Considering the current advances of AI how likely do you think an AI might replace you in your job?

how likely do you think an AI might replace you in your job in 10 years?


  • Total voters
    29

CultureCitizen

Well-Known Member
Joined
Feb 14, 2023
Messages
124
To my knowledge, right now only some musicians and graphical designers, copywriters, and people working at call centers are facing a high risk of being replaced.
 
As a retiree, I agree. But I also can't see AI replacing my old job as a Grounds Maintenance supervisor - I doubt they'll ever be able to correctly prune trees...
 
This is such a great question! We have an author publishing a book this month related to CHATGPT capabilities and the impact on our culture. It is certainly an interesting time in AI.
 
In ten years I hope to be retiring and I'm sure an AI can do at least a third of my job already.
 
As a retiree, I agree. But I also can't see AI replacing my old job as a Grounds Maintenance supervisor - I doubt they'll ever be able to correctly prune trees...
That will take longer at least 20 years.
 
Speaking of my old profession of software development, I don't foresee AI eliminating jobs. I view AI as providing augmentation to manual and automated tasks, but not eliminating them. I suspect that soon we are going to see a large backlash against AI as some of its failings come into public view.
 
Speaking of my old profession of software development, I don't foresee AI eliminating jobs. I view AI as providing augmentation to manual and automated tasks, but not eliminating them. I suspect that soon we are going to see a large backlash against AI as some of its failings come into public view.
Faster coding means less developers are needed. So individually a single person might not see the change, the situation is different on the aggregate level.
 
Faster coding means less developers are needed. So individually a single person might not see the change, the situation is different on the aggregate level.
To date, the advances in software development has led to more software being developed, not fewer developers. I've seen multiple approaches that were supposed to lead to the end of coding as we know it, and they have all been forgotten.

If we see AI systems that run on a single laptop, then I can see widespread adoption of AI augmented software development. Currently, I don't see using the tools at the scale of current world-wide software development. Even then, I suspect the uses will be in the more rote and repeatable areas of coding and not in the overall construction of programs. When it comes to upgrading and enhancing existing programs (which consumes the vast majority of software development effort), I am not sure that AI capabilities extend to be able to consume and understand an entire program.
 
To date, the advances in software development has led to more software being developed, not fewer developers. I've seen multiple approaches that were supposed to lead to the end of coding as we know it, and they have all been forgotten.

If we see AI systems that run on a single laptop, then I can see widespread adoption of AI augmented software development. Currently, I don't see using the tools at the scale of current world-wide software development. Even then, I suspect the uses will be in the more rote and repeatable areas of coding and not in the overall construction of programs. When it comes to upgrading and enhancing existing programs (which consumes the vast majority of software development effort), I am not sure that AI capabilities extend to be able to consume and understand an entire program.
The first change was gui development... honestly it was a lot simpler to program a terminal.
The second wave of change was web development... it not only requires a different skill set, but mostly different languages unless you pick a full stack js stack.
Python / Julia : they are both concise languages, they save a lot of coding, but again you have to learn new languages and they have a limited scope ( mostly backend stuff ).
Chat gpt writes good enough code most of the time but you still have to adjust it.
Certainly chat gpt is not able to adjust existing programs, but 10 years back AI was completely unable to write code at all.
 
The first change was gui development... honestly it was a lot simpler to program a terminal.
The second wave of change was web development... it not only requires a different skill set, but mostly different languages unless you pick a full stack js stack.
Python / Julia : they are both concise languages, they save a lot of coding, but again you have to learn new languages and they have a limited scope ( mostly backend stuff ).
Chat gpt writes good enough code most of the time but you still have to adjust it.
Certainly chat gpt is not able to adjust existing programs, but 10 years back AI was completely unable to write code at all.
Architectures and languages evolved, but the basics of software development are largely unchanged. In the 1990s, CASE (computer assisted software engineering) systems were going to replace programmers; people would simply draw UML diagrams and the code would be generated from that. About the same time, there was a push for object librarians; a few really smart people would create object libraries and everyone else would simply wire them together. This was supposed to greatly reduce the number of of programmers needed. There were and are "non-programmed," configured systems that quickly required specialists in the system who were paid more than programmers and the systems also quickly top ended in the scope of what they can do. More recently, there are systems like Microsoft's Entity (I forget the name of the Java equivalent) that are supposed to bypass the need for database development and learning SQL. The result is the need to learn an almost SQL language with far less available resources. It also makes a complete mystery of what a new deployment will do and whether it requires an outage. Of course, there are code wizards which provided canned routines to generate baseline code. I suspect if AI scaling becomes feasible, AI-based code wizards will the be result.

