What is AI?

So what is AI?

To get great ideas about what problems we can solve with it, we need to briefly revisit what AI is and what it can and cannot do.

You may - or may not - be surprised to hear that AI, as we know it now, started as a concept back in the 1950s. It began with a question from Alan Turing's groundbreaking paper: "Can Machines Think?"

Alan Turing was a UK visionary, a war hero, and widely considered the father of modern computing and artificial intelligence. He helped crack Nazi codes during World War II and died tragically in 1954, at age 41. He was posthumously honored and since 2021 appears on the UK 50 pound note.

His theory was simple but profound: If a machine could carry on a conversation indistinguishable from a human, we might reasonably say it could think. That idea became known as the Turing Test.

From there, the story of AI unfolded in waves. In the 1960s and seventies, we saw the rise of rule-based expert systems. Then came the AI winters when funding dried up and progress stalled. In the nineties and two thousands, machine learning reignited interest with breakthroughs like IBM's Deep Blue beating chess Champion Gary Kasparov. The 2010's brought deep learning and huge leaps in image and language processing. And now in the 2020s, we've entered the era of generative and agent AI systems that not only create, but also act.

ChatGPT launched in 2022 and was the fastest tech adoption in history. And yes, AI has now passed the Turing Test.

So what actually is AI?

Artificial intelligence is a broad term for any technology that does things we'd normally rely on human intelligence for - like spotting patterns, answering questions, or making decisions.

Inside that, we have machine learning. Unlike traditional programs that follow set rules like A equals B plus C, machine learning analyses large amounts of data to find its own patterns and make decisions without needing to be explicitly told what to do.

Then inside that is Deep Learning, a more advanced kind of machine learning that uses layered neural networks. This is the part of AI that reminds us of the human brain or constellations of stars in the night sky. Billions of connections lighting up together. The inspiration behind Southern Sky AI.

Right at the center is Generative AI, the most explosive leap we've seen in recent years. It doesn't just recognise patterns, it creates, it writes, draws, builds, and composes. This is the type of AI you're using when you talk to ChatGPT.

So what can AI actually do?

You can think of AI superpowers in four main buckets.

Let's start with Natural Language Processing or NLP. To bring this to life, I've borrowed an analogy from the US Game Show, "The Price Is Right." In one game, a contestant drops a disc into the top of a Plinko board, and it bounces unpredictably through rows of pegs before landing in a price slot. At first, it seems random, but over many drops a natural pattern emerges.

NLP works in a similar way. It breaks down language into tokens - units of meaning - and predicts the next one based on probability, context, and patterns that are seen before. Each decision point is like a peg on the board and large language models like the model underlying ChatGPT, don't just have a few pegs, they have billions of them, called parameters.

The early GPT-4, for example, is trained on hundreds of billions of words, using more than a trillion parameters to understand and generate human-like language. The model navigates this vast breadth of data to find the most likely and meaningful response. Mathematics on the grander scale.

The next is Computer Vision, AI's ability to understand and interpret images. One of the classic tests for these systems is the age old question, "Is it a cat or is it a dog"?

On the surface, it seems simple, but to a machine, every image is just a grid of pixels made up of colour and light values. Just like language models predict the next word based on patterns in text, Computer Vision models learn to recognise patterns in pixels, shapes, edges, textures to determine what they're seeing.

They're trained on millions of labeled images, learning what makes a cat, or a dog, a dog, even when the photos vary in lighting angle or background. And over time the model doesn't just spot animals. It can identify defects on a ship's hull, monitor safety gear, or scan containers for pests. The logic is the same as with language: with enough examples, the AI learns what to expect and where - just now it's doing it with vision.

The third AI superpower is Reinforcement Learning and Robotics. Unlike language or vision models that learn from large static data sets, reinforcement learning is dynamic. It's all about learning by doing. Here, an AI interacts with its environment, takes an action and receives feedback in the form of a reward or penalty. Over time, it learns to optimise its decisions by repeating what works and avoiding what doesn't, very much like how we train a puppy.

This is the core mechanism behind autonomous systems, self-driving cars, warehouse robots, and even AI in gaming. A powerful example comes from Boston Dynamics - the robotics company originally spun out of MIT, later owned by Google, and now part of Hyundai. In this video, you see one of their robots learning to walk, crawl, jump, and run, not because it was pre-programmed to move that way, but because it learned through thousands of tiny mistakes and micro corrections.

The fourth and newest AI superpower is Agentic AI, our emerging frontier. This isn't just about asking ChatGPT to summarise an email or write a recipe. Agentic AI can follow instructions, plan steps, and execute tasks autonomously without human input at every stage. It's like having a digital teammate that doesn't just answer questions, but actually gets things done.