Hidden Problems with AI That (Virtually) Nobody Talks About
Real issues affecting people right now
Not All AI is Equal AI companies are run two-tier systems (many more even). People with money get faster computers, better servers, and premium services. Everyone else gets the cheap/overloaded version.
Here's the problem: when AI systems get busy, they get dumber. The cheap services handle more users per server by tuning quality parameters (among other measures), so the AI gives worse answers because it used less cycles, or stopped early, or was a little more repetitive. Schools with tight budgets buy the cheap version. Their kids get worse help with homework than rich kids using the premium version at home.
The same thing happens at work. Your boss tells you to "use AI to be more productive." But if your company buys the bargain version, you're automatically behind colleagues/competitors who have access to better AI tools. Most people don't even realize this is happening.
You Can't Tell When You're Talking to a Bot More companies use AI to answer phones and chat with customers. The AI sounds human, but it's not. When people find out they wasted time talking to a computer, they feel stupid or tricked.
This damages trust over time. One bad human employee can hurt a company's reputation. Now imagine that happening everywhere, all the time, as people realize they're being 'fooled' by machines. Requiring declaration is a fine first step but it just makes the experience all the more excruciating for the person stuck on the phone with this system explaining to them for the twelth time that they're talking to an AI, it doesn't solve the problem that that org should have had a human there but didn't because it cost less to use a cloud chatbot service they configured in 3 minutes by uploading a couple of corporate documents because they believed the pitch of the AI sales bod.
"AI Makes Mistakes" is Just an Excuse Tech companies have a clever trick. They say "AI is sometimes wrong, so always double-check the answers." This sounds reasonable, but it's really about avoiding lawsuits.
All these companies train their AI on the same (often stolen) data from the internet. So their AI makes the same types of mistakes. Instead of fixing the problems (because they'd have to fix all the underlying source data to do it for real), they just warn users to be careful. Then when something goes wrong, they say "we told you to fact-check."
The truth is, if you need AI to be right about something important, you need to build a custom model from scratch and test it properly. But companies prefer selling one-size-fits-all AI because it's more profitable, and it's easy to add to the cloud account.
Blocking Bad Words Ruins Good Conversations Companies put heavy filters on AI to stop it saying offensive things. But these filters are crude. They don't just block swear words - they make the AI confused about related topics.
Take mental health. If the AI can't talk about depression, self-harm, or trauma, it also gets weird about normal emotional topics. The filters create blind spots that make the AI give worse advice about relationships, therapy, or personal problems.
The media only cares about whether AI says rude things. They ignore how censorship makes AI dumber about important human experiences. It's also crucial that despite most people (an as yet it seem unknown amount) asking gen-ai for personal advice, the services can't sell it as that. Even google's latest series of podcast ads were about how you can ask it for bland football opinions, when virtually nobody is doing that, because they can't call it a digital therapist despite that being a massive use case in reality.
Free AI Sites Are Dangerous Lots of sketchy AI apps pop up on social media promising "uncensored AI" or "free premium access." People, especially kids, think these work like ChatGPT and dump their personal information into them.
These sites have no rules, no security, and no oversight. People upload photos, share secrets, connect social media accounts, and even enter passwords (tools already offering to automate your life if you give them all your info). All this data goes to random servers run by unknown people, probably never to be deleted.
This teaches terrible online safety habits. A whole generation is learning that it's normal to tell AI systems their deepest secrets.
Your Social Media is a Scammer's Goldmine Scammers used to need luck to find good targets. Now they use AI to scan thousands of social media profiles and build detailed files on potential victims.
The AI reads your posts and figures out your fears, hopes, and weak spots. Then scammers use this to create perfect fake messages. They don't need to make deepfake videos. They just need to sound like someone you might know or trust.
Anyone with public social media posts is vulnerable. The AI can create a psychological profile in minutes and generate a personalized scam approach. Yet politicians keep saying it's your fault for falling for "sophisticated" attacks.
Teaching AI Makes You Replaceable When you use AI at work, you're training it to do your job. Over about a year, the AI learns your methods and thinking patterns. You get more productive, which seems good.
But your employer now owns an AI version of your expertise. They can use this AI to train cheaper workers or move the work somewhere else entirely. You've essentially taught your replacement.
This isn't the same as old automation that replaced manual work. This is capturing how you think and solve problems, then selling that thinking to someone else.
All Your Eggs in One Basket The main AI services already get hacked or crash regularly. But governments and companies are rushing to put all their important work into these systems.
If your local government uses Microsoft's AI for everything, one server problem could shut down benefits, licenses, and emergency services. The scary part is that when AI breaks, there's often no backup plan. Workers have been retrained to ask the AI instead of doing things the old way.
Nobody remembers how to do the job without the computer. So a brief AI outage becomes a major crisis.
These problems are happening right now, but most news coverage focuses on science fiction scenarios instead of real damage being done today.
AGI is not a threat (yet) but systems that look like AGI definitely are. There's not much point complaining about how AGI will definitely eventually create Terminators, while there's already enough raw potential to completely rewrite how everyone lives and works anyway.
Just looking at deployments so far in the handful of orgs I've seen get their staff into adoption, only 1 or maybe 2 really has enough of an existing review cycle and active introspection to actually get productivity gains. Most real people are stuck using it for its more gimmicky or personal features, and finding that its output isn't really reliable enough (at least when under load) to be used in anger, creating just as much work tidying up its output as it would have been to do the work in the first place....
Somehow tho everyone's supposed to find double productivity under orders from leadership who have barely seen gen-ai actually in use, and usually over someone's shoulder or entirely based on things they read in the newspaper (or worse, were told about how supposedly awesome and all-power it is by someone on their whatsapp).
The entire chat dump behind this summarised document can be found here
https://gist.github.com/bobbigmac/cd9ff3ca9527e3a2b6f0f908e494258b