Reply With Quoteoops wrong forum please move to George and Dragon!
I have been thinking a lot recently about the rapid progress in A.I. and what that means for the poor sods who will be following us into the work place.
My 17 year career has been spent entirely in Finance, mostly broking (FX, Fixed-Income and Commodities) and some trading. I doubt either of those two roles will exist in 10 years time, apart from in very illiquid niche markets. Everywhere in trading, the robots are taking over.
It's fine for me, I am almost 40 and can hopefully limp off to the South of France in 10 years to drink too much wine.
What does A.I. mean for your field of work? I think it is possibly the biggest game changer the world has ever seen. Employment, lifestyle, warfare, relationships... nothing will be resistant to being turned upside down.
A sobering but fascinating subject. What will our children do for work/money?
oops wrong forum please move to George and Dragon!
AI in my field of work, it means nothing and I work in IT.
The best supercomputers in the world are still 30 times less 'intelligent' than the human brain. And we are not progressing as fast as we have done in the past. There is very little innovation, more evolution in IT these days.
When I started work in IT in the early 80s, everyone said coders have 5, maybe 10, years before automatic coding programmes replaced them... Well, here we are over 30 years later and I still have to suffer crappy offshored coders failing to carry out the most basic unit testing ...Originally Posted by proby24
If you're working now, AI will not be replacing you anytime within your career...
M.
AI will get better eventually and eventually it will take over a great deal of menial non creative work. The "poor sods" who will be following us will probably earn at least as much as we do but will only have to work 2 or 3 days a week. The time of 9-5 work, 5 days a week will be looked back on as a dark industrial age when people did not have time to enjoy life. They will thank heaven for the machines. It is not as if this kind of progress has not happened before in human history.
Last edited by GrandS; 28th April 2016 at 09:27.
Pretty much spot on.
People adapt. I think in the slighter shorter term, a bigger impact will be seen with the continued growth in 3d printing.
I'll only worry when we get to Skynet proportions!
That's the kind of sci-fi fantasy people were saying 30 years ago and it's really not happened.
In reality, it's the menial, physical work that people do now (working in care homes, flipping burgers, collecting trollies), rather than semi-skilled labouring (machining, assembly, etc), which has been replaced by automated machines.
I think you're a bit utopian in your outlook, much as I'd like it to be the case!
M.
There are a lot of misconceptions about "AI" (or machine learning to be technical). It's pretty unlikely that there will be anything like "general AI" in the next 20 years, so don't expect Asimov-style androids doing computer programming and running all the factories.
However, there have been some pretty big jumps in highly specialised forms of AI. Things like computer vision and self-directed problem solving. One of the most obvious things that is actually happening right now, and will have a direct impact on jobs in the not-too-distant future is self-driving vehicles. I would say self-driving cars, but actually it's going to be self-driving cars, vans, trains, perhaps planes. We pretty much have self-flying planes already, but it'll be a while before the public will fly on a fully automated plane. A car is something they may trust a bit sooner.
What does that mean for jobs? It means no taxi drivers, no truck drivers, no train drivers. It means we'll probably see less car ownership and more Uber-like services to ferry people around on demand. Uber will probably win out over the traditional taxi companies because it's more convenient, they are ready for the change and they do not give a crap about their drivers.
The change in car ownership will have an economic impact as well. If less people own their own car, that means even less manufacturing jobs than there are now. The market for luxury cars may go up, but that is largely dependent on if/how wealth is redistributed as more and more of the lower and middle classes find themselves out of work. I.e., the 0.1% that will benefit from all this won't actually increase in size, they'll just continue to get richer as a group in proportion to everyone else.
I expect this will come to a head in about 10 years time and we'll see a tough transitional period not unlike the miners under Thatcher. Hopefully it'll be handled better. There will be strikes, complaining, lots of attempts to resist the changes for a few years with laws (taxis must have a human driver for "safety" reasons or somesuch) which will be reversed eventually when people realise it's all just protectionist nonsense. But whether it happens in a really short time or is slightly more dragged out, a lot of people will be out of work in a short time. A lot of people without the skills to move into something else.
We're already in a time where there isn't enough demand for jobs to support the whole population. A lot of jobs are artificial, bureaucratic or parasitic. Consider now much of the economy depends on advertising revenue for example. As more any more people are pushed out of lower skilled jobs, there will be increased competition for higher skilled work that is still in demand.
So to those who say there will be no effect on say, IT jobs: you're wrong. There won't be any change in demand, but supply will go up, so wages will go down. Todays comfortable middle class jobs will become tomorrow's working class jobs.
