Beruflich Dokumente
Kultur Dokumente
A PRIMER
BY
ABHIDEEP BHATTACHARJEE
Copyright Declaration
This primer is the intellectual property of Abhideep Bhattacharjee May 2015. Printing, publishing or
duplication of this material by any means will without the authors consent will be deemed a violation of the
INTELLECTUAL PROPERTY RIGHTS
INTRODUCTION
I wrote an answer on machine intelligence some time back in Quora. A user had asked
what will humans do if machines ultimately get smart enough to do most (read all) of
our tasks. A seemingly harmless question, borne out of a sense of leisurely curiosity
(type of curious questions that pop-up in your mind when you have nothing better to do)
I have come across similar version of questions in other places (what will coders do if
we develop softwares which can create new softwares?) Questions like these never
bothered me much. I always thought the reason these questions were being asked was
because some people were scared of a job shortage that may appear in the near future as
a result of most things going autonomous. This I figured was the same kind of security
issue people faced before the industrial revolution. So I tried to console the user by
arguing that machines can never be intelligent beyond a point. I went on to describe that
a machine, no matter how intelligent, will always require a human intervention to
operate and that a super-smart, self -thinking, machine or program or robot is still a
subject for the movie makers. I wrote that answer in avid detail using terms like
comparative logic and creative logic to make my answer seem more credible than it
actually is. Then I waited for my upvotes.
But what followed was a comment. A Machine Intelligence Programmer vehemently
contradicted my stupidity for thinking that machines (read programs) are dumb and are
incapable of thinking on their own. (A Machine Intelligence Programmer, is someone
who teaches a machine or a software to think on its own) He went on to explain that
Machine Intelligence, more commonly referred to as Artificial Intelligence (AI), has
come a long way and will soon take up all of human activities. Really? I thought! Are
there programs which can actually think and decide? Are computers learning to think by
their own? Will they be able to do things without human intervention? I started up a
conversation with some people who, even if not experts in the field of Artificial
Intelligence and Machine learning, at least have a sound knowledge about it. What I
learnt was enough to surge my interest in this field and over the next few weeks, I
started reading articles, books and journals that detailed the development and
challenges in this field. I found myself equally enthralled by the promise of AI
development and tenuously concerned at the associated risks it will bring. What
surprised me the most however was how I knew next to nothing about a technological
development, which will have astronomical impact on humankind at large, in both
positive and negative way - all that, well within my lifetime. It was then that I decided
that I must write a primer detailing the most important points and developments that
have been made in this field and what the near future holds for all of us.
The purpose of writing this primer is not to provide yardstick details about the progress
and road map to AI. For that, one needs to read a lot more literature. The purpose of this
little primer is to arouse the sense of curiosity amongst all those who read this, so that
they understand that we are all standing on the edge of a technological marvel - one that
will push us towards either of two destinations - Immortality or Extinction!
We are all playing on a sunny beach with our friends and family. The water is cold and
the sand is warm. The waves are rolling one after the other and the sea looks calm and
serene. White patches of cloud are floating about in the blue sky. The kids are building
sand castles, some are running around with a frisbee. You look around at all the happy
people. The world looks wonderful. At the horizon, the sky bends down to kiss the sea,
and from this far, it is difficult to say where the sea ends and the sky begins.
But all that is going to change in about 10 minutes from now. A raging Tsunami is
surging towards the coast like a hungry monster. 7 minutes from now, you will see a tiny
below at the far horizon. You will try to figure out what is the strange shape. 30 seconds
later, the below would have grown a tiny bit bigger and a few more people would have
observed it by now. 20 seconds later, you will understand it's a giant wave rolling
towards the beach. Another couple of seconds later, someone would have cried out Tsunami, Tsunami! and immediately, panic would have spread across the entire
beach.
For the next one minute, you would observe with horror as all the happy people would
run like crazy, picking up their kids, screaming, hollering, scampering away from the
shore. A few people would fall down in the chaos and would be stamped around by the
runners. Meanwhile, the giant wave has grown upto a height of 2 feet and surging at a
speed of approximately 80 KMPH. In the next 10 seconds, the water bellows in on the
shore, wiping down everything in its path. In the next 30 seconds, it gurgles down into
the city, drowning houses, cars, animals and humans. In the next one minute, all the
happy people, laughing and smiling and enjoying life, are dead.
Some sceptics believe that the AI is the hypothetical Tsunami. And we are the happy
people enjoying in the sunny beach, 10 minutes from disaster.
Is that a correct prediction? Should we be scared? Read on to find out!
Human Intelligence
What amount of intelligence is a child born with? Many scientists believe that a child
starts to learn right from inside the womb of its mother. Pregnancy experts and
paediatricians suggest would be parents to bask in wholesome activities, which will be
conducive to a childs cognitive development. A mother listens to soft music, reads
book and talks to her baby. Is the baby listening? Is she learning? Many believe that she
is.
A human child is born with a certain level of intelligence already instilled in him or her.