Keep in mind that ChatGPT uses somewhere north of a quarter million CPUs. This is not a system that can be easily deployed for a single organization and load demands on a single or small number of such systems to support current software development activities will exceed their capabilities.
 
Architectures and languages evolved, but the basics of software development are largely unchanged. In the 1990s, CASE (computer assisted software engineering) systems were going to replace programmers; people would simply draw UML diagrams and the code would be generated from that. About the same time, there was a push for object librarians; a few really smart people would create object libraries and everyone else would simply wire them together. This was supposed to greatly reduce the number of of programmers needed. There were and are "non-programmed," configured systems that quickly required specialists in the system who were paid more than programmers and the systems also quickly top ended in the scope of what they can do. More recently, there are systems like Microsoft's Entity (I forget the name of the Java equivalent) that are supposed to bypass the need for database development and learning SQL. The result is the need to learn an almost SQL language with far less available resources. It also makes a complete mystery of what a new deployment will do and whether it requires an outage. Of course, there are code wizards which provided canned routines to generate baseline code. I suspect if AI scaling becomes feasible, AI-based code wizards will the be result.

Keep in mind that ChatGPT uses somewhere north of a quarter million CPUs. This is not a system that can be easily deployed for a single organization and load demands on a single or small number of such systems to support current software development activities will exceed their capabilities.
Graphical interfaces have gotten more complex over time , not less.
Some CASE tools were quite useful but rarely ever gained sufficient market share and died quietly.

ChatGPT used a quarter million CPU, to TRAIN. Once the model is trained it can generate the output with a single high-end graphics card.
And models keep getting smaller and faster with each passing year.

The greatest problem with current models is their window , currently sitting at 32 K tokens, which is exceeded by many modules.
This stops them from understanding the context of programs and making corrections.
I am certain this will change in a ten years span.
 
They can have it. Oh, wait what is my job?
It was always more of a process. The process of not actually having a job.

I have a problem with authority figures. Especially those who are less clever than I am; which has made it very difficult for me to maintain a job job.

My entire life has been structured around not actually having a job. It's been very hard work and I couldn't begin to imagine how to write an algorithm for it.
 
They can have it. Oh, wait what is my job?
It was always more of a process. The process of not actually having a job.

I have a problem with authority figures. Especially those who are less clever than I am; which has made it very difficult for me to maintain a job job.

My entire life has been structured around not actually having a job. It's been very hard work and I couldn't begin to imagine how to write an algorithm for it.
That's a good goal. A good question is if all machines can do the work , why should anyone work for any other reason that's not "for fun".
But the current system is structured around getting a job to have an income ( unless you have some business or asset that allows you to collect a rent).
 
I'm an airline pilot with a major U.S. passenger carrier. There's a silly fear among many of my colleagues that pilotless AI controlled passenger aircraft are coming very soon. That's a statement of fear of job loss, not one of real likelihood. There' some major hurdles to overcome in that arena that today, just are not technologically possible yet, if ever.

I'll use U.S. Airways flight 1549 (The Miracle on the Hudson) as a case study.

Captain Sullenberger did what an AI cannot do. He also demonstrated why human intervention in the event of major aircraft systems failures is the only way the people on board an aircraft in distress have any hope of seeing another sunset.

A pilotless AI controlled airliner would have to be a dependent system. It would have to be built on top of an autoflight systems infrastructure which would have to be more technologically reliable then is even possible with current engineering. It woudl have to be coded to make decesions based on procedure. In short, on even the most advanced passenger aircraft, autopilot , and other major aircraft systems including engines fail. They don't fail catastrophically that often but as with 1549, it does happen either from environmental causes, i.e. engine FOD ingestion, or just straight up random systems failures such as the B-737-800 MAX incidents.