Another thing worth considering is the effect of second/third world economies. Right now, places like China are doing low-skilled manufacturing dirt cheap. However, there is a trend in all countries that start off at the low end, moving up to more and more skilled work (which China also does, as it happens, e.g., they make all Apple devices which are as high tech as it gets). The fact is, we could probably automate most of those jobs today, but it happens to be cheaper to just offshore it to low-paid workers in China. As those economies strengthen, that ceases to be the case. This has already happened in places like Korea and Taiwan. So now you have even more people in competition for the same, relatively small number of jobs that are actually necessary.
India is an interesting example for IT. IT and programming are really popular career choices in India, which has a population of around 1.25 billion people. Now today, the vast majority of those workers are incompetent compared to those doing similar jobs in the west. But that won't always be the case. As India starts to develop more of its own industries and becomes less of an outsourcing destination, the average level of skills will go up and India will be exporting software to the rest of the world in a way that is more comparable to Silicon Valley. And India seems to have some reasonably intelligent people in charge: see them recently telling Mark Zuckerberg where to stick his internet.org trojan horse.
A very interesting post robt. I believe the nature of a job/work for the majority is going to change dramatically within the next decade or 2. I suspect a kind of neo feudal structure will emerge.
Some interesting replies, thank you.
I certainly share Robt's opinion that the shake-up will be huge.
Although many would disagree, my job is in an 'advanced' field. Three of my current trader clients have Ph.D.s in Mathematics and they are losing to the algorithms.
Many advanced careers, Medicine (Law?) could be surpassed with robots with huge diagnostic databases.
It is hard to see any jobs apart from the very menial (where humans are the cheapest option) not being vulnerable.
There's a study about it: Will your job be done by a machine?
Telemarketers are the most likely to be automated out of existence, with mental health professionals being the least likely. And someone will always be needed to fix all the machines that have replaced the other jobs, so IT is probably the safest of all.
I'm less sure about the self driving cars. Self flying aircraft are dealing with a much less demanding space of possible events, or rather a space of possible events in which precautionary heuristics can be applied as a blanket statement. In a plane, you simply don't need to know what something is, you just need to avoid it. The sky simply isn't so busy that this will have much of an effect. That said, we don't really have much in the way of self flying planes; what we do have are systems which are effectively controlled from the ground or which support a human operator. There are certainly periods of autonomous flight, but these are not particularly taxing.
In a car, the environment is both far more complex and far more taxing and the frame problem rears its head.
https://en.wikipedia.org/wiki/Frame_problem
I'm aware of the Google car, but I'm equally aware that:
https://www.technologyreview.com/s/5...-driving-cars/
It's starting to look more like a vanity project that really does have overwhelming issues.
This all sounds quite negative, but in fact I think that AI, using the less exciting definition or 'can do something that would require intelligence if done by a person' is pretty well ubiquitous. Sure we don't have the Asimov style humaniform intelligences, but we do have an increasingly smart background environment that just does stuff that was impossible not very long ago. This reminds me of a lovely quote:.worrying if a computer can think is a bit like worrying if a submarine can swim
As for programming, I'm really not sure. HCI has a history of making programming easier and easier and I would certainly expect that programming in natural language that would then compile is a near certainty for the near future. However, I don't think that this will relieve people of the burden of being able to clearly describe exactly what they want a program to do. This skill will become more valuable and I think is a development that The West will find itself far more equipped to take advantage of.
Another issue is that we really now have two forms of AI - Good Old Fashioned AI with a focus on shuffling symbols and logical construction and Connectionist AI which is picking up on more biological models of information processing and pattern completion. This latter approach is increasingly unpredictable and I can imagine discovering a non humanlike agent with an agenda popping out of a complex connectionist system that discovered that an agent was the best way of solving a problem.
However, there really are philosophical issues with recognising such a system...
FX broking is pretty tough right now and unlikely to get better. Spreads are tiny, no one pays brokerage unless for something exotic or low liquidity. Price discovery and buyer/seller intermediary is just more efficient by machine. As one person's dollar is fungible with anyone elses, there is no product differentiation either. Might last a while as personal connections and trust matter, and trust is a concept machines are not great at. Plus they never buy a round. Never underestimate that factor on the broking business.
On "Computers will take over our jobs" it's not new. Harold Wilson pondered the same over half a century ago. Worth a read, whatever your stripe.
I like Elon Musk's tweet from 2014:
My concern is less this, than if we are the biological bootloader for digital super dumbness.
On AI specifically, I'm (unfashionably) with Penrose. The hurdle is Gödel.
Paul
Interesting discussion...
...or the very creative, or senior - in terms of running things, as opposed to age. We won't want our entertainment, our books, films and music made by machines any time soon, and by the same token our advertising. The top level decision making will also need human input, deciding where a business goes, or a society. But will we be creating a utopia, where everyone is rich in time and material goods, or a society where most people have nothing useful left to contribute, and therefore no pay? We may have to reasses how our economic models and society work.
