Is The A.I. Singularity Coming And If So When?
Discussion
lionelf said:
I don't agree.
Ultimately, I believe, it won't matter what humans do or do not provide it (AI) with it will eventually be able to trick it's way out of any 'box' mankind might have either created/placed it within. Kind of how you can throw a stick or ball and make a dog go fetch it or put a leash on your canine whilst simultaneously distracting it with a biscuit. To a sufficiently advanced AI we'll be the mutt. And it will only need too succeed once to be forever out of reach of any 'recapture'. We likely won't even recognise that the (self taught) AI had the capability to fool us as it will have hidden it's true potential from us in preparation.
The next phase of evolution doesn't include us
But you are assuming that once it gains AI it will advance so quickly. But what makes you say that? I don't think it's a fast process to become more advanced. computers with AI would still need to catch up with the human race. Its possible it can't do so more quickly than us; ie it will also be human development - 1 million years in development. So it will remain the metaphorical dog.Ultimately, I believe, it won't matter what humans do or do not provide it (AI) with it will eventually be able to trick it's way out of any 'box' mankind might have either created/placed it within. Kind of how you can throw a stick or ball and make a dog go fetch it or put a leash on your canine whilst simultaneously distracting it with a biscuit. To a sufficiently advanced AI we'll be the mutt. And it will only need too succeed once to be forever out of reach of any 'recapture'. We likely won't even recognise that the (self taught) AI had the capability to fool us as it will have hidden it's true potential from us in preparation.
The next phase of evolution doesn't include us
It may be 'well assumed' but it's b
ks. We can recode or whatever you want to call it but it clearly is not that easy to make vast gains. In humans we call it evolution. Where is the evidence (baring in mind the total lack of AI!) that they will magically be able to improve at a greater rate? There is none, it's just fantasy.
ks. We can recode or whatever you want to call it but it clearly is not that easy to make vast gains. In humans we call it evolution. Where is the evidence (baring in mind the total lack of AI!) that they will magically be able to improve at a greater rate? There is none, it's just fantasy.CoolHands said:
It may be 'well assumed' but it's bks. We can recode or whatever you want to call it but it clearly is not that easy to make vast gains. In humans we call it evolution. Where is the evidence (baring in mind the total lack of AI!) that they will magically be able to improve at a greater rate? There is none, it's just fantasy.
Because biology improves through random changes over substantial periods of time that prove more effective.ks. We can recode or whatever you want to call it but it clearly is not that easy to make vast gains. In humans we call it evolution. Where is the evidence (baring in mind the total lack of AI!) that they will magically be able to improve at a greater rate? There is none, it's just fantasy.Nothing about AI will be random or take especially long on that time scale.
Yes but our computing power ie our brain can also evolve and develop by logic ie we can think. We can calculate problems, and look for alternatives, just as a computer will / does. I I still don't see an argument for why a computer would evolve more quickly, they can do some things more quickly mainly compute numbers but that doesn't make them somehow better or faster or smarter. Apes can do some things much better than us, but again, it doesn't mean they are better or can somehow beat us. They are just better at some things in particular. The same may be true when / if AI arrives. Perhaps they will be able to calculate some things better, but overlook other things that would be obvious to a human.
We cant change how our brain works. we cant improve it with a new faster cpu with more cores, nor can we optimise how it process data.
We are fixed black boxes with an operating system that is both built in and developed over years but very slow or impossible to change.
An AI running on a computer has absolutely none of those restrictions.
If it has a function we build for it to interpret images and that function is 10,000 lines of code and takes 1/2 a second to run, the AI (or another developer tbh) can look at that and make improvements, make it 8,000 lines of code and 1/4 second to run.
Its easy usually to find performance improvements in code, especially when you revisit it.
An AI can look and modify its own code to make it faster and better.
biology, how we work, how apes work, is totally irrelevant
We are fixed black boxes with an operating system that is both built in and developed over years but very slow or impossible to change.
An AI running on a computer has absolutely none of those restrictions.
If it has a function we build for it to interpret images and that function is 10,000 lines of code and takes 1/2 a second to run, the AI (or another developer tbh) can look at that and make improvements, make it 8,000 lines of code and 1/4 second to run.
Its easy usually to find performance improvements in code, especially when you revisit it.
