<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0"><channel><title><![CDATA[Nitin Kumar K]]></title><description><![CDATA[Nitin Kumar K]]></description><link>https://nitinkk.hashnode.dev</link><generator>RSS for Node</generator><lastBuildDate>Sun, 22 Nov 2026 15:33:00 GMT</lastBuildDate><atom:link href="https://nitinkk.hashnode.dev/rss.xml" rel="self" type="application/rss+xml"/><language><![CDATA[en]]></language><ttl>60</ttl><item><title><![CDATA[Internet of Things (IoT): Connecting the Physical World with Digital Intelligence]]></title><description><![CDATA[How IoT works: architecture, sensors, MQTT and other protocols, cloud and edge computing, real-world uses, AI and security.]]></description><link>https://nitinkk.hashnode.dev/internet-of-things-iot-connecting-the-physical-world-with-digital-intelligence</link><guid isPermaLink="true">https://nitinkk.hashnode.dev/internet-of-things-iot-connecting-the-physical-world-with-digital-intelligence</guid><category><![CDATA[Internet of Things]]></category><category><![CDATA[iot]]></category><category><![CDATA[Cloud Computing]]></category><category><![CDATA[technology]]></category><dc:creator><![CDATA[Nitin Kumar K]]></dc:creator><pubDate>Mon, 09 Nov 2026 06:22:53 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af1583b68fd16fccf9bbafa/f3179ce6-497e-48ba-af23-c550dc1e6210.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>A thermostat that learns your schedule, a soil sensor that tells a farmer when to irrigate, a factory machine that reports its own vibration before it fails: all of these are examples of the Internet of Things (IoT). Computers once sat on desks and the internet connected people. IoT extends the network to physical objects, so the real world can be measured, reported on and, increasingly, controlled by software.</p>
<p>This article explains how that works from the ground up. We will cover IoT architecture, the hardware that makes up a connected device, the protocols devices use to talk, the role of cloud and edge computing, real-world applications, the link with artificial intelligence, and the security challenges that every IoT builder has to face. Along the way you will follow one running example, a smart greenhouse, to see how the pieces fit together.</p>
<h2>What is the Internet of Things?</h2>
<p>The Internet of Things is a network of physical objects ("things") that contain sensors, software and connectivity, so they can collect data, exchange it with other systems and sometimes act on it. A "thing" can be as small as a temperature sensor with a battery, or as large as a wind turbine.</p>
<p>The evolution is easy to trace. First came standalone embedded systems, such as microcontrollers inside appliances that did one job and never talked to anyone. Then those devices gained network connections, so their data could be collected. Next, cloud platforms made it practical to store and analyse that data at scale. Today, processing is moving back toward the devices themselves, which is the edge computing trend we cover later.</p>
<p>At its simplest, every IoT system performs the same loop:</p>
<ol>
<li><strong>Sense</strong>: a sensor measures something in the physical world.</li>
<li><strong>Communicate</strong>: the reading travels over a network.</li>
<li><strong>Process</strong>: software stores it, analyses it and decides what to do.</li>
<li><strong>Act</strong>: a person is alerted, or an actuator changes something in the physical world.</li>
</ol>
<p>Everyday examples include smart speakers, fitness bands, connected door locks, smart meters, parcel trackers and the sensors in a modern car.</p>
<h2>Understanding IoT Architecture</h2>
<p>There is no single standard diagram. Textbooks use three-, four- or five-layer models, and cloud vendors draw their own. Microsoft, for example, describes two broad patterns: a <em>cloud-connected</em> pattern, where devices talk directly to cloud services, and an <em>edge-connected</em> pattern, where devices connect to a local edge environment that processes data before optionally forwarding it to the cloud [1]. A four-layer model is a good way to learn the concepts:</p>
<img src="https://cdn.hashnode.com/uploads/covers/6af1583b68fd16fccf9bbafa/a4bcabcc-61a0-426a-8ed4-abdb6f7dc059.png" alt="Four-layer IoT architecture: perception, network, processing and application layers" style="display:block;margin:0 auto" />

<table>
<thead>
<tr>
<th>Layer</th>
<th>Job</th>
<th>Typical technologies</th>
</tr>
</thead>
<tbody><tr>
<td><strong>Perception layer</strong></td>
<td>Senses and acts on the physical world</td>
<td>Sensors, actuators, microcontrollers</td>
</tr>
<tr>
<td><strong>Network (communication) layer</strong></td>
<td>Moves data between devices, gateways and the cloud</td>
<td>Wi-Fi, Bluetooth LE, LoRaWAN, MQTT, CoAP, HTTPS</td>
</tr>
<tr>
<td><strong>Processing layer</strong></td>
<td>Stores, filters and analyses data</td>
<td>Edge/fog nodes, cloud IoT platforms, databases, analytics</td>
</tr>
<tr>
<td><strong>Application layer</strong></td>
<td>Turns results into something people use</td>
<td>Dashboards, mobile apps, alerts, automation rules</td>
</tr>
</tbody></table>
<p>In our greenhouse: soil-moisture and temperature sensors are the perception layer, a low-power radio link forms the network layer, a cloud service stores and analyses readings in the processing layer, and a dashboard that tells the grower when to water is the application layer.</p>
