Preface: I work in AI, and on LLM's and compositional models.
None, frankly. Where AI will be helpful to the general public is in providing tooling to make annoying tasks (somewhat) easier. They'll be an assisting technology, rather than one that can replace people. Sadly, many CEO's, including the one where I work, either outright lie or are misled into believing that AI is solving many real-world problems, when in reality there is very little or zero tangible involvement.
There are two areas where (I think) AI will actually be really useful:
Healthcare, particularly in diagnostics. There is some cool research here, and while I am far removed from this, I've worked with some interns that moved on to do really cool stuff in this space. The benefit is that hallucinations can actually fill in gaps, or potentially push towards checking other symptoms in a conversational way.
Assisting those with additional needs. IMO, this is where LLM's could be really useful. They can summarize huge sums of text into braille/speech, they can provide social cues for someone that struggles to focus/interact, and one surprising area where they've been considered to be great (in a sad but also happy way) is in making people that rely on voice assistants feel less lonely.
In both of these areas you could argue that a LLM might replace a role, although maybe not a job. Sadly, the other side to this is in the American executive mindset of "increasing productivity". AI isn't a push towards removing jobs entirely, but squeezing more productivity out of workers to enable the reduction of labor. It's why many technological advancements are both praised and feared, because we've long reached a point where productivity is as high as it has ever been, but with jobs getting harder, pay becoming worse and worse, and execs becoming more and more powerful.
Perhaps it's not possible to fully replace all humans in the process, but harmful content filtering seems like something where taking the burden off humans could do more good than harm if implemented correctly (big caveat, I know.)
Here's an article detailing a few peoples' experience with the job and just how traumatic it was for them to be exposed to graphic and distributing content on Facebook requiring moderator intervention.
The question of which jobs should be replaced by AI depends on societal values, priorities, and the potential impact on workers. Generally, jobs most suited for replacement by AI involve repetitive, high-volume tasks, or those where automation can improve safety, efficiency, or precision. Here are some categories often discussed:
Repetitive and Routine Tasks
• Manufacturing and assembly line work: Machines can perform repetitive tasks with greater efficiency and precision.
• Data entry and processing: AI can automate mundane tasks like updating databases or processing forms.
• Basic customer service: Chatbots and virtual assistants can handle frequently asked questions and routine inquiries.
High-Risk Roles
• Dangerous jobs in mining or construction: Robots can reduce human exposure to hazardous environments.
• Driving in risky environments: Self-driving vehicles could improve safety for delivery drivers or long-haul truckers in hazardous conditions.
Analytical and Predictable Roles
• Basic accounting and bookkeeping: AI can handle invoicing, payroll, and tax calculations with high accuracy.
• Legal document review: AI can analyze contracts and identify discrepancies more quickly than humans.
• Radiology and diagnostics: AI is becoming adept at reading medical scans and assisting in diagnoses.
Jobs With High Inefficiencies
• Warehouse operations: Inventory sorting and retrieval can be automated for faster fulfillment.
• Food service (e.g., fast food preparation): Robotic systems can prepare meals consistently and efficiently.
• Retail checkout: Self-checkout systems and AI-powered kiosks can streamline purchases.
Considerations for Replacement
1. Human Impact: Automation should ideally target roles where job transitions can be supported with retraining and upskilling.
2. Creativity and Emotional Intelligence: Jobs requiring complex human interaction, creativity, or emotional intelligence (e.g., teaching, counseling) are less suitable for AI replacement.
3. Ethical Concerns: Some jobs, like judges or certain healthcare roles, involve moral decision-making where human judgment is irreplaceable.
Instead of framing it as total “replacement,” many advocate for AI to augment human workers, enabling them to focus on higher-value tasks while reducing drudgery.
Any body-breaking heavy labour. Emphasis on body-breaking; there's nothing wrong with hard work, but there are certain people that believe hard work = leaving your body destroyed at 50.
Anti-Cheats. Train an AI on gameplay data (position, actions, round duration, K/D, etc.) of caught cheaters and usw that to flag new ones. No more Kernel level garbage, just raw gameplay data.
None. The current ones with internet content, reporting, and call centers are already making things worse. Just no.
It can definitely be a useful tool though, as long as you understand its limitations. My kids school had them feed an outline to ChatGPT and correct the result. Excellent
consultants generate lots of reports that ai can help with
I find ai useful to summarize chat threads that are lower priority
a buddy of mine uses it as a first draft to summarize his teams statuses
I’m torn on code solutions. Sometimes it’s really nice but you can’t forward a link. More importantly the people who need it most are least likely to notice where it hallucinates. Boilerplate works a little better
ai as in AI: aircraft auto-landing and pitch levelling. near-boundary ship navigation. train/ freight logistics. protein folding. gene mapping.
ai as in LLM/ PISS: hmmm... downlevel legalese to collegiate-, 6th-grade-, or even street-level prose. do funny abridged shorts. imo, training-wheels to some shakespearean writing is appreciated.
None, as long as society uses labor as a means to secure basic necessities. Shifting that towards some infinitely-scalable capital equipment owned by entities kinda fucks the whole system, don't it?
i think i read some posts like hackernews that they already use AI as a therapist. I have good conversations with chatgpt when i asked for some personal advise. I haven't tried talking to a real therapist yet but i can see AI being used for this purpose. The services may still be provided by big companies or we can host it ourselves but it could be cheaper (hopefully) compared to paying a real person.
Don't get me wrong, i'm not against real physicians in this field, but some people just can't afford mental healthcare when they need it.
Requirements revision review. It is the most mind-numbing part of my job and fortunately only a small portion of it.
A word changes, even just punctuation changes, can change the meaning drastically. And finding that change within a hundred page document is a task humans just plain suck at. Get a computer to compare revision A to revision B, highlight the changes, then pass it on to the human to interpret the change and decide what to do from there.
Realistically, a lot of the stock photo industry. If a few people can generate pictures on demand, you won't really need anyone doing sets, lights, wardrobe, etc for a series of generic photos .
I think many things that solicitors do could be easily replaced with AI since it's just parsing the contents of documents and then writing a few templated summaries.
Currently very few jobs should be replaced with AI. But many jobs should be augmented with AI. Human-in-the-loop AI amplify the finate resource of smart humans.
The only full job I can think of is assistant to a busy person. I don't think any whole jobs are done better by ai. Some of the jobs recommended in this thread would be better to be removed rather than replaced.
So, I think ai makes a better assistant to a person doing a job rather than a replacement to compete a job on its own. It can write rough drafts that a talented writer can expand and edit. It can quickly generate several plans that an experienced leader can pick from or discard. It can look through a designer's portfolio and spit out "new" combinations of their past designs that the designer can then build upon.
Any one of these jobs could give up and submit the AI's output as their own, but I think the quality of the results would suffer.
Illustrators. Actors. Animators. Writers. Editors. Directors. Let's make art impossible to sell so we can get back to proper starving, errr... I mean... making art as a form of expression rather than commerce.