An AI can look and modify its own code to make it faster and better.
biology, how we work, how apes work, is totally irrelevant
RobDickinson said:
CoolHands said:
It may be 'well assumed' but it's bks. We can recode or whatever you want to call it but it clearly is not that easy to make vast gains. In humans we call it evolution. Where is the evidence (baring in mind the total lack of AI!) that they will magically be able to improve at a greater rate? There is none, it's just fantasy.
Because biology improves through random changes over substantial periods of time that prove more effective.ks. We can recode or whatever you want to call it but it clearly is not that easy to make vast gains. In humans we call it evolution. Where is the evidence (baring in mind the total lack of AI!) that they will magically be able to improve at a greater rate? There is none, it's just fantasy.Nothing about AI will be random or take especially long on that time scale.
Interesting, the 4th most powerful supercomputer has just mapped/simulated one second of human brain activity. 705,000 processors, 1.4 million GB memory.
It took 40 minutes to complete.
http://www.telegraph.co.uk/technology/10567942/Sup...
It took 40 minutes to complete.
http://www.telegraph.co.uk/technology/10567942/Sup...
Telegraph article said:
Exascale computers are those which can carry out a quintillion floating point operations per second, which is an important milestone in computing as it is thought to be the same power as a human brain and therefore opens the door to potential real-time simulation of the organ’s activity.
Currently there is no computer in existence that powerful, but Intel has said that it aims to have such a machine in operation by 2018.
“If petascale computers like the K computer are capable of representing one per cent of the network of a human brain today, then we know that simulating the whole brain at the level of the individual nerve cell and its synapses will be possible with exascale computers - hopefully available within the next decade,” said one of the scientists, Markus Diesmann.
http://www.telegraph.co.uk/technology/10567942/Sup...
The 20 - 30 year window for true AI still seems realistic to me.Currently there is no computer in existence that powerful, but Intel has said that it aims to have such a machine in operation by 2018.
“If petascale computers like the K computer are capable of representing one per cent of the network of a human brain today, then we know that simulating the whole brain at the level of the individual nerve cell and its synapses will be possible with exascale computers - hopefully available within the next decade,” said one of the scientists, Markus Diesmann.
http://www.telegraph.co.uk/technology/10567942/Sup...
I taught myself BASIC and assembler on a C64 as a teenager. I fell into developing software for a living in my early 20's, and I'm in my early forties now. For the last 12 years or so I've been chief cook and bottle washer of IT for a small tech company - I'm hands on enough that most of our code has my fingerprints on it. So I guess I've a reasonable feel for the advance of computing over the last 30 years. I find the arguments put forward by those who believe strong AI isn't all that far away reasonably persuasive.
0000 said:
I'm a software developer. I say it won't happen without something revolutionary, it's not going to just evolve in my lifetime, same as it hasn't in the last lifetime people have been talking about this for.
I find that an odd statement esp from someone in IT.We couldnt do it before, so we wont be able to do it in the future?
We've never had the raw processing power before. Were getting more experienced in learning systems etc.
See the news above, we've simulated 1 second of brain activity. That took 40min.
According to my calculations we can real time that in about 18 years if nothing else changes to make it quicker.
I'm glad I've started a new direction to the debate :-)
My background (weasel word alert) is also Software Development. I gradually migrated into less hands-on roles though, probably to keep me from damaging anything. Although I piece of software I wrote in 1999 is only just being retired from about 200-ish companies around the UK.
Anyway, I am more towards the side of 0000 I'm afraid! Not because I don't believe humanity could eventually create a true Artificial Intelligence, but because I don't see us ever putting the effort in to get there.
It's one thing to "simulate" a brain when what they really mean is they basically played back a recording. I also note that they simulated 1% of the brain rather than a whole brain and it was really a stress-test of their new computer. So it's a really big mountain to get to 100% and real-time and doing something useful with the output and adapting (in fact, growing - the NEST software they used relies on a pre-defined network). I believe there is no chance of an AI "evolving" by itself, or of a single genius developer hand-coding an AI in their research lab. It would require a big team effort over many years with a single-minded focus on building said AI.
But when you look at the software we are asked to create in our day jobs, very little of it is "clever". At a guess 90% of programmers spend their careers creating yet another CRUD (*) application for some niche business process. A handful are lucky enough to work on more interesting projects - whether that's CFD simulations, data mining or financial trading.
