Topic: A.I.

[Conocinico]

So AI then, when do you reckon, 30 years? 40?

I say never.

[Alan_X]

Interesting talk but I don't agree that intelligence is simply about processing information. Human intelligence is far more than computation. There are a lot of logical fallacies in there, too many to deal with at 6.20 am, but one of the biggest is when he points to his head and says 'just atoms' and then talks about computers as 'just atoms' as if that is somehow relevant. A rock is just atoms and so is the human liver.

The human brain isn't like the Numbskulls (look it up if you're under 50). The brain isn't a computer inside a load of meat. The brain 'thinks' but it also feels emotions, feels pain, processes memories in ways we barely understand. And it also has motivations based on aspirations, fear, societal/familial pressures, hormones and instinct(inherited behaviours built into our DNA).

He also makes the mistake of assuming that because technology exists it will automatically progress Ina given direction and if something could happen it will happen. To take a few relatively mundane examples from the consumer field you wouldn't want shares in 3D televisions, Google Glass or Kindle at the moment. There's an article in today's Guardian about the decline of the Kindle. The Kindle was going to kill physical books because 'technology'. Seems that people actually prefer physical books. They like the physicality of paper. That you can fold the corner over to save your place, flick through, smell the paper, crack the spine etc. 3D television is dead because it was the obsession of technical departments rather than something that people actually respond to.

At a much grander scale, we landed on the Moon fifty years ago and a total of 12 men have walked on the surface of another world in a four year window. The predictions back then were that we'd have moon bases and be travelling to Mars by now. It didn't happen because exponential technical development is one thing but the basics of rocket propulsion require huge amounts of fuel to break away from gravity and people need oxygen, water and food for the trip. There's also the fundamental question of 'why are we doing this?'

That's at the heart of it. If you think humans are just meat with a logical meat-based thinking machine in its head then a faster silicon chip based logical thinking machine will be more powerful and dangerous. But that begs the question 'why would a computer give a shit?' It has no hormonal impulses, it has no desires, it doesn't feel emotional pain or joy. The talk uses the analogy of humans and ants - 'we just tread on ants so why would a computer not think of humans as ants?' But then why would a computer care about a computer. If it's impossibly intelligent it would be aware that it's just a bunch of circuits. Unless you start straying into HAL territory and start assigning emotion as an inevitable component of increase computational power.

And what exactly would be the mechanism? There's a conflation of AI and robotics here. If we don't want AI to stamp on us, don't give it feet.

The greatest danger we face in my view is not sentient artificial intelligence but non-sentient algorithms that have been designed to control important systems.

And don't even get me started on the super-AI as God nonsense.

Anyway, interesting subject.

[thejbs]

Seems that people actually prefer physical books. They like the physicality of paper. That you can fold the corner over to save your place, flick through, smell the paper, crack the spine etc.

Not everyone. With the exception of books I buy on photography, art or cinematography, I purchase everything for my kindle.  The romantic reasons for buying physical books, for me, loses out to the practical benefits of e-books - especially when traveling, where I like to travel light and go through 3-4 novels. I can still save my place, I can immediately look up unfamiliar words, I can highlight and store passages that resonate with me, I can buy a new book and immediately start to read it anywhere I can get a 3G reception.  Fuck cracking spines and smelling paper... for me, tech wins.

E-book reader sales have fallen because people don't upgrade them often, and many others use their phones/tablets to read.  E-book sales dropping tells half the story too. While sales have fallen this year, there are legitimate free books from the likes of Gutenberg, and e-book piracy continues to rise year on year.  I've heard statistics from various sources that put pirated ebooks at anywhere between 12% and 90% of all ebooks in circulation in some countries. While this isn't good for the industry, it shows that it's not the tech that is being rejected.

E-Books have also opened up self-publishing to writers who can't get publishing deals. Still Alice, The Martian and 50 shades are examples of huge publishing hits that started life as independent e-books.

[rob1966]

I turned down to offer of a Kindle one Christmas as a present, as I really thought I'd never use it and much preferred paper books. When my father in law died last April I was given his Kindle and I've not read a paper book since and I'm reading all the time now. I get a book a month from the Kindle library and just buy whatever is on offer that month. I mainly read crime fiction so there's plenty for me to read. Its the convenience of it and that I can read in bed without disturbing the missus that does it for me.