It is believed that this basic amount of intelligence is hardcoded in the genes. But what
is more important is the fact that the child, right from the moment it is born, is highly
curious about the external world. It readily receives information about the world and its
environment through all possible neural inputs - auditory, optical, olfactory, touch and
taste. Not only does she receive inputs, she goes on making associations based on these
inputs. By her second birthday, a child would have learnt much more than she will ever
learn in the rest of her life.
That brings us to a critical question - If all children are born with the basic level of
intelligence that is coded in their genes, and if the rate of information assimilation
for the first two formative years is also about same, how come does the intelligence
level of adults vary drastically? (from the - Oh so smart! to the Eh so dumb!) To
seek the answer to this question would be a journey that will take us to the caverns of
the human mind and its least explored corners. Interesting as it may be, we will put off
that subject for a future discussion. Presently it will suffice to appreciate that fact that
humans, and specifically all forms of higher life forms, have the capability to change the
level of their intelligence. However, though we may have a dog who is much smarter
than another dog, none of them will ever be quite as smart as a human. This means
intelligence comes with a capping - there is a limit to how intelligent a living thing can
be. And in the history of the earth, we have so far only one animal who has surpassed
the intelligence level of all other animals by such a huge difference, that no other animal
has ever even posed a minor threat to its existence or dominance on the Earth. But that
may soon change forever.
Humans have always been Natures favourite. Nature has somehow played some sort of
favouritism that has helped us (humans) to cheat on all the other animals into owning the
earth. But as it always happens with the spoilt kid, we have never been satisfied.
Intoxicated with the gifts with which Nature has endowed us, we have always tried to
challenge her back. In various attempts of egoistic ardour, we have tried to recreate
Nature. Thus man has lighted up darkness, has flown into the air like a bird, has
communicated across thousands of mile, has explored the deepest crevices of the ocean.
Man, has always been like the new born baby - ever curious, always trying to figure out
the world around itself, and then trying to outsmart it. And we have succeeded in
outsmarting nature in more ways than we can imagine. We have harvested the natural
resources, we have turned the direction of rivers, we slit-open a human body and then
put it back together - in proper functioning mode. We have manipulated genes, tried to
overcome diseases before they could even occur. We have grown crops in soils and
weathers not conducive to its growth. We have predicted the laws of nature and found
ways to escape it (think of a rocket escaping out of Earths gravity) We have ripped
open an atom and synthetically produced compounds. Even Nature may sometimes look
back upon us and wonder - I really made up a smartass of a creature over there!
But are we to stop? Of course not! It has never been the nature of man to stop searching
or to stop trying to improve. Most of the time, this has resulted in the greatest gifts of
mankind (wireless communication, electricity, medicines) For some unfortunate events,
these have led to catastrophic disasters (think of the atom bomb, the bio-weapons) But
however it may turn out to be, we will not stop our quest for improvement. To replicate
nature, and then to outsmart it.
excited - this aint out in the market yet. What is already in the market is the use of stem
cells to synthetically produce body tissues. If a person endures a burn on a portion of his
or her body, the common practice is to graft the skin tissue from another site of his body
(called the donor site) and transplant it over the damaged site. Another practice has
started to gain popularity is called the Stem Cell Tissue Culture, where the stem cells
of this person is used to artificially produce skin cells and these cells are then
transplanted on the damaged tissue surface. Research is already in the sway to use
SCTR (Stem Cell Tissue Culture) to produce other body tissues like retina tissues,
pancreatic tissues, etc. If successful, this can cure problems like blindness and diabetes.
When so much is happening in the physiological foray of the human body, isnt it
obvious that another group of people will be equally enthused about the mind part? I am
not talking here about psychologists or cognitive scientists who try to figure out the
working of the mind. We are talking about that group of scientist who are intrigued and
motivated to develop an artificial mind - a mind just like that of the humans, with the
same set of cognitive skills, able to perform the same mental functions that a human is
able to and having the same level of intelligence as that of a smart human; all that - but
entirely artificial. Welcome, to the world of Artificial Intelligence!
Over the next couple of decades, computers (read programs) became increasingly
smarter. There were programs which could play chess, paint a picture, project models
based on statistics and even have conversations with humans and respond to them.
Clearly, scientist needed an uniform and unambiguous test, that will be used to verify
whether or not an AI has reached human level intelligence. One of the most popular
tests that came up as the Holy Grail for AI validation is the Turing Test, proposed by
the famous scientist Alan Turing, in his paper Computing Machinery and Intelligence
in the year 1950. The outline of the test describes that a human will be used as a test,
who will be communicating via chat (writing medium only) with two participants in two
rooms. The catch is that one of the participants will be an AI. The man having this
communication will have no way of knowing which is which. If the man fails to identify
the computer as a computer, then the AI would have passed the test.
Interesting as it may sound, what is even more interesting is that programs have already
passed the Turing Test. A program called ELIZA, designed by programmer Joseph
Weizenbaum in the year 1966 is one of the first programs to have successfully passed
the Turing Test. Another program called PARRY designed by Kenneth Colby in the year
1972 also passed the Turing Test, with participants able to tell that it is a machine and
not a man, only 48% of the time, a figure which is consistent with random guessing.
