school, defended my thesis at my fifth year, while pregnant.

• I moved with my then husband to Michigan, where both of us were offered postdoc positions at The University of Michigan - Ann Arbor.

• I had my daughter in Michigan, got divorced, my ex-husband went to live in Mexico, and I became a single mother for a very young child. I did not have family in Michigan, a permanent residency in the States, or a permanent job.

• I applied for very few tenure track positions at liberal arts schools, and I got a tenure track position at James Madison University (JMU). I moved to Virginia with my daughter when she was three.

• Right when I moved to Virginia a person very close to me died. Grieving their death while starting a new job in a new state with a small child was the most difficult thing I had gone through in my entire life.

• I went up early for tenure, and I got tenured, seven years after earning my PhD, four years after getting hired at JMU. I am now a full professor, a published author, working on the most stimulating projects in AI, data (all engineering, computation, applied math, business acumen, and building people relations, in my mind).

So how did it happen?

• I never listened to all the “no’s” I was told. No one knew my circumstances and what I was battling while being at my job. I was confident of what I was capable of, only when I got time to focus on it. I was very O.K. with not having time to focus on my job, I did not feel guilty. I had this deep confidence that I was playing the hand I was dealt in the best possible way. I did not try to compete or compare myself to other people at the same level of my career.

• I was put down more times than you can ever imagine, both personally and professionally: “You do not have the skills to pursue a job outside academia,” when I was contemplating leaving academia. “We want a professional to do this,” when I proposed doing a new project. And these were the nice put downs…

• I enjoyed my life a lot and valued what I had and what I had accomplished. I was living my dream, my most powerful persisting dream, the one I had since I was a little girl: to be free. I was free. No one knew what I had gone through to get to be free. So as amazing and attractive being a professor in math is, it’s nothing compared to finally living and being free, day in and day out.

• I kept on working on research projects even though I knew I did not have the time to finish or publish them. There is a seven year break from publishing on my resume. I could not afford to worry about that.

• I never subscribed to the publish or perish bad incentive machine in academia. I had my own version- publish nothing or something worthwhile. I also had my own comfortable timeline because why would we perish if we still have a brain and valuable training? Isn’t that a waste? Just because something is a long held principle doesn't mean it is true, or that it applies to all situations and times. Perhaps it is just old school.

• I got a job that valued my teaching, service, and was forgiving, at least for a bit, about my publishing break. When I was finally able to publish my projects, I went up for tenure and my application was successful.

• I applied for many small grants. When some got rejected I wrote to the grant directors and asked for feedback. I implemented their recommendations in my next proposals. I applied again. By the time I went up for tenure I had been able to secure over $90,000 in funding for various research and service projects. By the time I became a full professor, I had secured ~$365,000, which is a lot for a math professor in an R3 to R2 school. Note I had applied for more than $1million, so ~$700,000 grant applications were rejected, and that did not faze me to apply to other grants.

• I rarely was able to focus my brain on math in my pre-tenure years, not because I did not want to, but because raising my daughter and providing her with the best kind of security- emotional, financial, housing, and educational- were my priority. I homeschooled her too. I knew that I would get back to my job full force, however these formative years for her were never going to come back, and no career can substitute giving her a solid foundation in life. It’s also a fact that no child under the age of eight gives a single parent any free time, so it’s not like I had much of a choice.

• I documented everything I did in my career, as related to: teaching, research, service, grants, and mentoring students. This documentation came very handy when writing my promotion applications.

• I cared tremendously about my students’ success.

• I ran an Airbnb, but that is a different story.

• I felt grateful that I had a beautiful home, a beautiful daughter, amazing friends, a great sister who was incredible support via phone from Lebanon, a paycheck, and an adventurous life full of travels. I always felt that if being a math professor did not work out, it did not matter, since with what I have been able to overcome since I was so little has somehow prepared me to be anything I want to be- when I want it.

When you are born and raised in a war, when you experience various losses and hardships: forced engagements, culture shocks, graduate school, new beginnings over and over, changing countries, jobs and states, death, heartbreak, motherhood, single motherhood, and many other life events; getting tenure, or not getting tenure, promotion, or no promotion, acceptance, or rejection, becomes the least of your worries, because you deeply believe that your heart is strong, loving, and will survive.

I am now at a different place in my life and career, and I love it all. I am also more experienced and productive. I love my family, travels, books, projects, research, students, and service. My daughter is 13. My husband is awesome ❤️. I travel, take pictures, and write.

Photos: Birthday at Navy Yard in DC- Albi Restaurant- Michelin Star- Highly Recommend

They used the word mathematics, with its fancy formulas and plentiful theorems, both as guard to keep curious people away, and as a weapon to attack anyone who questions the assumptions underlying their models and the limitations to their predictive capabilities. Quantifying uncertainty, risk, and others, became a business that no one understands except the gatekeepers of knowledge, and whoever they accepted to their club, where the rules are: do not question the usefulness or faithfulness of these models to empirical reality. The same reality where markets crash, wars happen or do not end, and the gatekeepers create all kinds of awards to congratulate themselves- and win them. They create grants and award them to each other. The call for new ideas never ends, with incentivizing vocabulary like: revolutionary, transformative, disruptive,… but do they really want new ideas when that means relearning their entire fields, upending the whole foundation of their expert-less industries? Are they willing to at least communicate with honesty about the limitations of their trade? That most times, we do not know the rules of the game, and that made-up rules are just made-up?

I love math more than anything, and I believe that our universe was written in its language. Math is out there; we just happen to have the hardware and software to discover it. But nature didn’t reveal its own governing equations to us, or its underlying probability distributions from which all human behavior in all kinds of situations can be predicted. All that we can do is approximate, then hope that our approximations are not too far off.

My unconventional advise to you is be willing to unlearn, but if you are in too deep, then identify which bubble you are currently operating from: are you in a theoretical bubble, too far removed from empirical evidence? Are you purely experimental? Are you on the ground? Are you an executive? A strategist? An engineer? Are you looking for applications and use-cases for your work? Who are you? Now look at what people in other bubbles in your field are doing, and let the intersection of all, only if it remains appealing, guide the next thing to learn or do.

In the digital age with algorithmic networks and graphs ruling supreme, the world of old gatekeepers has changed. We now have the big markets of ideas and the big markets of data. Acquiring information, disseminating information, distilling experience into valuable knowledge, and building platforms follow different rules. And this is all good.

#connectivity #digitalage #transformation #gatekeeping #math #modeling #prediction #ai #data

Photo. At a beach on the Atlantic— will not locate because I love so many beaches in the world much more.