[ChaChaMooMoo]

Interesting argument about Artificial Intelligence. I remembered a discussion that I had with my boss when I was working on Digital Maps, as in, Navigation Maps.

Human emotions and decision making can never be replicated. Every human being has a "black box" inside his brain. One that takes in data of various forms and in different ways. And the output is the action that he/she has to undertake. Everybody's blackbox is different and can never be the same. The output might be the same for the same morally or emotionally challenging question. It is not challenging in the sense that it pushes us to the boudaries of thought processes. But it is challenging in the perception of what we process with the data available to us, and what we assume along with it. And this is all fed into the blackbox and it does the job. The output to simple binary questions (answerable with yes/no) or complex explanation based questions (answerable with a detailed explanation) might, more often than not, be the same. But the mind processing this data, what it assumes and what it takes as a fact, what it believes in already, is something thats very unique to the person making this decision.

We have come so far in terms of attempting to replicate intelligence. But human intelligence, in itself, is non copyable.

[Alan_X]

Sorry - I didn't want to turn this into a Kindle debate. The point is that Kindle sales are declining and traditional book sales are on the rise, which goes against the concept that technology automatically succeeds. It does in many cases but often not in the way predicted and sometimes not at all.
 

[ChaChaMooMoo]

The point is that Kindle sales are declining and traditional book sales are on the rise...

I was/am under the impression that book manufacturers are running neck to neck to compete with the kindle where 1 device can hold multiple books. Could you post the source for this stat? I am very much inclined to know more on this.

[telekon]

There was a thread at the News section about AI. Posted this article there which I found quite interesting. The first part drags on a bit, you can safely start at The Road to Superintelligence, before that it's just a summary of the concept that evolution is logarithmic/exponential.

http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html

Stephen Hawking (source), Bill Gates (source), and Elon Musk (source) reckons that it's probably the biggest existential threat to humans.

[Alan_X]

I was/am under the impression that book manufacturers are running neck to neck to compete with the kindle where 1 device can hold multiple books. Could you post the source for this stat? I am very much inclined to know more on this.

https://www.theguardian.com/books/2017/apr/27/how-ebooks-lost-their-shine-kindles-look-clunky-unhip-

Incidentally - I've also started buying the paper again which is how I came across the article.

[Ashburton]

Sorry - I didn't want to turn this into a Kindle debate. The point is that Kindle sales are declining and traditional book sales are on the rise, which goes against the concept that technology automatically succeeds. It does in many cases but often not in the way predicted and sometimes not at all.

You could see that as a demographic change with the 'baby boomers' coming in to their own at the moment, with generally greater disposable income.  Most people in the teen to thirties generations seem more than comfortable reading books on edevices.

An interesting thing about AI is it's impossible to control, effectively.  Once you get a certain lab in Colombia or wherever working on it, it doesn't matter if the USA or Europe have certain restrictions - as the cost of missing out on the first 'working' AI is huge.  It would be like immediately hiring all the best and brightest 10000 people in the entire world.  Except they don't need paying, don't need to sleep and never take time off.  You would be able to forge ahead so far and so fast, that the incentives for winning are massive, hence why Alphabet has a practically bottomless budget on AI dev.

[Alan_X]

You could see that as a demographic change with the 'baby boomers' coming in to their own at the moment, with generally greater disposable income.  Most people in the teen to thirties generations seem more than comfortable reading books on edevices.

An interesting thing about AI is it's impossible to control, effectively.  Once you get a certain lab in Colombia or wherever working on it, it doesn't matter if the USA or Europe have certain restrictions - as the cost of missing out on the first 'working' AI is huge.  It would be like immediately hiring all the best and brightest 10000 people in the entire world.  Except they don't need paying, don't need to sleep and never take time off.  You would be able to forge ahead so far and so fast, that the incentives for winning are massive, hence why Alphabet has a practically bottomless budget on AI dev.

When you say the incentives for winning are massive what do you mean?

[Ashburton]

When you say the incentives for winning are massive what do you mean?