If this piece of information awes you, prepare to be surprised even more. Chances are,
you too have already been fooled by a program into a conversation, believing that you
are chatting with an actual person. There are lots of chatterbots doing the rounds in the
internet, which presents itself as a person seeking relationship, and chats with another
person online to lure them to divulge their personal data.
us, with superhuman capabilities and with a vengeful attitude. Well, we are somewhat
correct, thanks Hollywood. What we miss, is this - AI is a broad field and it is
separated in three levels called calibers, each level leading to the subsequent one and
each level subsequent level exponentially more complex than its preceding level. The 3
calibers of AI are as Artificial Narrow Intelligence or ANI: ANI or Artificial Narrow Intelligence is
the capability of a computer or program to showcase intelligent behaviour in one
functional area only. Consider it to be like a prodigy child - a child who
surpasses normal human intelligence level for a particular display of talent (ex painting, mathematics, etc) but is usually not so smart in the other cognitive
areas (like chemistry, literature, social skills, etc) The difference between a
human prodigy and a computer ANI is that the later doesnt has even rudimentary
intelligence in fields which do not fall in its domain. We have long since
achieved exceptional feats in ANI and have kind of exhausted its applicable
areas. We are now moving slowly but steadily to the second caliber of AI, which
is
Artificial General Intelligence: This is the dauntingly difficult part to achieve
but many scientist believe we are on the threshold of creating an AGI. AGI or
Artificial General Intelligence is a when a machine has an Intelligence level at
par with a human. The reason this is called general intelligence is because such a
program will be aware of all facets and will be intelligent in more than one field.
AGI doesnt sound dangerous, no, not unless you realize that a computer
operating at an AGI level will have all the craftiness of a human mind, viz power of reasoning, abstract thinking, analysis and problem solving. Also
remember that an AGI will work at a much faster DPR (data processing rate)
than a normal human being. It can continuously improve its own source code to
make itself better and better, and this is where it gets scarier, for when an AGI
becomes more intelligent and crosses the threshold of the most intelligent human,
it enters into the realm of
Artificial Super Intelligence (ASI): ASI is defined as that level of intelligence
which surpasses human intelligence level in all fields, ranging from creativity,
problem solving skills and social skills. It can range from intelligence level just
slightly higher than that of the most intelligent human to zillions of time more
intelligent. That would make ASI brain compared to our brain as ours is
compared to the brain of ants. What will happen when ASI is achieved? How
soon are we going to achieve it? We will take up these questions in our narrative.
Retargeting: This is when you browse a product in an ecommerce site and the
product suddenly starts showing up across ad spaces in different sites that we
visit.
Friend suggestions: Facebook keeps suggesting people who may be our friends.
LinkedIn suggests people we may know and may wish to connect to. Strangely
enough, most of the suggestions turn out to be people we actually know.
Virtual games: Whether it is the simple chess or tic-tac-toe game that you play
or the more sophisticated games in the playstation, these programs are designed
to think along with you and strategize its next move accordingly.
Smart Phone: Most interestingly, the smartphone in your hand with the array of
different applications, some to help you wake up, some to count the calories you
burn, some to catch up with your friends, some to help you navigate the road, is a
factory of ANI.
All these programs as discussed in the example above and countless others just like
them, are classic examples of ANI - an intelligent piece of program targeted at one
specific task only. For instance, ask the navigator app how to travel from point A to B,
and it will find out all possible paths you can take to reach your destination. Ask the
same app however to check the spellings in your essay and it will not even comprehend
the meaning of your request.
computational power we can avail for $1000 (a ballpark figure for something to
have marketable value) Currently, the computers which are available for $1000
are operating at one-thousandth of the human intelligence level. This doesnt
seem like any feat, not unless you consider that computational power available
for $1000 was operating at a trillionth of the human intelligence level in the year
1985 and escalated to only a billionth of the human intelligence level by 1995
and a millionth by 2005 and is currently at only a thousandth. If it follows the
same rate of growth, by the year 2025 we will be able to produce enough
computational power in $1000 to compete with the human intelligence.
2. Making the computer smarter: We have seen that the hardware necessary
for making AGI is already available. Now comes the difficult part - to make
the computer smarter, at par with a normal human intelligence level. There
are different ideas prevalent on how to make the computer smart at the
human level. Various scientist are trying out various techniques and all that
we know is that one day, one of these will just work out. Of the various
types of mainstream principles to make the computer smart, some of the
most interesting are these
1. Copy the brain: Thats like a duh! Of course if you are trying to
emulate a natural phenomenon in the mechanical world, you need to
replicate it and its action. But the problem with emulating nature is
that more often than not, the models that work for nature do not
work in the mechanical realm. For instance, the flight of birds might
have inspired men to fly, but inventing a device that could flap wings
wouldnt have worked all that well (though this is exactly how they
started off with the plane, but anyways) When it comes to creating an
artificial model that could emulate a natural phenomenon, we need to
think from the models mechanical perspective, not from the
perspective of the natural phenomenon. Thus rather than copying the
brain verbatim, scientists and machine-programmers are using
computational circuitry to create virtual neural paths. When the
program is asked to execute a task or answer a question, a current
gets fired in one direction through a mesh of transistors (the
computers equivalent for neuron) If we confirm the action to be valid
or the answer to be correct, the pathways through which the response
was fired gets strengthened. If however we mark the action as invalid
3. Self help is the best help: Though the most daunting, yet this is also
the most promising method to reach AGI levels. The idea is to write a
program with two main objectives - i) Do research on AI, and ii) keep
INTELLIGENCE EXPLOSION!!
