AI content is becoming increasing pervasive in all formats, from text, to audio, photo and more recently video, both long form and short from. In all formats the quality is now such that its becoming increasingly challenging to distinguishing real from AI generated content.
If we focus on text based academic and university AI content detection, platforms like Turnitin claim high AI detection, but have suffered significant issues with false positives. Given the pattern-matching that fundamentally underpins AI, and the fact that most publicly accessible academic writing tends to be structured and written in a fairly specific way, this is a large part of what AI models are trained on, so given the models are trained on the very data that they are trying to mimic, its no surprise the results look very similar, and explain why AI detection in this are is challenging. Sure, there are certain 'tells' in grammar, sentence structure and phrasings, but ultimately the issue is with academic writing it is assessed on a specific format and structure, the same that AI is trained on, and optimised to mimic, and this is the same structure that AI detection is also aimed at detecting, in laymans terms... follow this structure perfectly to pass, but don't follow it too perfectly or we'll mistake it for AI, but don't deviate too much or it wouldn't pass, its an impossible dilemma, but not that consequential in the scheme of things.
What is more concerning is how AI generated influencers are not only getting wider followings, but are also getting less distinguishable from real people. I recently watched a program about a UK based creator who had made a AI influencer talking about history, they pointed out the strength of using an AI influencer in the context of history was that the AI character could be placed in historical events which would be unreal. whilst this was a great case study, the underlying issue is the character itself was hard to distinguish from a real person, especially given the short 15 second format of social media videos.
underpinning this is a fundamental issue, how can you tell who is real and who is AI generated? we would think this would be easy, the assumption is that we can tell from the expressions and emotions. but heres where i think something has been missed. AI is trained on videos and content created by real people with real expressions and real emotions. AI may not understand emotions, but like a psychopath, it can, through cognitive empathy (as a result of pattern matching) very effectively mimic emotions, and, just like we easily overlook and mistake psychopaths, it is just as easy to overlook and mistake AI influencers in the same way. And, we can expect this to continue as innovations in AI make the quality of the videos and character consistency improve.
I'm not sure what the solution or conclusion is to this observation, except that maybe we need a resurgence in human to human conversation, debate and learning, in the context of university, face to face Q&A and presentation shows learning and is much more comparable to real world than essay writing for example. In the context of influencers, maybe we just need to stop consuming such ephemeral low quality content.
firstly, i should point out i'm in no way a developer, i have no coding expertise beyond excel formulas and rudimentry HTML and Python. I am using this as a lay person trying to get things done as quickly as possible.
Google Gemini - I use this for brainstorming, fast insights, and helping to plan and structure slides using other inputs, so you can consider this kind of a 'work desktop'
NotebookLM - this is what i use for understanding and learning information, the podcast, slide video and auto summaries help me understand info fast, then i can Q&A the source material without hallucinations, in terms of the thesis, this is also where i can pull out the key talking points from the source material. I'll also have 1 soruce in 1 chat, but i'll have another chat thats an aggregated grouping of all sources, so i use the single source for deep diving, and the aggregated group for 'pattern analysis'
Jenni.ai - this has been fantastic as a way of helping me find new cite-able sources which i then put into notebookLM, it also acts as a first place for me to do my writing.
AI is coming to eat jobs, i saw a recent video where some factory workers had cameras strapped to their head to record them as they worked on highly dexterous tasks, pair that with the advancements in humaniod robots and you know thats really the end goal, replacing humans even on the more fine motor skill factory tasks.
In the white collar world, its in a way worse, Oracle recently laid off a 30,000 employees.
it got me thinking whats going to be left. much as i'd be supportive of universal basic income, i believe thats just false hope and fantasy, but also a story for another day.
so, if we ignore business owners and founders, what are we left with?
i suspect it'll be personalised, hands-on jobs with high stakes, where people trust in a human more than a machine, not necissarily rationally but simple emotionally, they trust a human more - doctors, teachers, psychologies are the first, but i think eventually vulnerable. what i think would be really hard to AI-replace is a physiotherapist.
my reasoning, every case a physio meets is unique, and a puzzle where an initial diagnosis is not always clear, at the same time, treatment is very physical and requires tactile skills and 'feel' that a machine cannot replicate easily. so given how tailored and abstract both diagnosis and treatment is, i think its one of the few jobs that would be impossible for at least the next 10 years to replace.
taking the same 'logic' of the job of a physio, i think blue collar jobs, specifically a mechanic, plumber, electrician are the most safe, again, it comes down to how the cases are very unique, theres a high degree of tactile skill involved, and some of that tactile skill cannot be easily replaced with current robotics, e.g. in the case of a physio, being able to detect a knot in a persons back, and know what pressure to apply to 'click' it, or in the case of a mechanic, how tight to screw a nut that it stays in place without being too tight.
I could be totally wrong, but for now, based on what i've seen this is my general conclusion.