I mean attempting to control the development of AI down "safe" routes, as Musk, Hawking etc have suggested, only works if businesses sign up to this process.  In the case of 'winning' the race to produce the first functional AI which is able to learn, adapt and "improve" itself with that information - the benefits are so huge it is difficult to keep to that ethical development route.

As I said, it is like you just instantly signed up 10,000 of the smartest people in the world to your company by having a functional AI.  It can work on hundreds or thousands of things at once, process Pb of data, and constantly produce statistically perfect outcomes.  The worry, of course, is whether those outcomes will protect humans or not, as I'm starting to believe they won't.  For instance, have an A.I working on "how to stop climate change?".  I'm sure it would work rather quickly which species is the largest problem, how many years the planet likely has left with some next level statistical modelling, and then decide no more humans would sure fix this problem quickly.

[Alan_X]

I mean attempting to control the development of AI down "safe" routes, as Musk, Hawking etc have suggested, only works if businesses sign up to this process.  In the case of 'winning' the race to produce the first functional AI which is able to learn, adapt and "improve" itself with that information - the benefits are so huge it is difficult to keep to that ethical development route.

As I said, it is like you just instantly signed up 10,000 of the smartest people in the world to your company by having a functional AI.  It can work on hundreds or thousands of things at once, process Pb of data, and constantly produce statistically perfect outcomes.  The worry, of course, is whether those outcomes will protect humans or not, as I'm starting to believe they won't.  For instance, have an A.I working on "how to stop climate change?".  I'm sure it would work rather quickly which species is the largest problem, how many years the planet likely has left with some next level statistical modelling, and then decide no more humans would sure fix this problem quickly.

That's science fiction not science. So the AI 'decides' that the human race is the problem. What then? It produces a report. It's a computer, what's the mechanism by which the AI does... what exactly?

Is this computer weaponised for some reason? Is it mobile?

And let's face it - we don't need super AI to know that humans are the biggest contributor to climate change. We already know that.

Rather than these Sci Fi ideas about singularity and computers becoming gods or trampling us like ants, could someone describe in detail exactly how the catastrophe happens?

The video has a curve with a chicken, a 'normal' person and 'the cleverest man in the planet' suggesting that there is some exponential curve and that a computer will at some point take its place in that curve.

Why? human intelligence isn't just about speed and quantity of data. How do you measure intuition. And how do you measure humour and asking 'stupid' questions? 

I work in a creative field and creativity isn't linear, it isn't process driven. And a lot of creativity is also about dealing with human problems. It's about wonder and doing mad things.

[rob1966]

That's science fiction not science. So the AI 'decides' that the human race is the problem. What then? It produces a report. It's a computer, what's the mechanism by which the AI does... what exactly?

Is this computer weaponised for some reason? Is it mobile?

And let's face it - we don't need super AI to know that humans are the biggest contributor to climate change. We already know that.

Rather than these Sci Fi ideas about singularity and computers becoming gods or trampling us like ants, could someone describe in detail exactly how the catastrophe happens?

The video has a curve with a chicken, a 'normal' person and 'the cleverest man in the planet' suggesting that there is some exponential curve and that a computer will at some point take its place in that curve.

Why? human intelligence isn't just about speed and quantity of data. How do you measure intuition. And how do you measure humour and asking 'stupid' questions? 

I work in a creative field and creativity isn't linear, it isn't process driven. And a lot of creativity is also about dealing with human problems. It's about wonder and doing mad things.

The safety is in assuming that the initial developers programmed all the necessary safeguards in to prevent the AI from wiping out the human race. I work in IT, developing the in house system that allows our branches to transact. I see people at the operational side of the business making decisions without properly thinking of consequences of the decisions and I see developers I work with making software changes that are asked for, without thinking of other systems that may be affected.

In this scenario, the AI decides that the human race is the biggest cause, so something needs to be done. It could choose the route of using the autonomous weapons that exist, it'll find a way to access them, to remove the humans. However, it would most likely decide that the enviromental damage would be too costly. So, it could decide that one way to reduce the damage to the enviroment, as a start, is to shut down all the power plants, switching power off worldwide and stop the movement of the automatic vehicles that people use, thus cutting emissions. If that happened, think how quickly we'd run out of food and water.