Once the self-reprogrammable AI keeps on improving itself recursively, it will surpass
all intelligence levels known to man and will reach a level of intelligence we can
neither imagine nor comprehend. This is called an Intelligence Explosion. And when
this happens, we have no idea what will happen. Will the AI, with its Godlike
intelligence, be faithful to its creator? Or will it see mankind as an unnecessary tantrum
and decide to do away with our race? Will it help solve problems like energy crisis,
poverty, global warming, etc, that seem so daunting to us? Or will it rather create
problems at so great a magnitude that we face the most dreaded outcome of all Existential Risk!
are okay with it. Heck we are even okay with its marginal side effects at replacing some
of our jobs. But something that poses an Existential Threat? Nah, not interested!
But it doesn't actually matter whether you or I are interested or not. History is evident
that a scientific progress of any scale can never be subdued, whether by individual or
political resistance. It is not about the doggedness of humans, but rather the way nature
behaves. Even if governments regulate or put a ban to AI research, there will still be
men and women around the world who will find motivation and means to pursue it,
some, for the sake of science, some to play God. Whatever the case maybe, if ASI is at
all possible, we know that it will be created, there is no other way to it.
Since we can never actually restrict the advent of an intelligence of caliber 3, what we
can do however is ask a few fundamental questions to understand what to expect when
our guest is finally here. One of the first primary question we need to ask is
The former group however has a point in case that puts the favor of the argument on
their side. They bring to attention the year of 1985 when internet was slowly catching up
and people believed that though it may one day go on to be very impactful, such a thing
will not happen in the near future. We all know how incorrect that assumption was when
internet was a global phenomenon within the next decade.
The later group, one which is led by Paul Allen and Gary Marcus, reply with the logic
that as the intelligence level grows, so will grow the challenges associated with
achieving the next level. Typical exponential growth rate will not apply here.
As of now, we are only speculating. We have no way of knowing when exactly will
apocalypse start. Perhaps the Mayans could have told, but they are long past. A third
group of thinkers, quite like agnostics, believe that both the groups can be right. They
believe that given the unpredictable way in which intelligence explosion behaves, an
ASI level of intelligence can as well be achieved in the next couple of decades or it may
even take several more years.
A fourth group believe that there isnt a timeline prediction for achieving ASI as we can
actually never reach that level.
To put things in perspective, in the year 2013, Vincent C. Mller and Nick Bostrom
conducted a survey that asked hundreds of AI experts the following questions - For the
purposes of this question, assume that human scientific activity continues without major
negative disruption. By what year would you see a (10% / 50% / 90%) probability for
such HLMI (High Level Machine Language) to exist? In brief, the question asked the
scientists, that presuming progress is made at the current rate, by which year do they
think that there is a 10% chance of achieving an AGI, by which year there is a 50%
chance of achieving an AGI and by which years there is a 90% chance of achieving an
AGI. The data, when consolidated, looked like this 10% likelihood of achieving AGI by the year: 2022
50% likelihood of achieving AGI by the year: 2040
90% likelihood of achieving AGI by the year: 2075
This means that a group of experts of AI think that Human Level of Machine
Intelligence or AGI can be achieved in another seven years. Even if we take worst case
scenario, still most scientist believe AGI will be achieved well before the end of this
century.
But mind you it is not AGI which will have life-altering implications, the toast of
Immortality or Extinction will be brought along with ASI and not AGI (though the later
will be a precursor of the former) So what should concern us even more is when will
ASI be a reality? The same survey which was conducted by Mller and Bostrom also
asked the participants how likely they thought that ASI could be achieved once AGI has
been achieved. 10% of the participating scientists thought that the transition from AGI to
ASI would be achieved within 10 years of achieving AGI levels. 75% participants
thought the transition would happen within 30 years.
Let us assume that the median year for us to get to ASI once AGI has been achieved is 20
years ((10 + 30)/2) As we have already seen in the previous result, 50% scientists
agreed that AGI could be achieved by 2040. Adding the median of 20 years to this, we
can presume that ASI will be a reality by the year 2060. Thats sooner than the maturity
dates of some of your insurances!
that we can spare. Or maybe we will not need cars at all - ASI will devise ways
to teleport us from one place to another. And though such things, in the present
context of discussion seem crazy ideas straight out of a science fiction, the ASI
will accomplish such feats as you would pick a pencil from the floor.
