Last month, all alarms went off in Google’s headquarters.
Even the New York Times has an entire article dedicated to this. According to them, ‘Code Red’ went off in the highest structures of the company.
Google is scared, very scared.
Artificial Intelligence has recently made a giant leap that could put at very risk Google’s core business, search.
The question is then inevitable.
Could we be soon witnesses to the death of one of the so-called ‘trillion-dollar’ companies, and with it the demise of entire industries like SEO, SERPs, and ultimately digital marketing?
Google, the first Internet monopoly, is highly exposed
Google is currently valued at a whopping 1.13 trillion dollars.
That’s a one and twelve zeros. But what’s even crazier, is that Google was, in November 2021, an almost 2 trillion dollar company.
That’s quite a decrease, but it still allows them to be the fourth biggest company in the world by market capitalization.
Of course, being so huge, one imagines that its revenues are also huge, right?
256 billion dollars in revenue in 2021. Not bad.
But how much money is that?
For reference, more than Portugal’s expected entire GDP for 2022.
In other words, Google’s revenues in 2021 were bigger than what Portugal managed to produce, as an entire country, for a whole year.
Seeing those numbers, one can’t act surprised by Google’s insane valuation.
Or should we?
Truth be told, Google’s business model has a catch.
Diversification isn’t Google’s strongest suit
It’s undeniable that Google has some impressive numbers.
As a more visual representation, let’s look at the following monographic:
If we check Google’s quarterly results for Google in June 2022, Google managed to achieve an income of 69.7 billion dollars.
Almost as impressive as their final profit, 16 billion dollars, which is a profit margin of 23%.
That’s huge, considering how Amazon, one of the other big guys, actually lost money that quarter.
This seems all amazing and great, right?
However, if you watch carefully, something’s odd.
Or at least, improvable.
Out of the 70 billion in revenue, 41 billion, almost 60%, comes from one unique source, Search advertising, the industry where Google holds around 92% of the market share.
And the problem is that this is, specifically, the market that AI can potentially disrupt forever.
Transformer AI models. But what on Earth is that?
Quick hint, they aren’t killer robots like the ones in the movies.
ChatGPT and what’s to come
You’ve probably heard or read about ChatGPT a lot recently.
Like, painfully, a lot.
But it’s understandable, the technology is unbelievable.
ChatGPT in 30 seconds
For those unaware, OpenAI, a non-profit company founded by Elon Musk and Sam Altman, among others, has published the latest version of its chatbot, ChatGPT, powered by the largest transformer language model ever assembled, GPT-3.5, with over 175 billion parameters.
In case you’re wondering, chatbots are robots you can talk to.
Actually, you’ve probably spoken to many recently, especially when engaging with call centers, insurance companies… and basically any customer service you can find today.
Those chatbots, in all sincerity, are pretty annoying and painfully limited.
But this one’s not.
One could say it’s the same animal but a totally different beast, but to me, it’s a completely different animal also.
ChatGPT, among other things, can respond to almost any question with extremely eloquent responses, code anything you wish in many different programming languages, write completely new bedtime stories, debug your own code, and others.
It’s so impressive, some have claimed it could be the very first intelligent, sentient AI model.
Short answer, fuck no.
The problem with that statement is that people mistake high predictability for sentience.
A probabilistic beast
GPT, just like any other neural network, is a probabilistic beast; it’s capable of predicting with an astonishing success rate the next correct word in response to a sentence, thereby creating perfectly-elaborated sentences while sounding very human-like when engaging with it.
But having high success in predicting eloquent responses is one thing, being capable of truly understanding what they’re responding to is another.
Don’t let sensationalist media fool you, state-of-the-art AI isn’t sentient, period.
But although an armed uprising of human-killing robots seems improbable today, there’s something that ChatGPT is actually poised to disrupt from the get-go.
And that’s Internet search, Google’s golden goose.
The death of Google Search and SEO?
Unlike Google Search, ChatGPT releases you from having to scroll endlessly through pages of links while giving you concise and direct responses.
This could make people resort to querying these systems instead of searching through Google.
In that scenario, the appeal for companies to promote themselves through Google could drop significantly, putting Google at risk.
But Google isn’t the only one.
Thousands of marketing agencies rely on their SEO expertise, the capacity to help their clients ‘be seen’ on Google. This sudden change in customer journeys to using AI instead of Google could be devastating for them.
Hence, seeing how impressive ChatGPT is already, and considering that a newer version of the underlying AI model, GPT-4, is going to be released early next year…
Are we prepared to ‘kill’ Google?
Differing what’s wrong from what’s right
Creating a good search engine isn’t a simple task.
These algorithms need to be continuously fine-tuned to avoid exposing fake news, racism, homophobia, or even plagiarism, in an effort to offer the searcher relevant but acceptable information.
This effort, in turn, seems like an even bigger challenge with AI.
Today, AI needs to figure out three things:
- Data Bias
- Avoiding making shit up
AI, the ultimate blackbox
As described earlier, GPT-3.5 has been fine-tuned for 175 billion parameters requiring 800 GB of storage.
In very simple terms, this means that trying to explain how these models work and make decisions or predictions is close to impossible.
Thus, if ChatGPT makes a racist comment, there isn’t really a way of explaining why.
Biased data, biased model
Also, as these models aren’t sentient but mathematical models that have learned to answer by brute force and vast amounts of data, they are very dependent on having unbiased data sources and diversified teams of data engineers, which is a huge challenge in today’s world.
A challenge why? Here’s a fact for you.
According to data from this September by Zippia, 70% of computer programmers are male, and 69% of all programmers are white.
Assuming the majority of them aren’t racist, nevertheless, this extremely skewed workforce is surely heavily culturally biased, which is not great for AI models that intend to become universal and applicable to all of society.
Google Search isn’t going anywhere… for now
In spite of all this, I don’t feel that Google’s reign as the supreme search engine is in doubt for now.
However, I do have a strong opinion that the future of search engines is going to be AI-led, so although today’s AI models are limited and dangerous for mass use, ChatGPT has shown us what the future, undoubtedly, looks like.
Luckily for Google, it has its own large language model, LaMDa, and surely has taken note of what LLMs are capable of thanks to OpenAI.
Nevertheless, all this proves how disruptive AI is going to be. But not only for you and me, but also for the biggest companies in the world.