<h2>Essential Components of IoT</h2>
<ul>
<li><strong>Sensors</strong> convert a physical quantity (temperature, humidity, light, motion, pressure, flow) into an electrical signal.</li>
<li><strong>Actuators</strong> do the reverse: they turn a command into physical action, such as opening a valve, switching a relay or moving a motor.</li>
<li><strong>Microcontrollers (MCUs)</strong> are small, low-power chips that read sensors, run firmware and drive actuators. Microsoft's IoT guidance distinguishes constrained devices built around an MCU from standard computing devices with a CPU or microprocessor [1].</li>
<li><strong>Embedded systems</strong> are the combination of hardware and dedicated software (firmware) inside a device, built for a specific function.</li>
<li><strong>Gateways</strong> sit between local devices and the wider network. They translate protocols, aggregate data, and can process it locally. Microsoft describes devices that connect through an edge-based gateway or through a protocol-specific connector such as OPC UA for industrial equipment [1].</li>
<li><strong>Cloud platforms</strong> provide device management, secure messaging, storage and analytics. Examples include AWS IoT Core [2] and Azure IoT Hub [1].</li>
</ul>
<h2>Communication Protocols</h2>
<p>It helps to separate two kinds of technology that are often listed together. <strong>Application protocols</strong> (MQTT, HTTP/HTTPS, CoAP) define how messages are structured and exchanged. <strong>Connectivity technologies</strong> (Wi-Fi, Bluetooth Low Energy, LoRaWAN) define how bits travel over the air. A real device combines one of each.</p>
<h3>Application protocols</h3>
<p><strong>MQTT</strong> (Message Queuing Telemetry Transport) is a lightweight publish/subscribe protocol. Devices publish messages to named <em>topics</em> on a central <em>broker</em>, and any client subscribed to a topic receives them. The OASIS standard describes it as simple, lightweight and suited to constrained environments where bandwidth is at a premium [3]. MQTT defines three quality-of-service levels (at most once, at least once, exactly once). Note that managed services may support only part of the specification: AWS IoT Core, for instance, supports MQTT 3.1.1 and 5.0 but only QoS levels 0 and 1 [2].</p>
<p><strong>HTTP/HTTPS</strong> is the protocol of the web. It is familiar, works through almost any firewall and is easy to debug, but its request/response style and larger headers make it heavier for tiny battery-powered devices. It is a good fit for devices that report rarely or that talk to existing web APIs.</p>
<p><strong>CoAP</strong> (Constrained Application Protocol), specified in RFC 7252, is a web-style protocol designed for constrained nodes and lossy networks. It keeps the request/response model and URIs of the web, adds built-in resource discovery, and maps easily to HTTP [4]. It typically runs over UDP, and the specification defines security using DTLS.</p>
<h3>Connectivity technologies</h3>
<ul>
<li><strong>Wi-Fi</strong> offers high throughput and uses existing infrastructure, which suits smart-home devices that have mains power. It is relatively power-hungry for coin-cell sensors.</li>
<li><strong>Bluetooth Low Energy (BLE)</strong> is designed for very low-power operation, using 40 channels of 2 MHz in the 2.4 GHz band and supporting point-to-point, broadcast and mesh topologies [5]. It is common in wearables, beacons and phone-connected devices.</li>
<li><strong>LoRaWAN</strong> is a low-power, wide-area (LPWAN) protocol for battery-operated devices. LoRa is the radio signal that carries data over the air (the physical layer), while LoRaWAN is the network standard on top of it that manages communication, data rates and security [6]. It suits small, infrequent messages over long distances, such as farm sensors, and it has been recognised by the ITU as Recommendation Y.4480, "Low power protocol for wide area wireless networks" [7].</li>
</ul>
<table>
<thead>
<tr>
<th>Technology</th>
<th>Level</th>
<th>Strength</th>
<th>Trade-off</th>
<th>Typical use</th>
</tr>
</thead>
<tbody><tr>
<td>MQTT</td>
<td>Application</td>
<td>Lightweight pub/sub, many-to-many</td>
<td>Needs a broker</td>
<td>Telemetry to cloud</td>
</tr>
<tr>
<td>HTTP/HTTPS</td>
<td>Application</td>
<td>Universal, easy tooling</td>
<td>Heavier per message</td>
<td>Web APIs, infrequent reports</td>
</tr>
<tr>
<td>CoAP</td>
<td>Application</td>
<td>Small footprint, REST-style</td>
<td>UDP reliability handled by the protocol</td>
<td>Constrained sensors</td>
</tr>
<tr>
<td>Wi-Fi</td>
<td>Connectivity</td>
<td>Speed, existing networks</td>
<td>Power use</td>
<td>Smart home, cameras</td>
</tr>
<tr>
<td>BLE</td>
<td>Connectivity</td>
<td>Very low power, phone-friendly</td>
<td>Short-to-medium range</td>
<td>Wearables, beacons</td>
</tr>
<tr>
<td>LoRaWAN</td>
<td>Connectivity</td>
<td>Long range, long battery life</td>
<td>Low data rate</td>
<td>Agriculture, metering</td>
</tr>
</tbody></table>
<h3>A minimal MQTT example</h3>
<p>This Python sketch simulates our greenhouse sensor publishing a reading over a TLS-protected MQTT connection (it uses the <code>paho-mqtt</code> library, and the broker address and credentials are placeholders):</p>
<pre><code class="language-python">import json, random, time