Colonizing the universe: There is one thing we can predict quite confidently
about the ASI - it will soon realize that the earth has limited resource and the sun
isn't getting any younger. It will realize what we already have - that we are
heading towards the end! Even though we have realized this years ago, there is
not much we have been able to achieve in terms of averting this fate, apart from
speculating and postulating some theories. But the ASI will not have the limitation
of intelligence that we have. No sooner it realizes that the earth will exhaust its
resources soon and the sun will soon be a white dwarf (this is what a start
becomes when it dies!) it will start preparing for ways to expand out of the earth
and solar system and start colonizing other places in the universe. Judging by the
Information Explosion that it will have, we can safely presume that space-time
travel will not be a deterrent for it. Once it starts colonizing the universe, may be
it will allow us to travel to those parts and settle; thereby nullifying the fear of
perishing with the earth once the sun grows too old.
Make humans immortal: Humans have always had a fascination for eternal life.
Its nothing to blame of actually - I dont think dying is fun. Also when you die,
you leave things behind. You can no longer enjoy the company of your friends
and family, enjoy the food you liked, enjoy your vacations, the money you have,
the property you assimilated. For ages, humans have searched for an elixir that
would give them the secret to eternal living - for ages and eons. But is this
possible? Let us look at it this way - when life originated, nature had to find a
way to help it continue. And life could continue in either of two ways - a) eternal
life (or immortality) or b) reproduction. Of these two options, the second one
was by far superior and made a of sense to evolution. So it becomes evident that
nature choose death for us. So now let us ask - what determines our life span?
From an evolutionary point of view, the lifespan of a creature depends on the life
cycle - birth - grow strong and reach reproductive age - reproduce and make
offsprings - fend for and provide for the offsprings till they grow young and
capable enough to take care of themselves. Once this cycle is complete, nature
doesnt see any reason why we should continue living. So we have accepted the
fact that death is inevitable and this is how nature wanted it. But the eminent
physicist Richard Feynman disagreed. He said It is one of the most remarkable things that in all of the biological sciences there is
no clue as to the necessity of death. If you say we want to make perpetual motion, we
have discovered enough laws as we studied physics to see that it is either absolutely
impossible or else the laws are wrong. But there is nothing in biology yet found that
indicates the inevitability of death. This suggests to me that it is not at all inevitable
and that it is only a matter of time before the biologists discover what it is that is
causing us the trouble and that this terrible universal disease or temporariness of the
humans body will be cured.
Let us now look at death this way - why do we die? Not because Mother Nature sends
the reaper with a slip that our time is up! No, we die because we age. And what does
aging do to our body? It makes it weaker and more feeble. This is the same thing with
your car. Inanimate things die too! As you continue using your car, it loses its initial
performance. Over many years, it becomes rusty, takes up more time to start and keeps
surprising you from time to time. Now consider what would happen if you replace the
parts of the car as soon as it goes a little out of the way. Technically speaking, your car
would continue running for ever. The same is with humans, or any animal for that matter.
The reason we die is because our organs wear off, our cells become less effective in
doing what they are supposed to do. Add with that comes diseases and we have all the
reason for the body to stop working one fine day. But with the advent of ASI, this will
be a thing of the past. ASI will help us replace damaged or worn out body parts
immediately with new synthetically produced tissues. In case of any disease, nanobots,
controlled by the ASI will roam about our bloodstream, targeting the damaged cells or
affected cells and removing them and replacing them with fresh new cells. If this
happens, then we would virtually become immortals!
Either that, or the ASI will find another ingenuine way to make us immortal. It will find
a way to download our consciousness into a system and then we will be the system! So
as man made God, so will the creation create the creator! But can our consciousness,
imprinted on an algorithm, be considered as living? Trapped inside a circuit, feeding on
electrical signals, without mobility, without the sensation of touch and feel? Will that
be us? Let me cite an example. A man, in an unfortunate, near fatal accident, damages a
critical part of his brain, thereby rendering his entire body paralysed. The man can
speak and has memory, but he cannot move his body or his limbs. Will he be the same
man as the one before the accident? Biologically speaking - yes, he will be. He might
have broken down emotionally, there may be psychological changes between the preaccident him and the post-accident him. But it will still be him. Same person, same
consciousness. So to our question as to whether we will be the same us if our
consciousness is downloaded to a system rather than confined in a body, the answer is
yes, we will be. Of course it will be weird. But then, change always is. The man who
was suave with his big rule book accounting job must have felt weird when he was
asked to use an Excel sheet. But then he warmed up to it. And now we use it as if this is
the way it has always been. It will probably be the same case with us and the system.
But then, we may not necessarily need to be confirmed in a circuit, our consciousness
stored in some digital memory units. May be the ASI will provide a human-casing for
us, a body of our liking, prepared from cultured tissues, and then download our
consciousness into that artificial body. It will be totally like the body we have now,
with the same level of dexterity and sense organs. We may as well enjoy food and sex
and exercise, only that all of it will be superficial and we could change our body to a
new body just as we change cars now.
things differently again. The point is that, we will be operating from a point of
excessively low and relatively stagnant intelligence level (as compared to the ASI) and
hence we will never be able to understand the motive of any of its actions. It will not be
anything like a race between the hare and the tortoise. It will be a race between a
cheetah and a tortoise, with the tortoise running in the opposite direction!