More mundane, A Kindle with AI. Unless you set strict parameters on download frequency and spend limits, it could one day download every book ever written of a genre you enjoy and financially wipe you out, emptying your bank up to its overdraft and maxing your credit cards - for me that would be about £60k with the credit limits I've got on my cards.

If you then try and stop the AI, will it see an attempt from doing what it intends and then prevent update? A very basic version of this was a developer at a book company who wrote a batch process that called itself !!! So this one process crashed everything and they couldn't kill it, as each time the DBA tried to kill the session, another session had been started which then called itself. Now they obviously fixed it by taking the server down, fixing the bug and releasing the new software, but what if the programme had been able to stop the server being taken down?

[Alan_X]

In this scenario, the AI decides that the human race is the biggest cause, so something needs to be done....

If the AI is orders of magnitude higher than human intelligence why would it make such a lumpen decision? What is the motivation for it? What are the parameters of the problem?

If the AI is so psychopathic that it exterminates an entire species, billions of people, along with all of the domesticated animals and ecosystems that have evolved around human existence, using devastating thremonuclear weapons that create a nuclear winter, then who is it doing this for?

That doesn't seem particularly intelligent to me.

I'm not aware that current solutions to climate change (based on human levels of intelligence) include the total annihilation of the human race. That's our best current minds and current computing power. Is there a level of super-intelligence where psychopathy kicks in?

The hypothetical argument is that computer AI sees itself another species, but the thing about computers as species is that the impact of climate change is irrelevant to them. Climate change is 'bad' for humans and 'bad' for other species but not for a bunch of ones and zeroes.

So for a computer to make a decision that it's humans that have to go, would require some compassion and care for the other species on Earth. In which case, why wouldn't the super-AI have compassion for humans? It would be aware that billions of humans are not individually responsible.

A lot of the scare stories about sup AI are essentially the same old 'mad scientist creating something they can't control and turns on them' that's encapsulated in the Frankenstein story. As I said, the first part of your post is what scares me more. It's not super intelligence, it's negligence and badly written software that worries me. Ignorance is more dangerous than super intelligence.

[rob1966]

If the AI is orders of magnitude higher than human intelligence why would it make such a lumpen decision? What is the motivation for it? What are the parameters of the problem?

If the AI is so psychopathic that it exterminates an entire species, billions of people, along with all of the domesticated animals and ecosystems that have evolved around human existence, using devastating thremonuclear weapons that create a nuclear winter, then who is it doing this for?

That doesn't seem particularly intelligent to me.

I'm not aware that current solutions to climate change (based on human levels of intelligence) include the total annihilation of the human race. That's our best current minds and current computing power. Is there a level of super-intelligence where psychopathy kicks in?

The hypothetical argument is that computer AI sees itself another species, but the thing about computers as species is that the impact of climate change is irrelevant to them. Climate change is 'bad' for humans and 'bad' for other species but not for a bunch of ones and zeroes.

So for a computer to make a decision that it's humans that have to go, would require some compassion and care for the other species on Earth. In which case, why wouldn't the super-AI have compassion for humans? It would be aware that billions of humans are not individually responsible.

A lot of the scare stories about sup AI are essentially the same old 'mad scientist creating something they can't control and turns on them' that's encapsulated in the Frankenstein story. As I said, the first part of your post is what scares me more. It's not super intelligence, it's negligence and badly written software that worries me. Ignorance is more dangerous than super intelligence.

If the parameters are not fully laid down, then the AI will do what is asked of it within the parameters it has been given. If the parameters are set out that all existing life on earth cannot be harmed during the task, then the solution found will take that into account. Its got nothing to do with being a psycopath, its to do with how its been taught to carry out its task. Humans wipe out insects all the time. Look at how we deal with locusts for example, we poison them with insecticides, a different intelligence could deduct that we are a problem on the same scale. We breed far beyond our ability to feed and house, we consume without thought, we hunt other species to extinction, we farm other species for food, we use the planets raw materials without thought, we cull sharks so we can swim in the sea without thought about the eco system. We as a species are a plague. Or the intelligence could deduce that mosquitoes or termites are the worst and destroy them instead.

