AI gedachten / AI thoughts

It is not very original to compare the current state of AI to the internet bubble (or dot-com crisis) of 25 years ago. Even Sam Altman (of OpenAI, the company behind ChatGPT) is already making remarks about the investments in AI, and the probability that the bubble will burst. But, except for the craziness in the financial markets, there are some more insightful comparisons to be made with 25 years ago.

About 25 years ago, ‘internet’ was something new and exciting. “Everything” was still possible, the world was about to change fundamentally, and nobody really knew what we were heading for. So we tried out many things. We still needed to separate the wheat from the chaff: new things were proposed that are now completely normal in our everyday lives. But there were also new things that we now look back at with a tender heart – we were so cute and naïve back then! The comparison with today’s search for ‘useful’ AI applications is straightforward.

Prehistory

In 2000, it was not straightforward for an average citizen to have access to the internet. In the Netherlands, the first commercial internet service provider was already founded in 1993, but in 2000 most users had to rely on their landlines to connect to the internet, while enduring horrible sounds from their modem. Internet was for nerds, hackers, scientists and people who worked at large companies. You probably had to be abit technical as well. For the normal consumer, there was not so much activity on the internet either: you had to know the programming language HTML (and some other things) to contribute something.

In 2022, almost nobody used AI. OpenAI was founded in 2015 by Sam Altman, Elon Musk and Peter Thiel, and the company introduced the GPT technology in 2018. Only data scientists who were fluent with the Python programming language were experimenting with it. AI was something for nerds, hackers, scientists and people who work with large (technical) companies. Under the hood, a lot of AI was already deployed though. But, as hoods are intended to hide things, this was largely invisible to the bigger audience.

The bigger audience

Internet changed as a result of some developments that happened at more or less the same time and that reinforced each other. Google changed ‘internet search’ fundamentally. The emergence of ‘broadband’ changed your internet connection: while you were “surfing the internet” (as we called it), your daughter could still call her friends using your landline. Interactivity was fostered by “web 2.0“, and social media (like Hyves in NL and Habbo Hotel worldwide) enabled normal users to contribute content to the internet without having to use HTML. Mobile internet and the iphone made the internet something that you could also use when outside. Internet had become a consumer product.

AI changed almost overnight by the end of November 2022 when ChatGPT was introduced.. The quality of the GPT3.5 language model had improved so much compared to its predecessors, that it suddenly became useful. (Or rather, seemed useful, but more about that in another blog). The secret: besides the smart things in their GPT motor (more about that in another blog as well), the use of enormous amounts of text as training data caused the model to perform significantly better than predecessors and competitors. ChatGPT has the worldwide record of the fastest growing consumer product: a hundred million new users in just two months. AI had become a consumer product.

Craziness

The availability of internet all over the place caused investors to throw huge amounts of money to companies. The ‘new economy‘ seemed in reach: eternal growth without depleting resources. Making profits was frowned upon: a startup that made profit obviously did not have enough imagination to invent new services.

The amount of investments in AI is unprecedented. We don´t call it ‘new economy’, but the impact of AI on work is causing feverish phantasies already. Ideas like the ‘zero person company’ are seriously discussed: a company where the only employees are AI bots. The first experiments were not yet very convincing, but in China we can already witness ‘dark factories’ that have ‘lights out production’: no lights are needed because all work is done by robots.

At the time of the internet bubble, wild ideas were explored. The Dutch Ministry of Traffic and Water (that also goverend Telecommunication), discusseed the plan to give every Dutch citizen a ‘free email address’. Earning money on the internet was just starting; we now know that advertisements are the default model for that, but back in the day many people explored the concept of ‘micropayments’ as the go-to solution for buying content: make it easy to pay 10 cents to read one article. There was a service Third Voice’ that gave users the possibility to comment on a website independent of that website – Google Reviews or Trustpilot were not yet there. And there was the ‘battle of the portals’: a portal was a collection of links and services that hoped would serve as the main entry for consumers to start using the internet. The domain name ‘sportal’ (a portal aimed at sports) was sold for a lot of money.

I worked at KPN (the Dutch incumbent telco) at that time. We were search for ‘the killer app’: an app that would encourage mobile data usage. Preferably in combination with ‘location based services’: when you walked past a bakery shop, you would receive an offer (“20% off when you show this message”).

So many bright ideas. Not relevant anymore.

We are here!

In 25 years, we will look back on this period in time. For some applications of AI we can hardly imagine that we could ever do without them. We will also smile with some compassion at other applications, the ones that we thought were brilliant and very much needed, but were just a dead end street. Which applications will fall into which category? I’m already curious by now!