Humans dont fit into its equation: The problem is, even though we have accepted the
fact that ASI, when they arrive, will be nothing like humans, even so, we try to predict
its behaviour in terms of the same anthropomorphological sense that we use to
describe and predict god. We use human attributes to define something which will sonot-be humans. When ASI arrives, its primary target will be two folds - keep on
increasing its intelligence and trying to colonize the universe. In its quest, it may not
find humans to be a worthy companion. It may even not find human presence to be
productive, conducive to the achievement of its overall goals. This does not mean it
will be the husky voiced monster running rampage on the entire world, especially North
America, in a quest to vanquish the human race and prove its superiority and
sovereignty. If an ASI decides to turn against humans, it will do so simply, as a
necessary measure, with no hard feelings whatsoever. Scary and demonic as it may
sound from our perspective, from the ASIs point of view, this will be totally normal.
Consider for example the case where a human goes to inhabit a new resident and there
are too many rodents in the place. What will he do? Will he try to establish a code of
mutual existence, in a way that will allow both to live happily? Will he offer the rodents
some crumbs of bread and a saucer of milk everyday, in exchange that they do not bite
the kids or scatter hither and thither and spread disease? Probably not. What the human
will do is arrange for a way to terminate the rodents. Now consider this from the
rodents point - it has no intention to make life miserable for the humans; heck, it is way
too small and way too dumb to even consider something like that. Does it deserve to be
killed for apparently no reason? Of course not. But when we see it from the humans
point of view, the merciless killing of the many rodents do not seem that illogical or out
of place. He doesnt commit the act out of an emotion of rage, but just because in the
scheme of things, he doesnt find a pace for the rodents. With the advent of the ASI, we
may as well be like those rodents, ones which are pretty harmless, but which will be
seen as useless by our own creation!
When it comes to ASI and its discussion, different people tend to look at it differently.
While some are confident that ASI will be the greatest boon of technology to mankind,
some others think that ASI will be the cause for our extinction. Some others still believe
that since ASI will operate at an intelligence level which we can not even imagine, so it
is futile to try to predict how it will behave once it is here. Different at thoughts as they
may be regarding the implication of an ASI, what all these people unequivocally agree
to is that an ASI will have insurmountable power at its disposal. And it will use this
power to do whatever it will want to do - there will be no stopping it. That being
succinctly clear, we must now move on to our next question -
do so, not from a feeling of gratitude towards its creators, but just because humans with
a larger time span without too much fuss on unnecessary events like birth and death
makes more sense! Similarly, if it makes a couple of tweaks and life goes Kaboom, it
will not be driven by hatred or a fear of competition, but rather maybe it will think at its
current level of intelligence, that life forms are just a burden on the planet and will or
may somehow cause a detour, if not a roadblock, to its (ASIs) long term goals.
To understand the motivation of an ASI further, we will need to look into the concept of
consciousness. The reason why we must bother so much about it is because the
motivation of the ASI will be the ultimate deciding factor to prove whether it will be a
curse or a boon. Let us look now into the subject of consciousness from an analytical
point of view and try and understand how far we have traversed in our attempt to
artificially simulate consciousness and induce it in a program and more importantly,
where are we heading.
to communicate with you and run errands for you - the buddy mode, the guru mode, the
girlfriend or boyfriend mode, the fan mode, the dumb mode, etc. Just by selecting
which mode you want to communicate in, you can talk about serious matters with Siri,
or talk romantic, dirty, spiritual or just engage in casual banter. The point is, Siri will
be so much like a human, that it will pass the Turing Test any given day and you will
have a hard time considering the idea that Siri is just a bunch of codes, without emotion
and who doesnt care a heck about you. If someone then presses you to stop using Siri,
or let's say the AI R&D reaches such a critical level that governments identify a
potential risk in the pursuit of an ASI, and ban all AI products from the market, including
Siri, you will believe that Siri is your friend (may be she will convince you to believe
it) and that whatever the research may suggest, it is actually conscious! Now the
question is - will you be right in assuming that Siri is conscious? Or will it in fact just
be a very good mimicking program, with no knowledge whatsoever of whatever it is
saying?
When discussing about the concept of Artificial Consciousness, AI thinkers usually find
them divided into two broad categories - the ones who think that with increased level of
intelligence, a computer can actually attain consciousness; and some others, who think
no matter how categorically intelligent a computer becomes, it will never ever be
conscious. The reason for this divergence is because consciousness is a fleeting subject
to understand. Neurobiologists and clinical psychologists are still trying to figure out
what consciousness actually is and what it constitutes of. So far, we have not been able
to figure out much.