When those events occur, pilot experience and instinct takes over. You can't code that, not yet. Sully instantaneously thought outside the box when those geese shut down those engines. I am typed rated and current in the A320 family of aircraft. Starting the APU was the first thing he did, however, that wasn't procedure. The procedure for a dual engine failure has the APU in a all engines fail event started as the forth checklist item. This decision probably saved the lives of every man woman and child on that aircraft. Everything on the Airbus is electric. The emergency power systems are designed to function when there is more time for the crew to digest the situation and come up with a plan, they're meant o give you a chance at survival from cruise altitude, not right after takeoff. We basically lose 90% of every electronic resource we have in the flight deck if both engines shut down. The APU restores 100% of the aircraft's electrical system to full functionality. That actually enables the aircraft flight control system to remain in full operation under what's called "Normal Law." Had they run the procedure by the book, it's very likely that the flight control system would have degraded all the way down to basic "Mechanical Backup" which, I can tell you first hand from simulator training, would not have been manageable that close to the ground. That crew threw the book out the window because of human instinct and judgement. They saved everyone on board. AI can't do that yet, period.

What Sully also did was make a decision of self-sacrifice. That aspect got little attention by the media. The numbers of successful transport category jet aircraft ditchings are dismal. Flight 1549 was the only one ever to happen where everyone lived. Sully knew that, but he made a choice in an instant he had to make, put that jet into an icy river probably killing all if not most people on board, or try for land and definitely kill everyone on board and probably a lot of bystanders on the ground.

Pilotless passenger airliners? That's a LONG WAY off.

Pilotless cargo aircraft? That's another conversation entirely as that's already here.
 
I'm an airline pilot with a major U.S. passenger carrier. There's a silly fear among many of my colleagues that pilotless AI controlled passenger aircraft are coming very soon. That's a statement of fear of job loss, not one of real likelihood. There' some major hurdles to overcome in that arena that today, just are not technologically possible yet, if ever.

I'll use U.S. Airways flight 1549 (The Miracle on the Hudson) as a case study.

Captain Sullenberger did what an AI cannot do. He also demonstrated why human intervention in the event of major aircraft systems failures is the only way the people on board an aircraft in distress have any hope of seeing another sunset.

A pilotless AI controlled airliner would have to be a dependent system. It would have to be built on top of an autoflight systems infrastructure which would have to be more technologically reliable then is even possible with current engineering. It woudl have to be coded to make decesions based on procedure. In short, on even the most advanced passenger aircraft, autopilot , and other major aircraft systems including engines fail. They don't fail catastrophically that often but as with 1549, it does happen either from environmental causes, i.e. engine FOD ingestion, or just straight up random systems failures such as the B-737-800 MAX incidents.

When those events occur, pilot experience and instinct takes over. You can't code that, not yet. Sully instantaneously thought outside the box when those geese shut down those engines. I am typed rated and current in the A320 family of aircraft. Starting the APU was the first thing he did, however, that wasn't procedure. The procedure for a dual engine failure has the APU in a all engines fail event started as the forth checklist item. This decision probably saved the lives of every man woman and child on that aircraft. Everything on the Airbus is electric. The emergency power systems are designed to function when there is more time for the crew to digest the situation and come up with a plan, they're meant o give you a chance at survival from cruise altitude, not right after takeoff. We basically lose 90% of every electronic resource we have in the flight deck if both engines shut down. The APU restores 100% of the aircraft's electrical system to full functionality. That actually enables the aircraft flight control system to remain in full operation under what's called "Normal Law." Had they run the procedure by the book, it's very likely that the flight control system would have degraded all the way down to basic "Mechanical Backup" which, I can tell you first hand from simulator training, would not have been manageable that close to the ground. That crew threw the book out the window because of human instinct and judgement. They saved everyone on board. AI can't do that yet, period.

What Sully also did was make a decision of self-sacrifice. That aspect got little attention by the media. The numbers of successful transport category jet aircraft ditchings are dismal. Flight 1549 was the only one ever to happen where everyone lived. Sully knew that, but he made a choice in an instant he had to make, put that jet into an icy river probably killing all if not most people on board, or try for land and definitely kill everyone on board and probably a lot of bystanders on the ground.

Pilotless passenger airliners? That's a LONG WAY off.

Pilotless cargo aircraft? That's another conversation entirely as that's already here.
100% agree. Watching videos of Tesla drivers testing the latest "self driving" beta makes me more convinced than ever; we are nowhere near achieving these levels of AI.
 
