Ok, let's take a break from the frenzy of AI, AGI, and how it's gonna solve all our problems according to big tech moguls. For the common of mortals, like me, we personally have seen progress made only in Generative AI: a neural network capable of predicting "tokens" (next word, syllabus) to generate texts with an impressive accuracy. These texts, not only are perfectly formulated sentences in any language spoken on the globe, they also are built on the biggest dataset the humanity ever had.
Internet and the third industrial revolution
For the people of my generation, the "millennials", we still remember the cathodic television, the pagers, the fax, or the soothing music of the modem accessing the internet. Haaa such good memories they are, this little music which was giving you the time to realize you were going to access the world straight from your home. Of course, most of the time, prefixed by "Mom, get off the phone!".
That was the beginning of internet, the beginning of the third industrial revolution.
Why it is important to acknowledge that? Because, this revolution was also a digital one and the shifts we are seeing today with Gen AI are similar.
At that time we weren't opening NetScape browser to read the news, engage on social media with the world or watch our last favorite TV show. They actually were, for each of these previous services an already existing physical equivalent: the newspapers, people gatherings and the television.
Services which all have seen a major shift in how we consume them. Let's make a parallel with the newspaper industry, because I think that it is the closest from the industry I'm working in, or at least a part of it.
After some time of the democratization of the internet, paper newspapers saw a huge drawback from it, people started to read them online, for free, and just stop buying their physical form: the paper version. It was a huge shift for the industry, how to finance proper investigation and journalism if our main source of income disappear? Well, all of them started to develop a digital form of it, at first free, then for a monthly subscription. The same way you would buy your paper newspaper.
But what happened to paper newspaper? I can only tell it for France, but they still exist. They live under IV drip from state subventions. Most of them got bought by billionaires which kind of use them to seed society with ideas which help their businesses, but it is another problem. My point is, that a necessity has become a political decision. The necessity of providing information to people has become a political decision of financing them through tax payer money even if the businesses were at loss.
Gen AI and the new use of websites
There are plenty of tools we can create with Gen AI and of course up to this date they are plenty of more to come that we are not aware of yet. But Gen AI has been around for long enough to see what kind of tools emerged and which ones are probably here to stay.
AI coding assistant: Like journalists had to adapt from publishing long, supported, documented investigations, sometimes having to travel the world to cover one event, they now can investigate from the comfort of their offices and have to adapt to an audience with diminished attention spans. This emerged from the ability of anyone being able to be a "journalist". News, blogs, texts were everywhere, opinions were everywhere, comments were everywhere, anyone started to have a voice, all over the web. We are now seeing a similar shift as developers. Code is everywhere, anyone can code, code is accessible and ideas which stayed at the stage of ideas become reality, for cheap. Does that mean we don't need journalists? Does that mean we don't need developers? Well, it's a societal decision to make. I personally think that investigation is still the proper way of doing journalism and just read opinions (which most newspapers are today) just to entertain myself, in the tram or when I have couple of minutes to kill. Will developers still be needed? We are already starting to see the limit of coding assistants, security issues, scalability issues ... AI coding assistants are to developers what internet is to journalists, a faster, easier way to produce text, information or code. Anyone can call itself journalists or coder because they are able to have a voice, though not anyone is able to produce quality information or code. Because, the latter is a profession, a profession that takes study and / or years to master. AI coding assistant and internet are awesome, because they put the entrance barrier down. What used to be an "elite" profession becomes more accessible to anyone.
AI search generated answers: This is to me the second most important revolution which emerged from the tools created from Gen AI. You've probably noticed it, we all noticed it, even more the data analysts, because people don't visit websites anymore. Yes, this AI generated summary of the websites retrieved from your query at the top of the answers of your favorite search engine. It is awesome right? Well, when we see the amount of cr*p we have to go through sometimes, just to extract the text of a website, no wonder why this tool has become one my favorite. I mean, this is literally as best as it can get for a search engine. The ability to browse, extract and summarize the data from several sources. Why searching for minutes for the information you need when an AI assistant can do it for you. Plus, we are not spammed with ads and this repetitive cookie notification banner imposed by GDPR in the European region. No popup, no spam, just straight to what we are looking for: information.
