odysseus2000 wrote:Hi Onthemove,
Sure computers since their invention have been much faster than humans, the difference now is that at some level they can think or perhaps more correctly mimic thought. That is new.
I'd strongly disagree with that.
In terms of thought, the brute force trying out all paths methods of 'good old fashioned AI', as used in the original chess programs that beat the grandmasters, and that you find in your typical chess computer, are probably actually more akin to 'thinking'. These actually do consider many paths into the future, and evaluate them.
On the other hand, deep learning / convolution networks are (when used e.g. for vision applications) more closely related to the optic nerve and visual cortex. These aren't areas of the brain that would ordinarily be considered as dealing with 'thinking'. Rather they are areas of the brain which perform subconscious / unconscious processing of input data.
I suspect neuroscience has come on a bit since I was at uni, but back then one of our course text books was actually a neuro science book that described the workings of the visual cortex, and the experiments people had done on (live) monkeys to identify what parts of the visual cortex responded to what types of visual stimulus.
The empirical understanding that people have about how deep learning networks are working, is pretty much along the same lines as how it is believed the brain processes visual information from the eye through the optic nerve to the visual cortex at the back of the brain. But all that visual processing is considered to be 'low level' i.e. unconscious - consciously you have no awareness at all about how the brain has processed the information from your eyes. Consciously you just see the reconstructed representation of the objects in the world - much like the reconstructed representation a self driving car has of the cars around it.
Moreover, AI tends to also talk in terms of 'biological plausibility'. This means, for any algorith, or suggested approach to atificial intelligence, how plausible is it that it could be representative / modelling how the brain actually 'thinks'.
And when it comes to deep learning and convolutional neural networks ... ok, the 'deep' aspect is probably more biologically plausible than earlier 'shallow' networks. And the way the network (seems to) build up further an further aspect as you go from the inputs, also matches a lot how the visual cortex, etc, is structured.
But the 'convolutional' aspect which is now used widely in deep learning, at least when applied to vision applications, is a step back from biological plausibility. That's to say there isn't (at least not currently) any plausible explanation of how a 'convolution filter' could be trained in the brain, and then repeatedly applied across the entire visual field.
"...it seems, at least to me, not impossible to believe that many of the managers could before too long be relegated to the role of checking the AI and later to no role at all. I want to believe that AI will be nothing new, that humans will still be needed, but extrapolating forwards I am not so sure."
There is definitely a large fear of the unknown in that position :^)
It is true that some jobs will become redundant. Taxi drivers, HGV drivers, etc, are the most obvious.
But also low paid jobs that basically use the human vision system. For example, sorting rubbish for recycling. That is an incredibly difficult job for old style vision systems to do - that's why it hasn't been properly automated before now. That's why people are paid on the minimum wage to stand there all day identifying and picking particular categories of rubbish and sorting them for recycling.
But in reality, all the person is being employed for is their vision system - their eyes, optic nerve, visual cortex, that is able to classify what they can see on the conveyor in front of them, and their two manipulators (arms) that enable them to pick the identified items off of the conveyor.
That should now be definitely automatable - and with any luck we can go back to stuffing all our rubbish into a single bin and letting the AI sort through it at superhuman speed (the speed of some industrial robots these days is phenomenal - they can look like a blur when moving at full speed! You certainly could react if you had your arm in the way and it suddently started moving)
Fruit picking is another one which the new AI can now enable to be fully automated. The ability to identify fruit on the plant / tree, decide if it is ripe, then coordinate a robot arm to pick it. You could easily imagine such AI guided arms on a rail alongside the grow bags in a massive industrial greenhouse, constantly traversing, looking for the ripe fruit and picking it when found. That is a very real prospect now. There are already companies working on this.
So where jobs are dependent upon the human providing the identification / classification, etc, and / or coordination with a manipulator, there is the likelihood of them being completely taken over by AI and robotics
There will also be a loss of jobs with old style computer vision experts who think that they can demand a high salary and large respect for telling you you need to spend thousands of pounds for a very accurate lighting system to allow you to read barcodes, QCodes, etc, and identify things. We have one such guy where I work. One day the managers are going to twig that they are being spun a nonsense when they realise their £100 phone is able to read any barcode you wave in front of it in any lighting condition, yet this guy tells them they need thousands of pounds worth of highly accurate lighting. He is likely to become a victim of the current AI progress.