import paho.mqtt.client as mqtt

client = mqtt.Client(mqtt.CallbackAPIVersion.VERSION2,
                     client_id="greenhouse-sensor-01")
client.tls_set()                                   # encrypt traffic with TLS
client.username_pw_set("device-01", "&lt;token&gt;")     # placeholder credentials
client.connect("broker.example.com", 8883)
client.loop_start()

while True:
    reading = {"temperature_c": round(random.uniform(22, 30), 1),  # simulated sensor
               "ts": int(time.time())}
    client.publish("farm/greenhouse1/telemetry", json.dumps(reading), qos=1)
    time.sleep(10)
</code></pre>
<p>A dashboard or analytics service subscribes to <code>farm/greenhouse1/telemetry</code> and receives every reading, without the sensor needing to know who is listening.</p>
<h2>The Role of Cloud Computing in IoT</h2>
<p>IoT devices are small and numerous; the cloud is large and elastic. Microsoft describes the typical responsibilities of IoT cloud services as receiving sensor data at scale, sending commands to specific devices, provisioning devices and controlling which may connect, monitoring device state, and managing firmware [1]. In practice that breaks down into:</p>
<ul>
<li><strong>Data collection</strong>: a managed message broker, such as AWS IoT Core's MQTT support, accepts connections from many devices at once [2].</li>
<li><strong>Storage</strong>: time-series databases and data lakes hold raw and processed readings.</li>
<li><strong>Processing</strong>: rules and stream-processing jobs filter data, trigger alerts and route messages.</li>
<li><strong>Real-time monitoring</strong>: dashboards show live device state and flag offline or abnormal devices.</li>
<li><strong>Analytics</strong>: historical data is aggregated and modelled to find trends and forecast.</li>
</ul>
<p>Microsoft describes a progressive pipeline in which raw telemetry becomes structured, then contextualised (adding information such as location and asset type), then analysis-ready, and finally ready for AI [1]. That context step matters: a temperature number is far more useful when the system knows which greenhouse and which crop it belongs to.</p>
<h2>Edge Computing and Fog Computing</h2>
<p>Sending every reading to a distant data centre has costs: latency, bandwidth and dependence on a working internet connection. <strong>Edge computing</strong> moves processing close to where data is produced, on the device itself or on a nearby gateway. <strong>Fog computing</strong> is a related idea. NIST's conceptual model describes fog as a resource layer that sits between smart end-devices and the traditional cloud [8]. NIST also notes that no consensus exists on how fog, mist, cloudlet and edge computing differ, so treat the terms as overlapping rather than strict categories [8].</p>
<img src="https://cdn.hashnode.com/uploads/covers/6af1583b68fd16fccf9bbafa/eb1921b7-2ac6-411d-b403-13d03dc9b323.png" alt="Diagram of devices, edge gateway, fog layer and cloud, with latency increasing toward the cloud" style="display:block;margin:0 auto" />

<table>
<thead>
<tr>
<th></th>
<th>Edge</th>
<th>Fog</th>
<th>Cloud</th>
</tr>
</thead>
<tbody><tr>
<td><strong>Location</strong></td>
<td>On or next to the device</td>
<td>Between the edge and the cloud</td>
<td>Remote data centre</td>
</tr>
<tr>
<td><strong>Latency</strong></td>
<td>Lowest</td>
<td>Low</td>
<td>Highest</td>
</tr>
<tr>
<td><strong>Compute capacity</strong></td>
<td>Smallest</td>
<td>Moderate</td>
<td>Largest</td>
</tr>
<tr>
<td><strong>Works offline</strong></td>
<td>Yes</td>
<td>Often</td>
<td>No</td>
</tr>
<tr>
<td><strong>Best for</strong></td>
<td>Instant reactions, filtering</td>
<td>Local aggregation across sites</td>
<td>Heavy analytics, long-term storage</td>
</tr>
</tbody></table>
<p>Platforms support this directly. AWS IoT Greengrass is an open-source edge runtime that brings cloud logic to edge devices, can operate with intermittent connectivity, and lets devices transmit only high-value data [9]. Microsoft's Azure IoT Operations can run AI inference on the edge cluster for response times measured in milliseconds, in scenarios such as quality inspection and safety monitoring [1]. In the greenhouse, an edge gateway can close a vent immediately when temperature spikes, while sending only summarised readings to the cloud.</p>
<h2>Real-world Applications</h2>
<ul>
<li><strong>Smart homes</strong>: thermostats, lighting, locks and energy monitors that respond to schedules and occupancy.</li>
<li><strong>Smart cities</strong>: connected street lighting, parking sensors, traffic management and waste-collection monitoring.</li>
<li><strong>Healthcare</strong>: wearables and connected monitors that stream patient measurements, with clinical oversight. BLE has standard health profiles for this kind of data.</li>
<li><strong>Industrial IoT (IIoT)</strong>: machine monitoring, asset tracking and quality control on factory floors, often over industrial protocols such as OPC UA through edge gateways [1].</li>
<li><strong>Smart agriculture</strong>: soil-moisture, weather and livestock sensors, often over LoRaWAN, to guide irrigation and feeding.</li>
<li><strong>Environmental monitoring</strong>: air-quality, noise and weather stations, and sensors in forests or rivers, where devices run for long periods on batteries in remote places.</li>
