AI That Understands Reality, Not Just Words

You may already be using AI to draft emails, summarise documents, answer customer questions, or organise information. These tools are useful, but they mainly work with language. They predict patterns in text and help you communicate faster.

Many of your hardest business problems are different. You may need to reduce production waste, predict equipment failures, plan delivery routes during heavy rain, improve warehouse layouts, or understand why product quality changes between batches. These are physical problems involving time, temperature, movement, materials, and changing conditions. A chatbot alone cannot properly model all of that.

A recent development from Accelerated Understanding Inc shows where AI may be heading next. The company is building AI around physics and physical processes rather than language. Its founders say the system handled 5 trillion pieces of data in one prompt, a scale far beyond the typical context capacity described for leading language models.

TL;DR

Physics-based AI is designed to predict how real-world systems behave, including weather, machinery, materials, and movement.

For Malaysian SMEs, the practical lesson is to collect reliable operational data now, then apply AI first to one measurable process where prediction can improve decisions.

What This Means

Most popular generative AI systems are built around language. They learn from large collections of text, code, images, or other digital material and identify patterns. When you ask a question, the system generates a likely response based on those patterns.

Physics-based AI takes another approach. Instead of focusing mainly on what word comes next, it attempts to predict what will happen in a physical system over space and time. That might include how heat moves through a material, how a storm develops, how a machine behaves under pressure, or how a product changes during manufacturing.

The Reuters report describes a technology called neural operators. In simple terms, this approach is intended to learn relationships between physical conditions and outcomes. Rather than creating a separate mathematical model for every narrow situation, one adaptable system may eventually support multiple types of physical questions.

This does not mean the technology is ready to solve every business problem. It requires suitable data, powerful computing infrastructure, careful testing, and people who understand the business process. Predictions also need to be checked against real measurements. A confident-looking AI answer is not automatically a safe operating instruction.

The important shift is from asking AI what to say to asking AI what will happen.

How This Applies to Malaysian SMEs

Manufacturing businesses can use the idea to improve process control. If you operate a factory making food products, plastics, metal parts, furniture, or electronics components, your output may depend on temperature, humidity, machine speed, material quality, and timing. Today, staff may adjust settings based on experience and manual checks. A future physics-aware system could identify which conditions are linked to defects and predict the settings most likely to produce consistent results. You can begin with a simpler version by recording machine readings, batch information, defects, and corrective actions in one structured system.

Distributors and retailers can apply physical prediction to inventory and delivery planning. A delivery business does not only deal with addresses. It deals with traffic, road conditions, weather, vehicle capacity, loading sequence, driver availability, and product sensitivity. The Malaysian Meteorological Department provides weather information and warnings through its official services, while your own historical delivery records can show how rain affects travel time in particular areas. Combining these sources could help you plan dispatch windows and inform customers earlier when conditions are likely to cause delays. The first step is to record planned arrival time, actual arrival time, route, weather condition, and reason for delay.

Food, agriculture, and cold-chain SMEs have especially clear use cases. A frozen-food distributor, seafood supplier, nursery, or farm may need to manage temperature and humidity throughout storage and transport. A physics-oriented AI system could eventually model how long products remain within acceptable conditions under different loading, insulation, and travel scenarios. Until that technology becomes accessible, you can still improve readiness using sensors and automated alerts. Record temperature continuously, connect exceptions to delivery and product-quality records, and investigate repeated patterns rather than relying on memory.

Engineering, construction, and maintenance firms can use prediction to reduce disruption. Pumps, compressors, generators, air-conditioning systems, and production equipment often show signs of stress before failure. Vibration, heat, pressure, operating hours, and power consumption can provide useful clues. You do not need to build an advanced model immediately. Start by listing critical equipment, recording inspections digitally, and noting the exact symptoms that appeared before a breakdown. This creates the foundation for predictive maintenance and makes technician knowledge easier to preserve.

Service businesses can also benefit, even without owning machinery. A laundrette, commercial kitchen, car workshop, printing company, or cleaning contractor has physical workflows involving capacity, queue times, drying, curing, storage, and movement. A simple operational model can reveal bottlenecks. For example, you may discover that jobs pile up not because demand is too high, but because one preparation step, machine, or approval stage limits the entire process.

A Simple Data Foundation for Your Business

Business area	Useful data to record	Possible prediction
Production	Batch, settings, material, defect type, output	Risk of defects or rework
Delivery	Route, departure, arrival, weather, vehicle	Likely delay and arrival window
Equipment	Runtime, temperature, vibration, service history	Maintenance requirement
Cold chain	Temperature, location, loading time, delivery time	Product-condition risk
Energy use	Meter reading, operating hours, machine status	Unusual consumption or operating patterns

Practical Takeaways

• Choose one physical process first. Pick a process where delays, defects, downtime, or wastage are already visible.
• Define one measurable outcome. Examples include fewer late deliveries, fewer rejected batches, shorter repair response time, or more accurate completion estimates.
• Collect consistent records. Use the same names, units, timestamps, and categories every time. Inconsistent data will weaken any future AI system.
• Capture conditions, not only results. Record temperature, humidity, machine settings, staffing, weather, material source, and other factors that may influence the outcome.
• Keep human approval in the loop. Use predictions to support supervisors and technicians, not to remove responsibility for safety-critical decisions.
• Connect operational systems. Your sales, inventory, delivery, service, and accounting information should not remain isolated if they affect the same physical workflow.
• Test against history. Before acting on a model, compare its predictions with previous outcomes and document where it was wrong.
• Protect sensitive data. Control access to customer information, production formulas, supplier details, and equipment records.

What You Should Not Expect Yet

Physics-based AI is not a replacement for a qualified engineer, technician, safety officer, or operations manager. It may produce useful estimates, but real-world conditions can change suddenly. Sensors can fail, data can be incomplete, and a model trained on one machine may not work properly on another.

You should also avoid buying a complicated system simply because it uses advanced AI terminology. Ask the provider what data is required, how predictions are tested, what happens when information is missing, and how your staff can challenge an incorrect result. A useful system should fit your workflow, not force your team to create unnecessary administrative work.

The Bigger Picture

The longer-term direction is clear: AI is moving beyond documents and conversations towards operational systems that observe, predict, and recommend actions in the physical world. The Reuters report points to potential applications in chip design, robotics, extreme-weather prediction, and geological analysis. These are technically demanding fields, but the underlying principle also applies to smaller businesses: better predictions come from understanding the conditions that create results.

For Malaysian SMEs, this creates an opportunity to compete through operational discipline. You do not need a huge research team to prepare. You need clean records, clearly defined processes, reliable measurements, and a willingness to test small improvements. A company that knows why delays happen, which conditions create defects, and when equipment begins to behave unusually will be better prepared for advanced automation.

The best starting point is not asking, “How can we use the latest AI?” Ask instead, “Which physical business problem keeps repeating, and what evidence would help us predict it earlier?” That question can guide a practical automation project today while preparing your business for more capable AI systems tomorrow.

Ready to Streamline Your Operations?

Your business should run itself. AutoRunBiz deploys AI agents to automate your daily operations — WhatsApp orders, invoicing, customer follow-ups, and accounting. Book a free 15-min ops audit to see where automation fits your business →