Fundamentally what is Artificial Intelligence?
A book I read recently suggested that AI is about prediction: predicting the next word in a sentence (ChatGPT) or predicting the next move in a game like chess or Go.
However, I would disagree, and say it’s about calculation. Prior to the rise of neural networks, computer software like spam filters or Google Translate used statistical models like Bayesian filters. Prediction (spam or ham) is based on the outcome of those calculations. Prediction with modern neural networks (so-called deep learning) under-the-hood is still fundamentally calculations, just with better results than the previous generation of statistical modelling.
What makes neural networks different from other types of calculations? Neural networks succeed at looking at a very wide variety of data, and lots of it. 120 zettabytes of data will be generated this year, that’s 329 million terabytes of data created each day. This data generated by humans has been growing exponentially:
With a large amount of data, it needs to be processed efficiently. Even with the similarly exponential rise in compute power, we need algorithms that can take shortcuts in calculating big data, without impinging on the outputs. This is where multi-layered neural networks succeed.
In other words, AI applications are growing by leaps and bounds in recent years through the combination of increased data availability to train models on, increased processing power (thanks Nvidia), and improved mathematical models (neural networks). These calculations are often used to create predictions: is an email spam or not (creating a cleaner inbox) or creating content (generative-AI) like writing a first level response to support-desk email query. However it is not always predictions, under the hood it is still calculations: think the facial, image, or voice recognition on your phone.
What makes AI special is its ability to work with real-world scenarios. It can extrapolate from a real-world data, e.g. turnign something fuzzy like a low-light camera image into something meaningful: recognise the low-light image is you, and unlock your phone. The real-world isn’t black-and-white, and AI can work with that, providing meaningful output like voice, images, text, taking various forms like efficient routing, summarising meetings, or even healthcare diagnostics.
This allows for human labour to be freed up for the important work of interaction with clients. AI is never going to replace the human touch, and expert human time may be more available when we have AI tools doing parts of the job for us. However, one caveat is AI isn’t always intelligent: feed in poor quality data, and you’re certain to get garbage out.
About the Author
Matthew Whyte was born in rural Waikato and grew up in Tonga, before returning for tertiary studies at the University of Waikato. He completed a degree in computer science, including artificial intelligence, in the mid-2000s. At the time AI wasn’t very smart, so he focussed on web development and hosting. He has since been involved in the telecommunications and software development sectors, working in Aotearoa and the EU.