<li><strong>Water management</strong>: flow and pressure sensors to detect leaks, track consumption and monitor reservoir and treatment-plant conditions.</li>
</ul>
<h2>IoT and Artificial Intelligence</h2>
<p>When IoT supplies the data and AI supplies the interpretation, the combination is often called <strong>AIoT</strong>. IBM notes that machine-learning models can analyse data from sensors, IoT devices and operational technology to forecast when maintenance will be required and predict equipment failures before they occur [10].</p>
<ul>
<li><strong>Predictive maintenance</strong>: models learn what "normal" vibration, temperature or current looks like for a machine and flag drifts that precede failure. This can reduce unplanned downtime.</li>
<li><strong>Intelligent automation</strong>: instead of fixed thresholds ("water if moisture &lt; 30%"), a model can weigh weather forecasts, crop stage and past results.</li>
<li><strong>Data-driven decisions</strong>: aggregated data across many sites reveals patterns that no single sensor shows.</li>
</ul>
<p>AI can run in two places. Microsoft distinguishes edge AI, which delivers fast local responses, from cloud AI, which handles large-scale analysis and model training [1]. A common pattern is to train models in the cloud and deploy compact versions to the edge for real-time inference.</p>
<h2>Security and Privacy Challenges</h2>
<p>IoT devices are attractive targets: they are numerous, often unattended, physically accessible and built on tight cost and power budgets. The OWASP IoT Top 10 (2018 edition) lists the recurring problems [11]:</p>
<ol>
<li>Weak, guessable or hardcoded passwords</li>
<li>Insecure network services</li>
<li>Insecure ecosystem interfaces</li>
<li>Lack of secure update mechanism</li>
<li>Use of insecure or outdated components</li>
<li>Insufficient privacy protection</li>
<li>Insecure data transfer and storage</li>
<li>Lack of device management</li>
<li>Insecure default settings</li>
<li>Lack of physical hardening</li>
</ol>
<p>The main defences map onto this list:</p>
<ul>
<li><strong>Device authentication.</strong> Give each device a unique identity, for example its own X.509 certificate rather than a shared password. Cloud platforms such as AWS IoT Core are built around per-device identities of this kind.</li>
<li><strong>Encryption.</strong> Protect data in transit with TLS (or DTLS for CoAP) and protect sensitive stored data. This also addresses privacy, since sensor data can reveal when people are home.</li>
<li><strong>Network vulnerabilities.</strong> Disable unused services, segment IoT devices from the rest of the network and avoid default credentials.</li>
<li><strong>Data privacy.</strong> Collect only what you need, define retention limits and follow the privacy laws that apply where your users are.</li>
<li><strong>Secure firmware updates.</strong> RFC 9019 sets out an architecture for firmware updates on constrained devices, noting that updating is essential for fixing vulnerabilities, and calls for authentication and integrity protection of firmware images [12]. Automatic updates that need no human intervention are key to fixing flaws at scale [12].</li>
</ul>
<p>NIST's IoT Device Cybersecurity Capability Core Baseline (NISTIR 8259A) gives manufacturers a starting list of device capabilities, including device identification, device configuration, data protection, logical access to interfaces, software update and cybersecurity state awareness [13]. If you build or buy IoT products, it makes a useful checklist.</p>
<h2>The Future of IoT</h2>
<p>Several trends are already visible:</p>
<ul>
<li><strong>More intelligence at the edge</strong>, as smaller models and better hardware allow inference on gateways and even microcontrollers.</li>
<li><strong>Hybrid architectures</strong> that combine cloud and edge. Microsoft's current guidance frames them as complementary patterns, and many solutions combine both [1].</li>
<li><strong>Digital twins</strong>, which are live software models of physical assets used to monitor and simulate them. Cloud platforms now offer them as part of their IoT portfolios [1].</li>
<li><strong>Better security baselines and update practices</strong>, driven by standards work such as NIST's and the IETF's [12][13].</li>
<li><strong>Continued use of low-power networking</strong>, such as LoRaWAN and Bluetooth mesh, for sensing in agriculture, utilities and buildings [5][6].</li>
</ul>
<p>The open question is less about the technology than about trust: devices that are secure by default, updateable for their whole life, and transparent about the data they collect.</p>
<h2>Conclusion</h2>
<p>IoT connects the physical and digital worlds through a repeating loop of sensing, communicating, processing and acting. Sensors and microcontrollers form the perception layer; protocols such as MQTT and CoAP, carried over Wi-Fi, BLE or LoRaWAN, move the data; cloud and edge platforms process it; and applications turn it into decisions. AI makes the data more valuable by finding patterns and predicting failures, while security and privacy determine whether the system can be trusted.</p>
<p>A good way to start learning is to build the greenhouse example yourself: a microcontroller board, a temperature sensor, an MQTT broker and a simple dashboard. Then explore the references below.</p>
<h2>References and Further Reading</h2>
