AI is not new, so why is it getting so much attention now?

In 2006 I thought I had an original research question for my MSc dissertation on computer vision. I dusted off a space on the shelf for my future Nobel prize. Then I started my reading and found PDF scans of papers from the 1950s asking pretty much the same question as mine. That was an eye-opener.

Although AI and computing are now almost synonymous, mechanical devices such as automata and logical reasoning machines have existed or been theorised for millennia. Some of these are discussed in the book “AI Narratives: A History of Imaginative Thinking about Intelligent Machines”.

So what has changed to make the world so excited about AI now?

It’s not AI as a whole, but rather a specific type of AI called generative AI. Simply put, this is an area of AI that takes an input, such as text or an image, and generates an output, such as new text or a new image. Three things have come together to make generative AI so compelling: data, computing power and transformers.

Three major milestones for generative AI; internet data in the 1990s, computing power in the 2010s and the transformer architecture from 2016.

The Internet has provided huge amounts of data in the form of webpages, books, images, blog posts, news articles, etc.. Generative AI has access to this data and has ‘learnt’ from it. To give an analogy, you could not complete the phrase “Comment vous appelez-vous?” without knowing French. Generative AI cannot be expected to provide useful completions without having had access to data.

Generative AI needs the time to learn from this data. This is done by providing massive amounts of compute power and this is readily available through cloud computing. Before the 2010s, organisations needed to buy, site, power and maintain computer servers. With major cloud providers such as Microsoft, Google and Amazon offering compute on-demand, I can rent several thousand servers for a few hours if I needed to.

And finally, in 2017 a major advance in how the generative AI algorithms could work was described in a research paper called “Attention is all you need”. This dramatically improved how effective and relevant the generated output could be.

Those three factors have been instrumental in our current situation. Refinements and improvements are taking place every day as people and organisations learn to use and monetise generative AI.