What next, now that AI has outgrown humanity?

Elon Musk has revealed that AI has now “exhausted” all available human data for training. Let that sink in for a moment.

The entirety of mankind’s intellectual achievements has already been consumed by these insatiable thinking machines. From the great works of philosophy, literature and science right through to Fifty Shades fan fiction and that poetry blog you started in eighth grade. From quantum physics and ponderings on the mysteries of the universe, through to cute cat videos and conspiracy theories about milk. AI has seen it all and it still wants more.

Speaking on a livestream on his platform X, Musk stated that: “The cumulative sum of human knowledge has been exhausted in AI training. That basically happened last year.”

Hardly the most controversial thing he’s said recently, but it has potentially profound repercussions.

It’s quite humbling really that the sum of all human knowledge has been reduced to a mere amuse-bouche for our soon-to-be robot overlords. So, what's next on the menu?

DIY AI

The large language models behind the new iterations of AI that have emerged in recent years are trained on vast data sets to recognise patterns and predict outcomes. One of the reasons they’ve been able to become so advanced is the sheer quantity of data available out there.

And it’s a heck of a lot too. According to recent estimates, over 400 million terabytes of newly generated, captured, copied, or consumed data is created every single day. Wikipedia, the fount of all internet knowledge, is a measly 200 terabytes. That’s the equivalent of two million Wikipedias every 24 hours.

But not all data is created equal. Artists and authors, creators of much of the "good quality" data, have argued for compensation after it was revealed that many AI models had been trained on copyrighted or protected material without their consent. Perhaps that doesn't matter much anymore in the grand scheme of things.

Training of these advanced models won’t simply grind to a halt now they’ve ran out of data. Instead, AI will do what we feared it would all along and manage just fine without us.

Musk suggested that tech companies could substitute real human data with “synthetic” material created by AI to fine tune their new models. He explains: “it will […] write an essay or come up with a thesis and then will grade itself and […] go through this process of self-learning.”

In fact, this is already happening. Both Microsoft, Meta and ChatGPT creator Open AI have reportedly been using synthetic data to train their latest models.

Is it real?

However, relying on AI-generated material in the training process comes with its own rather obvious limitations. The most problematic being the so-called “hallucinations” prevalent with these advanced models, whereby the AI produces output littered with inaccuracies or total nonsense.

Musk sees this as the biggest challenge for developing and advancing the technology. When training models on material AI has created, how will you know if the answer is even real?

The scary and in all probability most likely answer is, it won’t matter.

Musk’s comments are just the latest in a series of alarming indicators that we are nosediving into a post-truth world. Whether an AI’s output is accurate or not will ultimately become immaterial.

Granted, if you want to cure cancer or fly to Mars, you need reliable, precise data. But for most of us, that’s not how AI is going to impact our lives.

Musk’s statement seems almost dismissive, nonchalantly alluding to a view that everything that came before – the sum of all human knowledge - is as disposable as a single-use plastic cup. To be crumpled up and thrown into an ever-growing sea of trash.

We are already in the midst of a Baudrillardian nightmare. We live in an age of simulation, misinformation and constructed reality.

A death spiral

Social media is now the number one source of information for people wanting to know what’s going on in the world. But this week, Meta chief Mark Zuckerberg announced he would be rolling back its third-party fact checking in favour of “community notes”, where the people get to decide what’s true or not.

The platforms Meta controls, Instagram and Facebook, have also recently repealed some of its user guidelines and censorship rules, seemingly opening them up to hate speech. Users are now allowed to say that gay people are mentally ill.

It’s a death spiral. The algorithm is unconcerned with accuracy, meaning the content shaping a user’s worldview – and their ability to discern fact from fiction - need not reflect reality.

There is even a theory growing in acceptance that large swathes of the internet are overrun with fake content and bots posing as people. It's called the Dead Internet theory. Look it up while you still can.

Arguably most if not all of the content consumed online today is an engineered representation or interpretation only loosely inspired by real events. Including this very article.

If this is where the sum of humanity’s knowledge leads us, then who can blame AI for outgrowing it…

***

Views own.

Copyright Joshua Allsopp 2025.

Cover image made with Microsoft Bing Image Creator.