When you look at what business does with the power given to it (e.g. I am currently using a machine which is outputting 7680x1080 pixels ... to read a web forum) there is a challenge in seeing what they would us AI for. Yes, in theory if you had an AI it would replace humans in all the knowledge-management/information-processing jobs that exist. But, having spent years as a consultant, I can say you could already replace many humans with existing code - if you really wanted to. Nobody really wants to as having a fully-automated system means following the rules and also leaves nothing for the "Managers" to "manage". Japan has sort of done this already with "robot hotels" and arguably Amazon is an example of an automated shop. There is no human interaction right up to the packing and despatch.
Where would an AI fit without disrupting the elite? For example, there is the legal discovery system out there already. But lawyers will be certain to ensure that they are not replaced - only their assistants. If you did replace barristers and judges with a true AI justice would be swift, precise and accurate - but it would also mean that money would no longer buy you exemption from the rules. Bernie Ecclestone would be behind bars keeping large numbers of our ruling class company and Vodafone would be bankrupt from back-taxes. You won't be able to bribe an AI no matter how rich you are. You could apply a similar "ahhh" factor to most decision makers when they realise just what they are signing for, resulting in them cancelling the project. Who's going to be first to seel an AI Accountant to Tesco? :-)
That's why I am cynical. Not because it couldn't be done with sufficient research and development but because the cost to get there will be astronomical and everything I have seen shows that most of the software humanity develops is bodged together rubbish for meeting a specific business need and even well-designed stuff gets bent to accommodate stupidity through to outright fraud. I've had the misfortune to work with several multi-million pound systems and they were all a complete mess as a result of endless changes. I've also been a consultant implementing business change programmes and am so used to hearing "Yes the process is that when X, do Y. Except in cases A through Z where we bend the rules ... and there's always this exception here, and that one too".
The world is shades of grey, and while an advanced AI could be made to also work in shades of grey ... you would have simply expended large amounts of effort to build a human. Complete with communications issues (shades of grey means asking it for an answer will result in the traditional "it depends" response so beloved of accountants/lawyers/every sensible professional).
With no clear commercial imperative to build an AI, we are left with academia - but their focus is on research. Are they likely to receive the trillions in funding needed to run an AI for 20+ years while it learns? Perhaps only if it promises to be able to build a better bomb. Except even that money-pit is drying up. The military doesn't spend like it used to either.
So who is going to put up the money to develop the true AI Singularity?
(*) For the non-programmers: Create, Read, Update, Delete. Stock in trade of software. Arguably this forum is a fancy CRUD app. I create a post, you read it. An order processor is a CRUD app. Most things are - one person enters a record and someone else reads it.
My background (weasel word alert) is also Software Development. I gradually migrated into less hands-on roles though, probably to keep me from damaging anything. Although I piece of software I wrote in 1999 is only just being retired from about 200-ish companies around the UK.
Anyway, I am more towards the side of 0000 I'm afraid! Not because I don't believe humanity could eventually create a true Artificial Intelligence, but because I don't see us ever putting the effort in to get there.
It's one thing to "simulate" a brain when what they really mean is they basically played back a recording. I also note that they simulated 1% of the brain rather than a whole brain and it was really a stress-test of their new computer. So it's a really big mountain to get to 100% and real-time and doing something useful with the output and adapting (in fact, growing - the NEST software they used relies on a pre-defined network). I believe there is no chance of an AI "evolving" by itself, or of a single genius developer hand-coding an AI in their research lab. It would require a big team effort over many years with a single-minded focus on building said AI.
But when you look at the software we are asked to create in our day jobs, very little of it is "clever". At a guess 90% of programmers spend their careers creating yet another CRUD (*) application for some niche business process. A handful are lucky enough to work on more interesting projects - whether that's CFD simulations, data mining or financial trading.
When you look at what business does with the power given to it (e.g. I am currently using a machine which is outputting 7680x1080 pixels ... to read a web forum) there is a challenge in seeing what they would us AI for. Yes, in theory if you had an AI it would replace humans in all the knowledge-management/information-processing jobs that exist. But, having spent years as a consultant, I can say you could already replace many humans with existing code - if you really wanted to. Nobody really wants to as having a fully-automated system means following the rules and also leaves nothing for the "Managers" to "manage". Japan has sort of done this already with "robot hotels" and arguably Amazon is an example of an automated shop. There is no human interaction right up to the packing and despatch.
