Amazon Just Proved AI Isn't The Answer Yet Again

You would think they would learn their lesson. But nope!

Photo by Abid Shah on Unsplash

Last week, Amazon showed Kim Kardashian how to actually break the internet (which is still a relevant reference… right?). Thanks to Amazon Web Services completely crapping the bed, everything from major online banking portals to the biggest video games like Fortnite and even some major social media sites went fully offline. And this wasn’t a momentary outage like we are used to, as it took AWS 16 hours to fix the problem. Some reports suggest over 2,000 businesses were impacted by this outage, and, due to its scale and the time it took to resolve, it has cost billions of dollars in lost productivity. So, what happened? Well, it looks like Amazon proved that AI can’t replace workers for the second time!

The official cause of this AWS outage was a DNS resolution issue. But that isn’t the full story, as typically these kinds of issues are relatively easy and quick to fix, especially for an industry leader like AWS. This outage should have been a momentary blip. So, why did it take 16 hours to resolve?

Well, it is painfully apparent to anyone paying attention what has happened: AI turned this issue into a major global event.

Amazon is 100% on the “replace workers with AI” bandwagon. Back in July, Amazon announced it would shrink its workforce as it rolled out generative AI to replace workers. At the same time, at least several hundred workers in the AWS cloud computing unit were laid off. Some reports suggest more than a few hundred were laid off. And now, exactly a week after the incident, Amazon has announced that 30,000 employees, including many at AWS, will be let go and replaced with AI. In short, Amazon is currently undergoing a huge restructuring to implement AI.

But did you catch that?

Just a few months ago, Amazon laid off a significant number of engineers in the AWS unit responsible for resolving outages. That timing is mighty suspicious…

However, we are missing critical context here.

AWS is massively expanding! The same AWS cloud computing which powers all these internet services that crashed is also used to ‘train’ AI. As such, to meet demand, AWS is spending an utterly gargantuan $100 billion on increasing its compute power in 2025 alone! To give you an idea of how huge an investment that is, AWS likely spent $108 billion on infrastructure from 2011 to 2022. Considering how modern infrastructure is more cost-effective, that means this new $100 billion expenditure could more than double AWS’s compute power.

Onboarding and operating that amount of additional computing power also requires onboarding a large, highly educated workforce. But AWS is doing the opposite and has instead been shrinking its workforce!

So, in real terms, AWS is severely understaffed. And you simply cannot bridge this gap by overworking either. AWS has to be replacing workers and filling new positions with AI. Just to be clear, I can’t prove that statement, but it is almost certainly the case.

And that explains why it took so damn long to fix the problem.

AI is inherently unreliable. In fact, it is so unreliable that AI programming tools actually slow programmers down, as constant “hallucinations” produce tonnes of bugs that are hard to identify and resolve. AI is also terrible at being accurate, as these hallucinations cause it to make up data and botch data insights. Naturally, AI cannot reliably resolve big, complex, novel tasks.

AI can’t be trusted to create basic programs, so how can it be trusted to troubleshoot and fix several billion dollars’ worth of offline cloud computing infrastructure? It can’t.

We can all read between the lines here, can’t we?

It is blatantly obvious that AWS is replacing critical workers with AI, and that last Monday, an error happened that it simply couldn’t solve, but because the team of workers who could actually fix the problem had been whittled down to nothing, it took them exponentially longer to fix this issue than it should have. As a result, what should have been a small blip turned into a devastating global event. It is so obvious that I have seen multiple AI-friendly media outlets come to the same conclusion.

Given that they have experienced this issue before, you would think Amazon would have learnt not to rely on AI like this.

Remember Amazon’s “Just Walk Out” grocery stores? The idea was that facial-recognition cameras, shelf sensors, and AI would track which items a customer had taken, then charge their Amazon account once they left, negating any need for a cashier or self-checkout. But there was an issue. A report found that over a thousand remote workers had to be hired to monitor the video feeds and verify 70% of the customers’ purchases, as the AI was consistently getting it wrong. This amount of labour isn’t cheap, even if it is outsourced overseas, and Amazon’s “Just Walk Out” AI became significantly more expensive than simply hiring regular cashier staff. Understandably, Amazon has struggled to sell the system to third parties and has had to switch its own grocery stores to a fancy non-AI self-scan system instead. Basically, the AI was too unreliable to replace workers.

All of this happened in the latter half of 2023. Since then, these AI tools have not improved much. For example, ChatGPT-4 had been released months before Amazon abandoned this “Just Walk Out” concept, and many consider more recent models to be of lower quality. There has been zero evidence that the problem which caused this venture to fail has been even remotely resolved.

In fact, OpenAI’s recent paper found that this unreliability (or the presence of hallucinations) is inherent to the LLM AI models that power the kinds of tools AWS is trying to utilise, and there is no solution. In fact, they found no way to even reduce the frequency of hallucinations from their current level. It doesn’t matter how much data or computing power you shove into them, how you engineer prompts, or how you optimise the model; hallucinations are here to stay.

This is something AI engineers have known for a while now. Yet, Amazon seems to not only be ignoring these experts but also forgetting its own previous, public and expansive AI failures and is instead rapidly placing this devastatingly bad technology at the core of its business.

When I say this is a catastrophic mistake, I don’t just mean for Amazon, but for everyone. AWS is a critical sector of the internet, as we saw with this outage. And, if AWS continues to fire its workers and blindly adopt AI, these kinds of outages are going to become more frequent and last longer. In short, everyone will suffer from this, all because Amazon is too pig-headed and greedy to learn from their own mistakes. How pathetic.

---

Thanks for reading! Don’t forget to check out my YouTube channel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!

---

Sources: Futurism, Reuters, UC Today, The Register, 80 LV, Reuters, Will Lockett, My London, DCM, IG, Will Lockett, The Guardian, Tech Crunch, FN, METR