I try to remind people that before the latest hype about AI (actually machine learning), computer automation of various tasks has been happening for years and it has led to certain functions being eliminated (before computers, mechanization was doing the same thing). There has been autocomplete for text in searches, texts, etc. for some time now. ChatGPT is merely an enhanced version.

One of the bigger challenges are that human senses have a great deal of overlap and redundancy. Automated systems typically fail spectacularly upon the failure of a single sensor, while humans adapt to multiple failures. Another issue is that humans adapt to new situations and, with high frequency, react in appropriate manners. For machine learning situations, it is unknown how the system will act when presented with a situation that it has not been specifically trained for. We simply do not know how human response mechanisms work.

Advances in computer technology are very interesting, but one should not overestimate what is actually being done. One should also not underestimate the amount of physical hardware used to accomplish current feats. It must also be recognized that machine learning is a two-phase effort, training then operation. Once in operation, machine learning does not improve, it is frozen. Lastly, don't underestimate the complexity of simple, human tasks.
 
I'm an airline pilot with a major U.S. passenger carrier. There's a silly fear among many of my colleagues that pilotless AI controlled passenger aircraft are coming very soon. That's a statement of fear of job loss, not one of real likelihood. There' some major hurdles to overcome in that arena that today, just are not technologically possible yet, if ever.

I'll use U.S. Airways flight 1549 (The Miracle on the Hudson) as a case study.

Captain Sullenberger did what an AI cannot do. He also demonstrated why human intervention in the event of major aircraft systems failures is the only way the people on board an aircraft in distress have any hope of seeing another sunset.

A pilotless AI controlled airliner would have to be a dependent system. It would have to be built on top of an autoflight systems infrastructure which would have to be more technologically reliable then is even possible with current engineering. It woudl have to be coded to make decesions based on procedure. In short, on even the most advanced passenger aircraft, autopilot , and other major aircraft systems including engines fail. They don't fail catastrophically that often but as with 1549, it does happen either from environmental causes, i.e. engine FOD ingestion, or just straight up random systems failures such as the B-737-800 MAX incidents.

When those events occur, pilot experience and instinct takes over. You can't code that, not yet. Sully instantaneously thought outside the box when those geese shut down those engines. I am typed rated and current in the A320 family of aircraft. Starting the APU was the first thing he did, however, that wasn't procedure. The procedure for a dual engine failure has the APU in a all engines fail event started as the forth checklist item. This decision probably saved the lives of every man woman and child on that aircraft. Everything on the Airbus is electric. The emergency power systems are designed to function when there is more time for the crew to digest the situation and come up with a plan, they're meant o give you a chance at survival from cruise altitude, not right after takeoff. We basically lose 90% of every electronic resource we have in the flight deck if both engines shut down. The APU restores 100% of the aircraft's electrical system to full functionality. That actually enables the aircraft flight control system to remain in full operation under what's called "Normal Law." Had they run the procedure by the book, it's very likely that the flight control system would have degraded all the way down to basic "Mechanical Backup" which, I can tell you first hand from simulator training, would not have been manageable that close to the ground. That crew threw the book out the window because of human instinct and judgement. They saved everyone on board. AI can't do that yet, period.

What Sully also did was make a decision of self-sacrifice. That aspect got little attention by the media. The numbers of successful transport category jet aircraft ditchings are dismal. Flight 1549 was the only one ever to happen where everyone lived. Sully knew that, but he made a choice in an instant he had to make, put that jet into an icy river probably killing all if not most people on board, or try for land and definitely kill everyone on board and probably a lot of bystanders on the ground.

Pilotless passenger airliners? That's a LONG WAY off.

Pilotless cargo aircraft? That's another conversation entirely as that's already here.
Modern airliners can almost fly themselves. After the flight plan has been inputted into the FMS the plane can practically fly itself. And if I'm not mistaken, in certain situations autonomous landings do occur. Computers have become an integral component of modern aircraft. Modern fighter jets wouldn't be able to achieve the performance they have without modern computer systems (controlled by wire). With that being said, when diversions or emergency scenarios occur you want a well trained (and rested) human being at the controls. Technological innovation is progressing faster than ever, the time between one innovation and the next becoming smaller, I think it's only a matter of time before AI would be able to act as an actual pilot. However, I doubt it'll be anytime soon.
 

Similar threads


Back
Top