<p>All links were last accessed on 9 November 2026.</p>
<ol>
<li>Microsoft Learn. <em>What is Azure IoT?</em> (updated 15 April 2026). <a href="https://learn.microsoft.com/en-us/azure/iot/iot-introduction">https://learn.microsoft.com/en-us/azure/iot/iot-introduction</a></li>
<li>Amazon Web Services. <em>MQTT</em> (AWS IoT Core Developer Guide). <a href="https://docs.aws.amazon.com/iot/latest/developerguide/mqtt.html">https://docs.aws.amazon.com/iot/latest/developerguide/mqtt.html</a></li>
<li>OASIS. <em>MQTT Version 5.0</em>, OASIS Standard, 7 March 2019. <a href="https://docs.oasis-open.org/mqtt/mqtt/v5.0/mqtt-v5.0.html">https://docs.oasis-open.org/mqtt/mqtt/v5.0/mqtt-v5.0.html</a></li>
<li>Shelby, Z., Hartke, K. and Bormann, C. <em>The Constrained Application Protocol (CoAP)</em>, RFC 7252, IETF, June 2014. <a href="https://www.rfc-editor.org/rfc/rfc7252">https://www.rfc-editor.org/rfc/rfc7252</a></li>
<li>Bluetooth SIG. <em>Bluetooth Technology Overview.</em> <a href="https://www.bluetooth.com/learn-about-bluetooth/tech-overview/">https://www.bluetooth.com/learn-about-bluetooth/tech-overview/</a></li>
<li>LoRa Alliance. <em>About LoRaWAN.</em> <a href="https://lora-alliance.org/about-lorawan/">https://lora-alliance.org/about-lorawan/</a></li>
<li>ITU-T. <em>Recommendation Y.4480: Low power protocol for wide area wireless networks</em>, November 2021. <a href="https://www.itu.int/dms_pubrec/itu-t/rec/y/T-REC-Y.4480-202111-I!!SUM-HTM-E.htm">https://www.itu.int/dms_pubrec/itu-t/rec/y/T-REC-Y.4480-202111-I!!SUM-HTM-E.htm</a></li>
<li>Iorga, M. et al. <em>Fog Computing Conceptual Model</em>, NIST Special Publication 500-325, March 2018. <a href="https://doi.org/10.6028/NIST.SP.500-325">https://doi.org/10.6028/NIST.SP.500-325</a></li>
<li>Amazon Web Services. <em>What is AWS IoT Greengrass?</em> <a href="https://docs.aws.amazon.com/greengrass/v2/developerguide/what-is-iot-greengrass.html">https://docs.aws.amazon.com/greengrass/v2/developerguide/what-is-iot-greengrass.html</a> and <em>AWS IoT Greengrass</em> product page, <a href="https://aws.amazon.com/greengrass/">https://aws.amazon.com/greengrass/</a></li>
<li>Stryker, C. and Kavlakoglu, E. <em>What is artificial intelligence (AI)?</em> IBM Think (predictive maintenance use case). <a href="https://www.ibm.com/think/topics/artificial-intelligence">https://www.ibm.com/think/topics/artificial-intelligence</a></li>
<li>OWASP Foundation. <em>OWASP Internet of Things Project</em> (lists the OWASP IoT Top 10, 2018). <a href="https://owasp.org/projects/internet-of-things">https://owasp.org/projects/internet-of-things</a></li>
<li>Moran, B. et al. <em>A Firmware Update Architecture for Internet of Things</em>, RFC 9019, IETF, April 2021. <a href="https://www.rfc-editor.org/rfc/rfc9019">https://www.rfc-editor.org/rfc/rfc9019</a></li>
<li>Fagan, M. et al. <em>IoT Device Cybersecurity Capability Core Baseline</em>, NISTIR 8259A, NIST, May 2020. <a href="https://doi.org/10.6028/NIST.IR.8259A">https://doi.org/10.6028/NIST.IR.8259A</a></li>
</ol>
]]></content:encoded></item><item><title><![CDATA[Artificial Intelligence: Understanding the Technology Transforming Our World]]></title><description><![CDATA[A beginner-friendly guide to AI: how it works, machine learning, deep learning, generative AI, real-world uses, risks and the future.]]></description><link>https://nitinkk.hashnode.dev/artificial-intelligence-understanding-the-technology-transforming-our-world</link><guid isPermaLink="true">https://nitinkk.hashnode.dev/artificial-intelligence-understanding-the-technology-transforming-our-world</guid><category><![CDATA[Artificial Intelligence]]></category><category><![CDATA[Machine Learning]]></category><category><![CDATA[Deep Learning]]></category><category><![CDATA[technology]]></category><dc:creator><![CDATA[Nitin Kumar K]]></dc:creator><pubDate>Mon, 09 Nov 2026 06:21:26 GMT</pubDate><enclosure url="https://cdn.hashnode.com/uploads/covers/6af1583b68fd16fccf9bbafa/bff03f9b-b17c-499b-95fe-d1f292e9c4e6.png" length="0" type="image/jpeg"/><content:encoded><![CDATA[<p>Artificial intelligence (AI) has moved out of research labs and into the tools most of us use every day: the keyboard that predicts your next word, the map that reroutes around traffic, the assistant that drafts an email. Behind these features sits a small set of ideas, such as learning from data, finding patterns, and generating new content, that are easier to grasp than the headlines suggest.</p>
<p>This guide walks through those ideas from the ground up. You will learn what AI actually is, how it differs from ordinary software, which technologies power it, how tools like ChatGPT work at a high level, where AI is used today, and what its limits and risks are. Each claim is tied to a primary source listed at the end, so you can check it and keep reading.</p>
<h2>What is Artificial Intelligence?</h2>
<h3>A working definition</h3>
<p>The computer scientist John McCarthy, who is credited with coining the term "artificial intelligence" in 1956 [2], described AI as "the science and engineering of making intelligent machines, especially intelligent computer programs" [1]. IBM's current explainer gives a more applied definition: technology that lets computers and machines simulate human learning, comprehension, problem solving, decision making, creativity and autonomy [2].</p>