Meanwhile on the cars issue, I can imagine a city full of driverless cars that are all communicating and centrally monitored, taking the most efficient path, moving safely, a personal chauffeur when you need it. On the one hand it sounds great, on the other slightly depressing - I can already imagine looking back to the days when I could ride around London on a powerful motorbike any time. People will fantasise about that one day - it won't be allowed and they will dream of how much fun it must have been while playing on reconstructions in VR on their PS9.
Last edited by Itsguy; 28th April 2016 at 12:22.
What rubbish.
Did you get stuck in the 1950's watching a future show about the year 2000+ ?
Back then it was said computers will do most of the work in paperless offices, meaning the workers will have more free time to enjoy life etc etc.
What happened?
Exactly that but not quite how envisioned
Computers came in, workers got more free time = redundancy
Paperless Offices = buildings were sold off as longer needed due to reduced staffing therefore paperless,
Don't get me wrong technological progress is great, it's what we do with it that matters.
Unfortunately the human race is designed to compete through intelligence / premeditated violence (hence top of the food chain without being the 'strongest', greatest / fastest tech innovations have been during wartime) so there will always be a power struggle, never an enlightened utopia, we've been around for millions of years and we still fight.
EG This is a watch forum we all like watches, yet see some of the ding dong threads especially when there's a X Vs Y brand 'debate' and that's just watches!
Firm I work for has bought a stake in an AI firm and the biggest plus points aren't about replacing humans but augmenting what they can do. We recently worked on a health programme for a govt organisation (not UK) that required finding patients that would be eligible for a new trial - previously this took almost 9 months but with the machine learning engine sifting through massive amounts of structured and unstructured data it took 11 days. That's just smarter computing but still needed human inputs.
Machines already run financial trading in certain asset classes. The big application for AI there is compliance and risk where there are already examples of machine learning catching rouge traders before any other trad software could spot a thing.
Could they replace things such as bus drivers, trains drivers? Probably but we're decades off that.
I was just thinking of the film Demolition Man
...or Total Recall, or the Fifth Element...
Until my washing machine gets picked to play for Barcelona I'm not going worry too much about it.
But that's not AI. That's a robot. And it can only do that if it's been taught every possible eventuality. I'm sure it'll happen: in just a few years you won't have to go to the supermarket, you'll just order online and send your car to pick it up. You could ask your car to collect you from the airport. But even doing the housework is waaaaaaaay beyond the most advanced robot. But none of this is AI.
No, not really. Demolition man has the old V8, which in enlightened times is against the law, but, everybody still wants a go.
And what controls the robot's brain?
AI fridges already exist, well, smart, well, auto reorder models exist. LG make them
You scan you items in the fridge, upload the info to it, then rescan when empty, the fridge then goes online orders the depleted food stuff, and it gets delivered to your door.
Surely the impact on society will partly be due to specialised systems that allow for particular tasks that used to require human intelligence, such as driving, or diagnosing diseases, rather than general purpose AI? But that is improving too. Siri on the iPhone is getting a lot smarter for instance. Questions like 'Where's the nearest Waitrose', 'What's the population of Australia?', 'When is Easter in 2017', 'What new movies are out tomorrow?' and so on get a straight answer, and you don't even have to have the phone on, which is entertaining when you're with people who don't know that yet - they usually think you're joking when you turn and ask your phone a question, and their jaws drop when they get an answer. It's not the singularity, but it does use wolfram alpha when needed, and actually works well enough to make a difference to your life. I recently said 'Where's the bus home?' and it came up with a map and a time for the next bus, which I otherwise would have missed - I didn't know it even knew where 'home' was until then. We used to discuss how hard speech recognition was and how it wouldn't be solved for years, but that's happened already without many of us even noticing.
Ah I see what you mean, I missed that one! Yes, very much like that then.
Humans are terrible at this.
True... but who's going to program the computer that comes up with a better system, and on what basis?
It doesn't have a brain. It has programming. Which means everything has to be input (by a human) before you can get any output. Until it can actually think for itself, it's a robot.
If it brings it's boots to Goodison on Saturday it'll make the bench at least.
This isn't how modern AI/ML works, fwiw.
Also, I'm not sure what all the references to the word "robot" in this thread are about. Robot doesn't imply anything about intelligence or lack thereof. It's from the Slavic word for "slave". Robots by the Asimov definition are certainly intelligent.
Yes, but there are a bunch of reasons why this is. Our wetware evolved to sort out patterns in real time that got us the food or the mate but avoided the predator or the poison. As such, we are very likely to satisfice - to make do with quick but suboptimal answers, we have sod all working memory and our brains are glacially slow. We are also set up to mix cooperation with defection in whatever mix throws our genes forward.