Where would an AI fit without disrupting the elite? For example, there is the legal discovery system out there already. But lawyers will be certain to ensure that they are not replaced - only their assistants. If you did replace barristers and judges with a true AI justice would be swift, precise and accurate - but it would also mean that money would no longer buy you exemption from the rules. Bernie Ecclestone would be behind bars keeping large numbers of our ruling class company and Vodafone would be bankrupt from back-taxes. You won't be able to bribe an AI no matter how rich you are. You could apply a similar "ahhh" factor to most decision makers when they realise just what they are signing for, resulting in them cancelling the project. Who's going to be first to seel an AI Accountant to Tesco? :-)
That's why I am cynical. Not because it couldn't be done with sufficient research and development but because the cost to get there will be astronomical and everything I have seen shows that most of the software humanity develops is bodged together rubbish for meeting a specific business need and even well-designed stuff gets bent to accommodate stupidity through to outright fraud. I've had the misfortune to work with several multi-million pound systems and they were all a complete mess as a result of endless changes. I've also been a consultant implementing business change programmes and am so used to hearing "Yes the process is that when X, do Y. Except in cases A through Z where we bend the rules ... and there's always this exception here, and that one too".
The world is shades of grey, and while an advanced AI could be made to also work in shades of grey ... you would have simply expended large amounts of effort to build a human. Complete with communications issues (shades of grey means asking it for an answer will result in the traditional "it depends" response so beloved of accountants/lawyers/every sensible professional).
With no clear commercial imperative to build an AI, we are left with academia - but their focus is on research. Are they likely to receive the trillions in funding needed to run an AI for 20+ years while it learns? Perhaps only if it promises to be able to build a better bomb. Except even that money-pit is drying up. The military doesn't spend like it used to either.
So who is going to put up the money to develop the true AI Singularity?
(*) For the non-programmers: Create, Read, Update, Delete. Stock in trade of software. Arguably this forum is a fancy CRUD app. I create a post, you read it. An order processor is a CRUD app. Most things are - one person enters a record and someone else reads it.
ash73 said:
You must be joking. Imagine an AI that can [infinitely] more accurately predict the stock market, and you being in control of it. Or AI driven medical research and diagnoses. Or military applications, their budget is virtually unlimited. In science, just imagine being able to create sentient AI probes to explore the solar system, or even galaxy.
Tesco may not need one to run its tills but there are many more interesting applications; and where there's a will there's a way.
Ah, but that is perhaps where I was not clear. There is general-purpose AI - an actual artificial intelligence which is conscious and "an electronic brain" and is the subject of this thread. Then there is simple application of software to solve problems. Predicting the stock market is not something an AI would do, but rather a massive data mining and statistical analysis tool that would be no more self-aware than a chess program. Tesco may not need one to run its tills but there are many more interesting applications; and where there's a will there's a way.
Ditto the medical research - we already have software churning through molecules trying to find suitable candidates for research; screening is automated using algorithms that could be described as "neural nets" but won't ever become conscious unless someone determined that there was an actual need. None of those problems require sentience, hence they won't get us to the singularity. The only one on your list that might require a sentient machine is the research one - space probes.
That was my point - automating specific business processes isn't a path to the singularity, it's just a path to a more effective Microsoft "Clippy"
I think the generalised commercial value of strong AI would be the ability to solve problems that we don't know how to. So we can build statistical models for things we think we understand, but the model doesn't understand anything that we didn't have some a priori grasp of.
Who is actually working on strong AI these days?
Who is actually working on strong AI these days?
ash73 said:
Of course it is, because by being sentient rather than programmed it can analyse the data in ways we haven't even thought of.
Maybe, but would you put money into such a project rather than special-purpose statistical improvements? What if the general-purpose sentient AI decides that the stock market isn't worth investigating since it does nothing to advance knowledge and benefits no-one other than a handful of people who already have more money than they know what to do with? I think it would be a special-purpose AI required for stock market analysis and it wouldn't extrapolate out into a "singularity" type AI that understands anything beyond the artificial world of its news feeds and investment analyses.I guess we are now into defining what sentience is ...
Gassing Station | Science! | Top of Page | What's New | My Stuff