<p>Notice what these definitions leave out. AI is not one product or one algorithm. It is a field, and McCarthy himself stressed that intelligence is not a single thing you either have or lack; machines can handle some mechanisms of intelligence and not others [1].</p>
<h3>How an AI system works</h3>
<p>Most modern AI systems follow the same loop:</p>
<ol>
<li><strong>Collect data.</strong> Examples of the thing you want the system to handle, such as emails, X-ray images or sensor readings.</li>
<li><strong>Train a model.</strong> An algorithm adjusts the internal parameters of a mathematical model until its outputs match the examples well.</li>
<li><strong>Run inference.</strong> The trained model is given new, unseen input and produces a prediction, a decision or generated content.</li>
<li><strong>Monitor and improve.</strong> Performance is checked on real-world data and the model is retrained when it drifts.</li>
</ol>
<h3>Traditional programming vs AI</h3>
<p>In traditional programming, a person writes the rules. In machine learning, the person supplies examples and the system works out the rules. The difference is easiest to see in code:</p>
<pre><code class="language-python"># Traditional programming: a human writes the rule
def is_spam_rule(message):
    return "free money" in message.lower()

# Machine learning: the rule is learned from labelled examples
from sklearn.feature_extraction.text import CountVectorizer
from sklearn.naive_bayes import MultinomialNB

messages = ["free money now", "claim your free prize",
            "lunch at noon?", "meeting notes attached"]
labels = [1, 1, 0, 0]  # 1 = spam, 0 = not spam

vectorizer = CountVectorizer()
model = MultinomialNB().fit(vectorizer.fit_transform(messages), labels)

print(model.predict(vectorizer.transform(["win a free prize"])))  # [1]
print(is_spam_rule("Win a FREE prize"))                           # False
</code></pre>
<p>The hand-written rule misses "win a free prize" because nobody anticipated it. The learned model catches it because "free" and "prize" appeared in the spam examples. Real systems use far more data, but the principle is the same.</p>
<table>
<thead>
<tr>
<th></th>
<th>Traditional programming</th>
<th>Machine learning</th>
</tr>
</thead>
<tbody><tr>
<td>Input</td>
<td>Rules + data</td>
<td>Data + expected answers</td>
</tr>
<tr>
<td>Output</td>
<td>Answers</td>
<td>Rules (a trained model)</td>
</tr>
<tr>
<td>Strength</td>
<td>Predictable, easy to audit</td>
<td>Handles messy patterns that are hard to describe by hand</td>
</tr>
<tr>
<td>Weakness</td>
<td>Breaks on cases the author did not foresee</td>
<td>Depends on data quality; harder to explain</td>
</tr>
</tbody></table>
<h2>Types of Artificial Intelligence</h2>
<p>Researchers usually group AI by how broad its abilities are [2]:</p>
<ul>
<li><strong>Narrow AI (weak AI).</strong> Built for a specific task or set of tasks: voice assistants, recommendation engines, spam filters, image classifiers. This is the kind of AI that exists today, and it powers the systems most people use.</li>
<li><strong>Artificial General Intelligence (AGI, or strong AI).</strong> A system that could understand, learn and apply knowledge across a wide range of tasks at human level. IBM describes this as currently theoretical, with no known system approaching it [2].</li>
<li><strong>Artificial Superintelligence (ASI).</strong> A hypothetical system that would surpass human intelligence. It is a topic for philosophers and safety researchers, not an engineering product.</li>
</ul>
<p>When will AGI arrive, if ever? McCarthy noted that most AI researchers believe new fundamental ideas are required, so no date can be reliably predicted [1]. Treat confident timelines in either direction with caution.</p>
<h2>Understanding the Core Technologies</h2>
<p>AI, machine learning and deep learning are often used interchangeably, but they are nested ideas. IBM describes machine learning as sitting directly under AI, deep learning inside machine learning, and generative AI built on deep learning [2].</p>
<img src="https://cdn.hashnode.com/uploads/covers/6af1583b68fd16fccf9bbafa/53fb98a8-a339-4425-8427-8736a54e102a.png" alt="Nested diagram showing generative AI inside deep learning, inside machine learning, inside artificial intelligence" style="display:block;margin:0 auto" />

<h3>Machine learning</h3>
<p>Machine learning (ML) builds models by training an algorithm on data so that it can make predictions or decisions without being explicitly programmed for each task [2]. It comes in three main flavours:</p>
<table>
<thead>
<tr>
<th>Paradigm</th>
<th>What the system gets</th>
<th>Example</th>
</tr>
</thead>
<tbody><tr>
<td><strong>Supervised learning</strong></td>
<td>Inputs paired with correct labels</td>
<td>Classifying emails as spam or not spam</td>
</tr>
<tr>
<td><strong>Unsupervised learning</strong></td>
<td>Inputs with no labels; it finds structure itself</td>
<td>Grouping customers with similar buying behaviour</td>
</tr>
<tr>
<td><strong>Reinforcement learning</strong></td>
<td>Rewards and penalties for actions in an environment</td>
<td>Training an agent to play a game or control a robot by trial and error</td>
</tr>
</tbody></table>
<h3>Neural networks and deep learning</h3>