AI prompting: The first tool that introduced us to LLMs. A direct prompt. Why browsing the internet when you can just query an LLM to retrieve, extract and structure the information for you? Well, because it comes with limitation: the knowledge cut off. Models are trained on an amount of data at a T time. Meaning they are not able to retrieve most recent information. A limitation, big tech already found a solution for: url context and ground search. We have the tool, the LLM capable of reading, understanding, extracting and summarizing the information, what we missed was the ability for it to search most recent information by itself (without having to train it constantly on new information, and always being behind). That's when came agents. Behind the terms "ground search" or "url context" are simply little code snippets allowing to search the internet and scrap the content for you, feeding the LLM with most recent information.
So, now we've stated the current state of the art on the internet, the question everybody ask itself is: Do we still need websites or not?
Two types of websites: Informational and Functional.
Functional websites: Let's start by eliminating the type of websites which are not directly impacted by Gen AI, at least, not from AI prompting or AI search generated answers: the institutional forms, marketplaces, browser games, betting platforms, basically any website which offers a functional service to its users. Let's nuance it with the fact that some websites are a combination of both, functional and informational, this is the case of most institutional websites which most often include functional services through forms.
Informational websites: Those are the ones the most impacted by Gen AI and its tools: institutional websites, newspapers, blogs, forums and forum likes (such as stack overflow) etc. Their main purpose is to deliver up to date information to their users. The last published laws, the last published news, the last online showcase of an institution or a business. Let's not forget this latter, websites are also a way for organizations to showcase themselves, publish their values, show their team, brand themselves etc.
Towards the slow death of Informational websites?
Well, if you ask anyone in your organization for which the number of visit per day of their website matters, they will surely tell you about their concerns. Since the development of the AI tools mentioned above, and for the reasons detailed above, they've seen striking drops in their analytics. It's a game changer, people are more and more using AI search generated summaries and AI prompting, before actually, when it happens, visit a website.
With a wild guess, I can imagine that google's teams are currently thinking about a new way of feeding their LLM with last up-to-date data. Because that is what we are leaning toward: search engines are switching toward AI prompting.
And that is the question in any manager's head, is it still worth it to invest millions, in development, in an informational website?
That's where we get back to our parallel with paper newspapers. Of course big tech has no value, else than scrapping your data, from your website. Of course they are probably thinking about new ways of making you giving them the necessary data for them to provide accurate information to their users in a more efficient and manageable way for them. Scrapping your website at query level, after your query, is absolutely not efficient and is part of the long time LLMs take to generate their answers (when they think). I'm sure they are thinking of an Internet 3.0 where data are directly sent to the source, where they can harvest them, where they can control them, at least, in their shoes, I know I will. It is what happened with paper newspaper ... you remember?
So here again, what was a necessity for organizations becomes a political decision. Are we still maintaining an expensive website for the sole purpose of controlling our data?
The echo chamber trap
That will be the conclusion of this personal analysis of what I think is the fourth industrial revolution. We can still see, in little letters, and we surely should definitely ALWAYS see them, the mention: "AI can make mistakes". This is a reminder to the users to check the initial data, the raw data, the ones the models used to generate its answer, the only reason in my opinion for which organizations should still maintain their own public data, under whichever form they see fit.
Because, unfortunately, the more we'll be using these tools, the more we will trust them (because they are good), the more we will forget about these little letters at the bottom of the software. Like the small clauses of a contract we never read, the ones which by their size are paradoxically the most important.
We should not forget that with the death of informational websites comes an echo chamber. If people are not offered, through their feeds or through paper newspapers, like it used to be, new ways of seeing the world, they will only get stuck in their own questions. At the end, an LLM will only answer the questions you ask it, it won't ping you with the last news from the world if you never ask ...