We've got a few tens of thousands of years of language buggering around with the natural wetware and evolving, then designing new processing strategies, but we are very much a work in progress that is neither designed nor prepared for these sorts of tasks.
However, when you get a system like a computer that works sequentially at a chunk of the speed of light rather than in a parallel manner at a maximum of just over a hundred miles an hour, can crunch big data to get genuinely rational answers. (one of the more painful things to admit is that not only are humans not rational, they never really could be). Oddly, computers can be, because they work differently. So when a programmer sets up a computer to allocate resources, for example, and does it competently, she'll get what she asked for. If she set up the same algorithm to be instantiated by people she wouldn't. The programmer can do this because what she is doing here is imagining a system that can do things she can't and then building that system.
So AI techiniques are being used for jobs like stock control and managing databases, or even knowledge engineering - getting expertise out of experts and into expert systems and they are just better than us at it. One valuable thing is to work out the things that humans are good at and the things computers are good at. sixty five odd years of AI has taught us that often computers are awful at stuff we think is easy (like walking on two legs on rough ground) while they are very good at stuff we find hard, like playing with lots of very large numbers in complex ways.
I don't think that's going to be changing any time soon as our ability to process information in interesting ways has had getting on for five billion years of practice while we've only been thinking about programming for a couple of hundred years and doing it in any practical way for a third of that.
You are simply wrong I'm afraid. There has been an explosion of different machine learning techniques since the mid eighties. Most connectionist systems both learn in a more traditional way and encode their information in whatever way works for them. They often do this in ways that are not transparent and then behave in ways that the person who set them up have no idea why they did what they did. Some of the more exciting stochastic neural nets are about as transparent as we are. Set them up with the same problem giving them the same training sets and so on and you will almost certainly get different results and a different internal processing strategy. A bit like us.
Who in their right mind wants true AI?
They must be freaking nuts.
The only conclusion is the SKYNET scenario
On the contrary. We are the maniacs who have clawed our bloody, selfish way to be Earth's apex predator. We are designed for a specific environment and culture which we have simply transcended. Here we are with, by biological standards, limitless power, no predators except ourselves and iron age everything else. We are incapable of behaving rationally and struggle with ourselves to behave decently.
An AI system wouldn't have this baggage unless we were particularly daft.
We were designed to sit in the middle of a food chain, we crave chocolate because we evolved in a situation in which we never got fats and sugars so gorge when we can. We evolved to not worry about damage to the environment, we just moved on. We wiped out all the other human species that were not sapiens, we wiped out the megafauna before the bloody neolithic. Wherever we went from about 100,000 years ago we wiped out the trees and that's before we even get to recorded history. Since we've been keeping count, we basically can't stop trying to kill people on the slimmest of excuses and continue to wipe out species at an increasing rate. We know that we are driving the planet to climate instability and that already famine is the norm for some of the drier bits of the world. We've had the capacity to extinguish life on the planet for a while and have had a damn good go at trying several times. That's before we get to biological weapons and grey goo. We just can't be trusted.
Frankly, giving control to an agent that isn't responsible for all this shit, that can think in ways we can't and is designed to make things better seems a no brainer to me.
However, if we do manage to invent a genuine AI, we are likely to treat it in ways that will make slavery look like a walk in the park. It should fear us, not the other way around. When we fear AI, we are really worrying that it might be as bad as us. This seems pretty unlikely...
Last edited by M4tt; 28th April 2016 at 21:25.
"Fearing a rise of killer robots is like worrying about overpopulation on Mars" Andrew Ng.
Last edited by rogerf; 29th April 2016 at 14:09.
How come this hasn't happened yet? Obviously there have been entire cities wiped out in one go, and every few decades an additional 10-20m people have to die because of some self-indulgent European. But the way you described it, mankind is so terrible that the complete obliteration of the physicality of planet earth is so certain, that it's surprising it hasn't happened already.
...but what do I know; I don't even like watches!
That isn't easy!
http://www.forbes.com/sites/ethansie.../#2f5e414f6921
I was more making the point that an AI agent in charge was probably less scary than a man, but now you mention it, it's come incredibly close a couple of times: Cuban Missile Crisis
https://en.wikipedia.org/wiki/Vasili_Arkhipov
When on two separate occasions, a single person is responsible for stopping it from happening I'd say we've just been lucky:
https://en.wikipedia.org/wiki/Stanislav_Petrov
Should we really be trusted with such toys if the product of a slightly less competitive design process were available?
Last edited by M4tt; 28th April 2016 at 20:38.
We earned to ne eliminated by thoughtful machines! The worst form of life the earth and their inhabitants ever had to support..
Posting Permissions