<p>A neural network is a model made of interconnected layers of simple units ("nodes") loosely inspired by the brain. Each node combines its inputs, applies a small mathematical function and passes the result on. <strong>Deep learning</strong> uses networks with many layers. In a widely cited 2015 review, LeCun, Bengio and Hinton explain that such models learn representations of data at multiple levels of abstraction, using backpropagation to adjust each layer's parameters [3]. In an image, early layers might detect edges, middle layers shapes, and later layers whole objects.</p>
<h3>Natural Language Processing</h3>
<p>Natural Language Processing (NLP) is the branch of AI that works with human language: translation, summarisation, question answering, sentiment analysis and chatbots. Modern NLP is dominated by deep learning, especially the transformer architecture described below.</p>
<h3>Computer Vision</h3>
<p>Computer vision teaches machines to interpret images and video: recognising objects, reading text in photos, detecting defects on a production line. LeCun and colleagues note that deep convolutional networks brought breakthroughs in processing images, video, speech and audio [3].</p>
<h2>Generative AI</h2>
<h3>What is generative AI?</h3>
<p>Generative AI refers to deep learning models that create new content, such as text, images, audio or video, in response to a prompt. IBM explains that generative models encode a simplified representation of their training data and then draw on it to produce new work that is similar to, but not identical with, the original [2].</p>
<h3>Large Language Models and the transformer</h3>
<p>A <strong>large language model (LLM)</strong> is a generative model trained on huge amounts of text. Most are built on the <strong>transformer</strong>, an architecture introduced in 2017 in the paper "Attention Is All You Need". Its authors showed that a network based entirely on attention mechanisms, with no recurrence or convolution, could be trained faster and in parallel [4]. Attention lets the model weigh how relevant every other word in a sentence is to the word it is currently processing.</p>
<p>Scale mattered too. The 2020 GPT-3 paper described an autoregressive model with 175 billion parameters and showed that sufficiently large models could perform new tasks from just a few examples placed in the prompt [5].</p>
<h3>How ChatGPT-style tools work, at a high level</h3>
<img src="https://cdn.hashnode.com/uploads/covers/6af1583b68fd16fccf9bbafa/ac6ce61e-72d6-461d-b5d9-c75c6cbfbe07.png" alt="Pipeline diagram: pre-training on text, tuning, then prompt and token-by-token generation" style="display:block;margin:0 auto" />

<p>IBM summarises generative AI in three phases [2]:</p>
<ol>
<li><strong>Training.</strong> A foundation model learns patterns from very large collections of raw data. For a language model, the core task is predicting the next token (a word or word fragment) given the ones before it.</li>
<li><strong>Tuning.</strong> The model is adapted to a purpose, for example by fine-tuning on example conversations or by reinforcement learning from human feedback (RLHF), where people rate responses so the model learns which ones are better.</li>
<li><strong>Generation and evaluation.</strong> When you type a prompt, the model generates a response one token at a time, each choice conditioned on everything before it. Developers evaluate outputs and keep tuning. Retrieval-augmented generation (RAG) can also supply relevant outside documents so answers are not limited to training data.</li>
</ol>
<p>This explains both the fluency and the weaknesses of these tools. The model is optimised to produce plausible continuations, not to look facts up in a verified database.</p>
<h3>Images, audio and video</h3>
<p>Image generators commonly use diffusion models, which learn to add noise to images until they are unrecognisable and then reverse the process to produce new images from a prompt [2]. The same family of ideas, along with transformers and other architectures, is applied to audio, music and video, and multimodal models can accept or produce several kinds of content at once.</p>
<h2>Real-world Applications</h2>
<table>
<thead>
<tr>
<th>Sector</th>
<th>Typical use</th>
<th>AI technique involved</th>
</tr>
</thead>
<tbody><tr>
<td><strong>Healthcare</strong></td>
<td>Flagging suspicious regions in medical images; summarising clinical notes</td>
<td>Computer vision, NLP</td>
</tr>
<tr>
<td><strong>Education</strong></td>
<td>Adaptive practice that adjusts to a learner's level; feedback on writing</td>
<td>Recommendation models, LLMs</td>
</tr>
<tr>
<td><strong>Finance</strong></td>
<td>Fraud detection by flagging unusual transactions or logins [2]</td>
<td>Anomaly detection, ML</td>
</tr>
<tr>
<td><strong>Manufacturing</strong></td>
<td>Predictive maintenance: forecasting equipment failure from sensor data [2]</td>
<td>Time-series ML</td>
</tr>
<tr>
<td><strong>Transportation</strong></td>
<td>Driver assistance, route optimisation, autonomous-vehicle research</td>
<td>Computer vision, reinforcement learning</td>
</tr>
<tr>
<td><strong>Agriculture</strong></td>
<td>Spotting crop disease or water stress from field and drone images</td>
<td>Computer vision</td>
</tr>
<tr>
<td><strong>Cybersecurity</strong></td>
<td>Detecting abnormal network behaviour and phishing</td>
<td>Anomaly detection, NLP</td>
</tr>
</tbody></table>