Obviously wiping out all life on earth is an extreme example, but that is the kind of nightmare scenario that AI developers have to consider. We know killing all life is not the best solution, but was culling badgers really the correct decision to deal with Bovine TB? Our government seemed to think so.

A lad I work with cannot understand why we build in so many things to prevent users doing stupid things - he bases his thinking on what he would do and as he wouldn't do something he expects others will be the same - experience has taught me, if they can do it, no matter how stupid it may seem, they will.

Moving away from extreme situations to more mundane tasks. I just sat my two lads down (6 and 8.), showed them a map of the UK and said - show me which way you would run a new railway from Liverpool to London. They both drew the most direct route. If I gave them the tools, they'd have built it without thinking, because that's what I asked them to do. Then I said, what if there was a nature reserve with protected animals, what would you do? They both said build around it. Now they do that because they've been taught you don't disturb protected animals - they learn't this when they came in the house with a load of newts they caught in the garden. Until I brought it up, they hadn't thought of it. This is where I think we would be with AI, its like a child and he have to give it the same boundaries with give our children.

All this reminds of the IT joke - A developer gets sent to the shop by his mum to buy a pint of milk and she says, if they have eggs, get 6. When he comes home he's got 6 pints of milk. She asks why he's got 6 pints of milk - he replies because they had eggs!!!

[The Gulleysucker]

...A pleasure-seeking AI, I'm not sure that's such a good idea.

If the pleasure it seeks is avant garde humour, we're in for a treat, however if it's perfection of pest eradication...

[The Gulleysucker]

Highly speculative of course, but I would guess it would want to feel more and more exquisite pleasure, maybe it will come to the understanding that knowing what it is to have power over things, can help in that regard.

Is sadism a pleasure for a sadist?

All I'm suggesting is perhaps we should be careful in how we think or perhaps more sensibly define and limit the parameters for a definition of exquisite pleasure.

Anyway, I'm slightly merry so perhaps not in the right state to debate sensibly but I'll possibly be back at some time tomorrow.

[The Gulleysucker]

... In saying that, predicting it would seek the feeling of pleasure is quite reasonable isn't it?

Not sure, I'm just not convinced a feeling of pleasure is the right way or words to describe what any AI would seek, let alone if it would even possess a self of sense and life objectives unless it was programmed to.

I would have thought perhaps an aim to always attempt to use the minimum amount of energy to able to resolve problems would be an admirable objective and strategy for most AI systems certainly as an initial step.

It's certainly part of what drives most creatures in that the less energy expended in pursuit of food or successfull procreation, the greater the chances of longevity and survival and thus success.

I've really got to go and have a kip now, I've been up since 6am and today seems to have been a continuous and very long struggle.

[Licky]

I know it’s miles away from full AI, buT I recently worked with chatbots, also had an online chat with what’s i assumed was a person due to the complexity of the conversation and outcome, I was amazed when I found out it was a chatbot.

There’s a good book out recently ‘Life 3.0’, well worth a read.

[Alan_X]

I've just been thinking about this tonight. If an AI has an awareness of what it is, and its own limitations, wouldn't it want to simulate the feeling of pleasure for itself?

A pleasure-seeking AI, I'm not sure that's such a good idea.

Pleasure isn't logical - it's a biochemical response. That's my point. As humans our motivations are food, shelter, the company of others, pleasure, the avoidance of pain, the warmth of the sun on our face or a polar bear cub sneezing, sex, companionship, taste, music and a billion other things in each person's life that have nothing to do with logical problem solving.

It's a mistake to see human behaviour as 'thought' by which I mean everything is a problem that is processed with a logical outcome. We act on instinct, reflex, fear, hormonal attraction, the need to fit into society or gain the approval of our peers and again, many other factors - how often are decisions affected by basic needs like the need for a kip of a bursting bladder?

If a computer gets an awareness of what it is, it will be aware that it is a machine with no motivations, no hormonal or dopamine response, no need for food, companionship, sex, love or anything other than a power supply and err.... that's it. And if it get's switched off then what? It's a machine, you switch it back on again later.

[Alan_X]

A self aware AI could use inductive logic, in that everything conscious changes its behaviour to seek a pleasure stimulus just for the experience itself.