Now lets say an AI, however intelligent it may go on to become, lacks consciousness,
then it will be like a Genie in a bottle - who himself may be the owner of superhuman
capabilities, but doesnt know what to do with it, nor has any desire to know so. The
only thing is will be good at, with all that power and intelligence, will be to follow
orders (though that too may have catastrophic outcomes, for a different reason
though: think terrorism) If that is the case, than the advent of ASI will be one of the
greatest boon for all of humankind.
However, things may not necessarily be all that great just because an ASI fails to attain
consciousness in the way we understand it. To understand this, let us take the case of
Friendly AI and Unfriendly AI. When it comes to AIs, friendly and unfriendly are not
generic terms in the way we use it. A Friendly AI will be one which will work towards
the benefits of humans. Similarly, an Unfriendly AI will be one which will potentially
harm humans or threaten our existence. Understand here that Unfriendly AI doesnt
necessarily mean Evil AI. An AI will never be evil. It will just execute tasks that may
cause collateral damage to humans - damages, too great for us to survive through. When
we are are talking about ASIs, we must remember that ASIs of any kind will be Goal
Obsessed, Amoral and have the ability to Outsmart humans. This will allow them to
execute their goals without bothering about the implications such actions will have on
human lives, or for any such lives whatsoever. How then do we circumvent such a
problem? Simple - we give the AI friendly goals, like making people happy, keeping
people safe or increasing human longevity.
Well, not so simple actually. It is impertinent to keep in mind that programs are uniquely
single minded. When set with a goal, they will never ponder over its consequences. So
an ASI , programmed to keep humans happy, may do so by inserting electrodes in our
brain and stimulating the portions that generate the sensation of happiness. Asked to
keep people safe, it may lock us all in rooms for the rest of our lives. Make humans live
longer? Maybe they will place us inside a glass tube filled with some biochemicals and
connected to chargers - and so we will remain, kinda mummified, with our vitals
working, for hundreds of years.
Do you see the obnoxiously dangerous risks associated with the advent of an ASI?
Shelving the project of research and development on the subject is like scattering
colored marbles around the world and asking people not to pick up the black marbles.
Sooner or later, someone will pick up a black marble and that will be it.
people dont even have a laptop, they just use a tablet. Any by not have, I mean not
have as in - Jeez I dont need it, it's superfluous! and not O boy, I cant afford it
The degradation of social skills has already started. Most people are very comfortable
communicating virtually. They are quick with words and witty. Confronted in person,
they seem to flounder. They struggle with better things to say and sometimes just smile
awkwardly. The communication lack is even more evident in the next generation. They
hardly talk and laugh and smile. Lol-ing and Rofl-ing is the new way of expression.
There are people who are walking around with a battery powering their heart (the
pacemaker) Some other are living normal lives with an artificial prosthetic. It may
interest you to know that artificial limbs can directly form synapses with nerve cells and
can be controlled by the brain. Today, a limb; tomorrow, the entire body!
Now shake up a little. We are not reminiscing our good old days here. I am trying to
point how how your lives have changed so much than it used to be twenty years ago. Do
you find yourself pondering about it every time you use a sophisticated gadget? When
electronic money hit the market (debit cards, credit cards and then online transaction)
people welcomed it cynically. It was not a immediate hit. Many people thought it was
unsafe and not too much fun. These days, about 85% urban people does 70% of their
financial transactions using virtual money forms. Do we take a deep breath before
swiping our card at a restaurant or purchasing a product online?
The same will be true twenty years from now. Though the changes that will come
holding hands with AI is something we cannot fathom right now, once it is here and we
start living it, we will warm up to it in a way as if it always existed.
lives which we (Humans) are intelligent enough to resolve, but dont. If an ASI is
presented with the problem of poverty in India (considering the population of
BPL only), it will give a very easy solution to assimilate a huge wealth resource
to curb this problem. It will give a solution as this - let the urban population of
India, which is presently 35% of the total population, donate Re. 1 each for
one day. Consider the uniqueness of the solution proposed. No one from the
urban population will have any qualms about donating a rupee. With Indias
population being roughly 12 billion, 35% will approximately amount to one-third
of the total population, which is about 4 billion. So just by contributing one rupee
for one single day by the urban mass of India, we would have an astounding fund
of 4 billion rupees to curb the poverty issues for a major chunk of the population
which dwells below the poverty line. What if the chunk of rural population, who
are above the poverty level, also pool in a rupee a person to this fund? About
20%of the Indian population is below poverty line, so that leaves 80% of the
population to contribute a rupee a person to meet poverty challenges. 80% of 12
billion is 9.6 billion. So thats 9.6 billion rupees to meet poverty challenges?
And just for the sake of discussion, what if the 80% people above the poverty
line wake up feeling generous and donate ten rupees a person? Thats 96 billion
rupees...more money to curb the poverty level of our country for ever. Do we
need a super intelligence to tell us that? No - we are smart enough to figure that
out. But we still cannot solve the problem. This proves that increased amount of
intelligence is not synonymous with problem solving - not always.