<p>Maturity varies. Fraud detection and predictive maintenance are established. Fully self-driving vehicles are, in IBM's words, still not perfected [2]. In regulated fields such as healthcare, AI is best seen as decision support that a qualified person reviews.</p>
<h2>Benefits of Artificial Intelligence</h2>
<ul>
<li><strong>Automation of repetitive work</strong>, freeing people for higher-value tasks [2].</li>
<li><strong>Faster insight from data</strong>, since models can scan volumes no person could read.</li>
<li><strong>Better decision support</strong>, with data-driven predictions delivered in real time.</li>
<li><strong>Fewer routine errors</strong>, for example by flagging mistakes before they happen.</li>
<li><strong>Round-the-clock availability</strong> for support and monitoring.</li>
<li><strong>Reduced physical risk</strong>, by taking on dangerous work such as inspecting hazardous sites [2].</li>
</ul>
<h2>Challenges and Limitations</h2>
<ul>
<li><strong>Bias.</strong> Models learn from historical data. If that data reflects unfair patterns, for example in hiring decisions, the model can reproduce and scale them [2].</li>
<li><strong>Data privacy.</strong> Training and running AI often involves personal data, so regulations such as the GDPR place obligations on how it is collected, stored and used [2].</li>
<li><strong>Security.</strong> Data can be poisoned or tampered with, and models can be stolen, reverse-engineered or manipulated [2].</li>
<li><strong>Hallucinations.</strong> Generative models can produce fluent text that is unsupported or false. A survey in <em>ACM Computing Surveys</em> reviews how hallucination degrades system performance and how it is measured and mitigated [6]. Always verify important claims.</li>
<li><strong>Employment concerns.</strong> AI changes how tasks are done, and some tasks will be automated. How this shifts whole occupations is an active, unsettled debate, so be wary of anyone offering exact predictions.</li>
<li><strong>Ethics and accountability.</strong> IBM lists explainability, fairness, robustness, accountability and privacy as recurring principles of responsible AI [2].</li>
</ul>
<h2>The Future of AI</h2>
<p>No one can forecast the future of AI precisely, but several directions are visible today:</p>
<ul>
<li><strong>Multimodal models</strong> that combine text, images, audio and video, and <strong>smaller, more efficient models</strong> that run on modest hardware [2].</li>
<li><strong>AI agents.</strong> Systems that do not only answer, but plan and use tools such as other applications and services to complete tasks on a user's behalf [2].</li>
<li><strong>Human-AI collaboration.</strong> The most reliable results often come from pairing a model's speed with human judgement, where the person sets goals, checks outputs and owns the decision.</li>
<li><strong>Responsible AI.</strong> The U.S. National Institute of Standards and Technology released its voluntary AI Risk Management Framework on 26 January 2023. It organises risk management into four functions: Govern, Map, Measure and Manage. NIST later published a generative-AI profile in July 2024 and notes that the framework is currently being revised [7].</li>
</ul>
<h2>Conclusion</h2>
<p>AI is best understood as a toolbox, not a single machine. Machine learning finds patterns in data, deep learning does so at scale with multi-layer networks, and generative AI uses those networks to create new content. Everything deployed today is narrow AI: powerful within its task, brittle outside it, and dependent on the quality of its data.</p>
<p>The practical takeaway is balanced. Learn what these systems can do, check what they produce, and ask who is accountable when they are wrong. If you want to go hands-on, start with a small supervised-learning project like the spam example above, then explore the sources below.</p>
<h2>References and Further Reading</h2>
<p>All links were last accessed on 9 November 2026.</p>
<ol>
<li>McCarthy, J. <em>What is Artificial Intelligence?</em> Stanford University, revised 2007. <a href="https://jmc.stanford.edu/articles/whatisai/whatisai.pdf">https://jmc.stanford.edu/articles/whatisai/whatisai.pdf</a></li>
<li>Stryker, C. and Kavlakoglu, E. <em>What is artificial intelligence (AI)?</em> IBM Think, published 9 August 2024. <a href="https://www.ibm.com/think/topics/artificial-intelligence">https://www.ibm.com/think/topics/artificial-intelligence</a></li>
<li>LeCun, Y., Bengio, Y. and Hinton, G. "Deep learning." <em>Nature</em> 521, 436-444 (2015). <a href="https://doi.org/10.1038/nature14539">https://doi.org/10.1038/nature14539</a></li>
<li>Vaswani, A. et al. "Attention Is All You Need." arXiv:1706.03762 (2017). <a href="https://arxiv.org/abs/1706.03762">https://arxiv.org/abs/1706.03762</a></li>
<li>Brown, T. B. et al. "Language Models are Few-Shot Learners." arXiv:2005.14165 (2020). <a href="https://arxiv.org/abs/2005.14165">https://arxiv.org/abs/2005.14165</a></li>
<li>Ji, Z. et al. "Survey of Hallucination in Natural Language Generation." <em>ACM Computing Surveys</em> 55(12), 2023. <a href="https://doi.org/10.1145/3571730">https://doi.org/10.1145/3571730</a></li>
<li>NIST. <em>AI Risk Management Framework.</em> National Institute of Standards and Technology. <a href="https://www.nist.gov/itl/ai-risk-management-framework">https://www.nist.gov/itl/ai-risk-management-framework</a></li>
</ol>
]]></content:encoded></item></channel></rss>