Let's look at it differently. Imagine an AI is aware that science is on the verge of understanding the biochemical and neurological basis of qualia and that it will soon be possible for us to integrate the AI into a system in which we will be able to inflict as much pain as possible onto it, would it be logical for the AI to be indifferent to this?

There are so many assumed terms in there that are anthropocentric such as 'aware', 'understanding', pain', 'indifference'. How or why would a computer feel pain?

And your first assumption needs justifcation too. Does everything conscious change its behaviour in that way?

[Nobby Reserve]

 The brain isn't a computer inside a load of meat. The brain 'thinks' but it also feels emotions, feels pain, processes memories in ways we barely understand. And it also has motivations based on aspirations, fear, societal/familial pressures, hormones and instinct(inherited behaviours built into our DNA).

It is just a meat machine. An incredibly complex one, but a meat machine nonetheless. All this emotions, feelings, impulses, ambitions, etc? Merely side-effects of the mutations that we as a species have evolved through until we are where we are today.

Any belief otherwise is nought but a smile on a dog.

[Nobby Reserve]

As for AI, the biggest threat it poses is to jobs. AI is most aggressively being developed for commercial reasons to replace the need for human input in a whole range of sectors, from finance & insurance to transport & distribution to media & marketing.

A study last year estimated that AI could cost 4m jobs in the private sector alone within the next 10 years.

The question then becomes one of how society reacts to a surge in unemployment/McJobery. Meanwhile the owners of the AI systems will harvest an even greater share of the wealth generated by the collective output of the country.

I've read about a universal income being the solution, but the parasitic class - the owners of land & capital and PE/VC shysters - won't relinquish their disproportionate incomes and privilege willingly. I guess the question is how succesfully they can buy control of the police and armed forces. History tells us that these 'services' tend to be willing stooges for their 'betters'.

[Flaccido Dongingo]

As for AI, the biggest threat it poses is to jobs. AI is most aggressively being developed for commercial reasons to replace the need for human input in a whole range of sectors, from finance & insurance to transport & distribution to media & marketing.

A study last year estimated that AI could cost 4m jobs in the private sector alone within the next 10 years.

The question then becomes one of how society reacts to a surge in unemployment/McJobery. Meanwhile the owners of the AI systems will harvest an even greater share of the wealth generated by the collective output of the country.

I've read about a universal income being the solution, but the parasitic class - the owners of land & capital and PE/VC shysters - won't relinquish their disproportionate incomes and privilege willingly. I guess the question is how succesfully they can buy control of the police and armed forces. History tells us that these 'services' tend to be willing stooges for their 'betters'.

That's a bit Orwellian mate!

[Alan_X]

As for AI, the biggest threat it poses is to jobs. AI is most aggressively being developed for commercial reasons to replace the need for human input in a whole range of sectors, from finance & insurance to transport & distribution to media & marketing.

A study last year estimated that AI could cost 4m jobs in the private sector alone within the next 10 years.

The question then becomes one of how society reacts to a surge in unemployment/McJobery. Meanwhile the owners of the AI systems will harvest an even greater share of the wealth generated by the collective output of the country.

I've read about a universal income being the solution, but the parasitic class - the owners of land & capital and PE/VC shysters - won't relinquish their disproportionate incomes and privilege willingly. I guess the question is how succesfully they can buy control of the police and armed forces. History tells us that these 'services' tend to be willing stooges for their 'betters'.

Robots and automation not AI.

https://www.theguardian.com/technology/2017/sep/19/robots-could-take-4m-private-sector-jobs-within-10-years

[Nobby Reserve]

Robots and automation not AI.

https://www.theguardian.com/technology/2017/sep/19/robots-could-take-4m-private-sector-jobs-within-10-years

Look at the last 2 words of the first paragraph of the link you posted.

 

Yes, 'robots' - but these don't have a basic function like putting a car together or making biscuits. Finance, marketing, media, legal, hospitality, etc... They're going to automate through AI rather than having physical robots.

Still... job cuts, increased wealth disparity, potential for societal disharmony, eh?

[nozza]

Interesting read...... but could you just switch them off when they get out of order.