AI Timeline View: Lens 1 of 3
What can a timeline of AI tell us about how fast this thing is going to move
My last two blogs, AI Impact Curve: Which way is that sucker going to go? and AI: Time to cross the adoption chasm? I set out arguments for why forming a view on the level of business change that AI will bring is important. Although the future is a tricky thing to predict, we can look at the present and the past through different lenses to help us peer into the dim mist that is the future.
My next three blogs will give three lenses through which we can look at AI. The first lens is a timeline view. By understanding the origins of AI we can begin to get a sense of the pace of change.
Definitions
Lets get the boring stuff out of the way before we start. When most people refer to “AI” today they mean the new generation of ‘intelligent’ chatbots like ChatGPT. These are powered by Large Language Models (LLMs) like OpenAI’s model and there are others such as Anthropic’s Claude, Meta’s open source Llama and so on. This article uses the terms ‘LLM’ and ‘AI’ reasonably interchangeably.
AI Timeline
Above is an extremely high level view of the origins of AI. As a side note, I value accuracy more than “being right” so if you have information that contradicts or corrects anything in this article please put this in the comments.
Early days of Machine Learning (1950s — early 1990s)
With computers becoming a viable tool, and the establishment of durable databases, the idea that machines could “learn” was a natural progression. Machine Learning, in an over-simplified nutshell is assigning different weights (a weight is sometimes rather confusingly also called a bias) to different pieces of data. Sophisticated statistical algorithms are then applied to this weighted data to make determinations or predictions. Based on the accuracy or otherwise of these predictions, the weighting is adjusted.
The Internet and the rise of Machine Learning (1990s to mid 2010s)
The Internet, a paradigm change in itself, was also a huge technology enabler. Computer and Data Scientists could now share their research online and receive feedback in real time. This, combined with the ever-increasing power of computers saw Machine Learning enter a kind of golden age. Neural networks, connecting different islands of Machine Learning processes began to spring up in companies like mushrooms.
We need to talk about Tay (23 March 2016–24 March 2016)
In the mid 2010s enough confidence had built around the power and flexibility of Machine Learning and Neural Networks that Microsoft released their chatbot, named Tay onto Twitter. It didn’t take users long to realise that Tay’s flexibility and willingness to please was its Achilles heel. Within hours, in response to savvy users knowing they could manipulate the chatbot by posting offensive tweets to it, Tay was spewing out the most offensive racist, sexist garbage you could imagine. Microsoft pulled it within 17 hours. It was a sobering lesson in both human nature and the limits of a machine’s ability to communicate using even state of the art machine learning models.
Birth of Large Language Models 2015–2022
In December 2015 OpenAI (the organization behind ChatGPT) was formed as a non profit. An early founder, Elon Musk is no longer with the organization, and has an active law suit filed against OpenAI claiming it has deviated from its not for profit mandate. In 2017, eight researchers, most of them from Google’s “Deep Mind” programme released a research paper named Attention is all you need. This paper broke new ground proposing a “Transformer” model for training models on large data sets.
Although as this article has tried to show AI has build incrementally on advances in computing, if you were to draw a line in the sand and talk about the “invention” of AI, the publication in 2018 of the OpenAI white paper “Improving Language Understanding by Generative Pre‑Training” could be said to be the lift-off point. Building on the “Transformer” concept of the Google research, the OpenAI paper and the research behind it added the “Generative Pre-trained” elements, and the world was introduced to the “Generative Pre-trained Transformer” (GPT). OpenAI’s internet chat based offering, called understandably enough “ChatGPT” was launched as a research preview in November 2022.
The Rise and Rise of AI 2023 —
Although still a research preview, it took only 5 days for ChatGPT to reach 1 million users after its release on 30 November 2022. In January 2023 Microsoft invested 17B USD into OpenAI, and has integrated OpenAI technology into its core product offerings. Although it seems like AI has been part of our lives for a while, features like the ability to search the internet from within an AI chat are actually recent product features, released only in the last couple of years.
To bring this article up to date the March 2025 tweet by Tobi Lutke, CEO of Shopify contained a memo point that showed just how far AI has come in a short space of time:
5. Before asking for more Headcount and resources, teams must demonstrate why they cannot get what they want done using AI.
The takeaway?
I wasn’t one of ChatGPT’s first million users, but I was a very early adopter, first playing around with this technology in January of 2023. I opened my web browser, typed in my first question: “what is the meaning of life?” and got a balanced, well reasoned reply. Other replies were a little hit and miss (this was the first every OpenAI model) but I immediately knew there was something extraordinary going on.
Having been in tech for over 30 years I could tell that things would never quite be the same again.
Follow me for the next two lenses — the Technical lens and the Tech Revolution lens.