In most other areas I genuinely believe that AI will be a partnership. That in actual fact workers will benefit from working with AI. The AI will handle the mundane or difficult stuff while allowing the human workers to be more creative. The AI is just another tool available to the worker. Or something that allows the worker to do their job in a different way.
For example, I believe Microsoft have already demonstrated AI being able to take a sketch and then infer what was being sketched and create a proper representation.... even simple things like when you are having a brainstorming session on a white board. If you use an 'AI enabled' white board, you can scribble a few really crude blocks, and the AI will transform them into proper squares with straight lines. It will read you scribbled text and translate it into proper typed text, which can then, for example, be searched for later. You are still the one providing the creative content. The AI is just making your job easier.
It may even try to do some basic analysis of your diagram so that at a later point you can ask the computer in quite vague terms to "Find you the early diagram for new widget we were developing" and it would then understand what you mean, and allow you to find the diagram even if you never properly organised it into folders, etc, yourself.
There are tools in development that can use AI to quite literally listen to the engine in your car as you drive along, and it is able to tell you various things like if the air filter needs fixing, or if the head gasket is blowing - purely from the sound that you hear in the cabin. In reality, you'll probably still need to take it to a garage to have the work done. So it isn't going to put your garage mechanic out of business. But the garage mechanic may be able to use the AI to help him better diagnose what's wrong.
Similarly in medicine. There are plenty of examples now of AI systems being able to perform diagnoses faster and more reliably that leading specialists in their fields. That will help avoid misdiagnoses. The AI isn't going to do the surgery for them. The surgeon is still needed. He may even be assisted by further real-time computer vision algorithms that real time highlight areas of interest while performing surgery. You could image that the surgeon might even use an augmented reality head setup with AI providing additional analysis of what the surgeon is seeing. But it is unlikely you'd completely get rid of the surgeon... you'd always have to have someone at least supervising any 'automated' surgery.
In video processing, Adobe has created the ability to automatically segment items out and auto-fill the hole that is left. This isn't going to put people out of a job - the artist still needs to decide what bits they want to cut out, etc. It is simply now much quicker and easier for them to do that. But they won't be out of a job. In reality, the lower cost, ease, etc, of doing this editing will simply mean more people can make use of it.
In effect, rather than putting managers out of a job, my gut feeling is that in human terms we are going to see a much higher percentage of 'managerial' jobs. Whereby the 'manager' simply creates and defines tasks for the AI workers under them. I think you'll effectively see a world in future where there are more chiefs than indians - genuinely so because the humans will be providing the coordination and adaption.
"there is a clear separation between them and their less Ai powered competitors such that at some level AI seems to be doing something powerful and new."
The new AI definitely povides something new. It's definitely not something that is going to be here today gone tomorrow. It is already becoming ubiquitous. For example consumer cameras aren't going to lose the capability to identify a face and treat that as the subject.
And using deep learning to spot patterns that people haven't seen is always going to give an advantage if you use it in addition to existing techniques as well.
People who don't use it (directly or indirectly), will definitely be left behind in almost every industry.
It is going to re-structure the economy. It is going to re-structure the jobs market. There's no doubt about that.
But with (e.g.) the NHS stretched for resources, and the majority of people working all hours to earn a living ... I'd say most people would be glad of some more free time.
AI could allow people to produce the same in far less time, providing more time for family and friends while still retaining the same material standard of living.
I guess the question is how this gets handled by society. Should the eperts work 50hour weeks to pay tax, while those who don't bother then take a universal income for doing nothing but enjoying themselves?
Or perhaps instead should we move to a 4 day week, so that the jobs and workload gets shared around, and we all get more leisure time. (France on their 35hr week already manages to produce in 4 days what it takes the brits 5 days to produce, and they have more leisure time)
Or perhaps 3 months statutory holiday per year. Again so that work gets shared around, but we all gain leisure time. I'd happily have more weeks holiday per year!
Or perhaps we could retire at an earlier age.
(Double bad luck though for anyone unfortunate enough not to survive to enjoy that retirement)