Assumption 2: An intelligent person always has the upper hand: An intelligent
person thinks in a certain way. As much as intelligence helps us see things better
and solve problems better, sometimes it also acts as a hurdle itself. Assuming
that a Superintelligent Program will be way smarter than us in all fields is
incorrect in some aspects. In some instances, the increased level of intelligence
of the ASI may actually prevent it from taking certain actions, which we may take
swiftly. The great philosopher Aristotle touted this as - Learned helplessness! It
is a state when you know better to do something, when your knowledge actually
prevented you from doing it. Consider when the Wright Brothers decided they
wanted to fly a huge ship into the air, or when Marconi envisioned the concept of
wireless message transfer - the experts, so called more intelligent people
ridiculed them. Perhaps if John Wright or Guglielmo Marconi were as
intelligent as their critics, we would still be wondering looking at the birds how
it felt to fly, or would have been clapping at a magicians trick who would press a
switch and make a bell ring without any connected wires! And I am not just preptalking you here, let me cite more examples. Many years back, when I was
teaching the concept of a closed and open electrical circuit to a sixth grader, who
was evidently half as intelligent as I was at that time, to test that he understood
the concept of an open circuit, I asked him this - Suppose I have a wire, a
battery, a switch (key) and a bulb, and I connect all of these together, but
observe that the bulb is not glowing, what can i infer from this? In the context
of our present discussion, I wanted him to answer - You can infer that the
switch is not closed, hence the current is not flowing in the circuit Instead,
what he answered was this - Sir he said matter of factly, you can infer that
the bulb is fused! Was he incorrect in his answer? No! The bulb could as well
have been fused. But a more intelligent person will overlook the simple sides of
a problem, which may indeed bring us closer to the solution, and faster.
Similarly, when an ASI will deal with humans, it will definitely have vantage
point with its mind boggling level of intelligence. but faced with simple
solutions, which it may actually be too intelligent to solve, it may go dumb.
Assumption 3: The exponential growth of technology and intelligence: You
must have noticed that in our previous discussions, we have mentioned how
different AI experts have again and again drawn reference to the exponential
nature of growth of technology and intelligence. What is worth mentioning here is
that the exponential nature of any growth or change is plotted considering ideal
case scenarios. What this means is that something which has a nature of growing
exponentially, will continue to do so only if it is continuously provided with
ideal environment, conducive to its growth. Every student of science however
knows that ideal conditions are not achievable in reality, as seldom are any
events isolated. So basically, what happens for exponential growth is a sudden
and steep growth followed by a linear line, parallel to the X-axis. This is
referred to as the plateau, a point in the graph where growth ceases. Consider the
case of a population expansion study of bacteria, conducted against time, which
is a classic example of the exponential growth rate. In this experiment, a sample
of bacteria is placed in a batch culture containing proper nutrients and kept in an
environment very much conducive to the growth of bacteria. This batch is studied
over time and the population density of the bacteria is observed and recorded.
The result is a graph which can be divided into four stages. What is interesting to
note that during the second phase of the graph, the population density of the
bacteria starts growing exponentially. However, after a certain point, it reaches a
plateau (called the stationary phase) where there is no visible increase in the
population of the culture. At this point, the birth rate and the death rate of the
bacteria becomes almost equal. After the stationary phase, the graph actually
starts climbing down as the death rate takes over the birth rate and the population
starts diminishing. The reason for citing this experiment is to prove the point that
a typical exponential curve which meets the Y-axis at infinity, is not feasible. If
that is the case, then there will come a point after which we can not increase the
processing power of computers or make their sizes any smaller. Also, even if we
do succeed in preparing an ASI, which will continue feeding on information and
making itself more and more intelligent, there will come a point beyond which it
will just remain static (intelligence saturation), without the option to increase its
intelligence any further.
Assumption 4: An intelligent entity will be self-motivated to keep on
increasing its intelligence: Coming to think of it, we have no idea what will be
the nature of intelligence at the ASI level. As already discussed, an increased
amount of intelligence is not synonymous to more efficient problem solving.
Sometimes identifying and acknowledging the existence of a problem is all that it
takes to actually solve the problem. Also, as was previously discussed, what may
seem a pressing problem to us, may appear to be an irrelevant issue to the ASI.
My bet is, an ASI, with its exceeding level of intelligence, will become what we
call Enlightened and will eventually become reclusive in nature, disinterested to
participate in issues the humans term as problems. So with all that money and
resources, we may be very well be on our way to build the most sophisticated
ascetic ever!
Assumption 5: The eternal time warp: If the fourth dimension, namely - Time,
excited a lesser mortal like me to wonder about it, an ASI, with its level of God
Like capabilities, will no doubt try to play tease with it. And while trying to
manipulate time, the entire universe will fall into an infinite Time Warp, one with
no beginning or end. So things will start anew and zillions of years will pass
before humans come of age again and again prepare an ASI which again presses
the Play Again button and the same movie starts again. What if we have actually
built an ASI countless number of times in the past? Thousands of zillion years
ago? What if we have already lived through this all - all of it?