Download `Will Artificial Intelligence Systems Ever Surpass Human Intelligence

‘Will Artificial Intelligence Systems Ever Surpass Human Intelligence’?
A dissertation submitted in partial fulfilment of the requirements for the
degree of Bachelor of Science (Honours) in {Software Engineering}
By Riaz Ali
Department of Computing & Information Systems
Cardiff School of Management
Cardiff Metropolitan University
May 2016
1
Declaration
I hereby declare that this dissertation entitled ‘Will AI ever surpass Human intelligence’? Is
entirely my own work, and it has never been submitted nor is it currently being submitted for
any other degree.
Supervisors Name: Jason Williams
Candidates Signature: …………………………………
2
Date: ………………………
Abstract
Artificial intelligence (AI) has been with us for over 40 years at least the philosophical idea of
it. AI in computers isn’t recent as we have been using the technology for digital devices for a
while now from having GPS maps to talking Phone apps. This study investigates and critically
assesses whether AI will advance on to becoming more intelligent than humans and to evaluate
the repercussions of this if it was to happen.
During the 1980s to 1990s, a whole new, different approach to AI was made rather than to copy
human intelligence; it was to mimic our intelligence, and how our brain performed in the real
world. So thus researchers and scientists have said that this approach to AI had the potential to
be faster and better than us. A full range of review from existing literature in this study gives
an insight of this field by researching the benefits and dangers of AI. This is investigated in
this study to foresee whether AI will ever be advanced enough to surpass our intelligence.
The method of approach was taken by researching past papers on the views and opinion on AI.
This was undertaken by two papers and the compared and contrasted between them for analysis
and findings. The questions that tried to be answered in this study were, could a computer ever
have more intelligence than a human? When is this likely to be achieved and what are the
repercussions of this happening? These questions will be researched and raised in this study.
As insight now the field of AI is ever expanding into unknown territories showing that it’s
evolving over time however certain technologies will need to catch up with AI to succeed.
Conclusions drawn from this study indicates that we are unsure as to when this will happen as
many scientists have different views yet the future of AI growing as technology advances and
is questionable whether humankind can create human-level intelligence systems.
Keywords: Artificial general intelligence (AGI), Ontology, Human-level machine intelligence
(HLMI), Modern AI
3
Acknowledgements
This dissertation was the biggest academic challenge I have faced during my time at Cardiff
Metropolitan University. Without the support of my family and friends during this year
achieving this would not have been possible. Firstly I’d like to thank my dissertation tutor, Dr
Jason Williams for guiding me through this process and offering me help throughout the course
of my dissertation.
I would like to thank my Family and friends also giving me the encouragement and
perseverance for this dissertation to complete it this year.
4
Table of Contents
CHAPTER 1-INTRODUCTION ........................................................................................................................... 7
1.1| INTRODUCTION ....................................................................................................................................... 7
1.2 | BACKGROUND ........................................................................................................................................ 8
1.3 | PURPOSE OF STUDY............................................................................................................................... 9
CHAPTER 2 - LITERATURE REVIEW ............................................................................................................. 10
2.1|INTRODUCTION ...................................................................................................................................... 10
2.2|INTELLIGENT SYSTEMS ........................................................................................................................ 11
2.3|APPLICATIONS OF INTELLIGENT SYSTEMS .................................................................................... 14
2.4|CAN A MACHINE THINK? ..................................................................................................................... 15
2.5|THE FUTURE OF AI ................................................................................................................................. 18
2.6|SUMMARY ................................................................................................................................................ 19
CHAPTER 3 – METHODOLOGY ...................................................................................................................... 20
3.1|INTRODUCTION ...................................................................................................................................... 20
3.2|RESEARCH PHILOSOPHY...................................................................................................................... 21
3.3|RESEARCH APPRAOCH ......................................................................................................................... 22
3.4|DATA COLLECTION METHODS ........................................................................................................... 23
3.5|SUMMARY ................................................................................................................................................ 24
CHAPTER 4 – FINDINGS AND DISCUSSION ................................................................................................ 25
4.1 |INTRODUCTION ..................................................................................................................................... 25
4.2.|ANALYSIS OF SURVEY 1 ...................................................................................................................... 25
4.3 |ANALYSIS OF SURVEY 2 ...................................................................................................................... 29
4.4 |SUMMARY OF FINDINGS ..................................................................................................................... 33
4.5 |FUTURE PREDICTIONS AND LIMITATIONS ..................................................................................... 36
CHAPTER 5 – CONCLUSION ........................................................................................................................... 37
5.1 |REVIEW OF AIM AND OBJECTIVES ................................................................................................... 37
5.2 |REFLECTION AND FURTHER RESEARCH ......................................................................................... 38
CHAPTER 6 – REFERENCES ............................................................................................................................ 40
6.1 |REFERENCES .......................................................................................................................................... 40
6.2 |BIBLIOGRAPHY ...................................................................................................................................... 45
CHAPTER 7-APPENDICES................................................................................................................................ 47
APPENDIX A-ETHICS APPROVAL ............................................................................................................. 47
5
List of Figures
Figure 2.1- How neural network feeds information ............................................................ 12
Figure 2.2- CBR cycle ......................................................................................................... 13
Figure 2.4- Moore’s law progressing through the years ..................................................... 17
Figure 4.2- Results from the survey .................................................................................... 27
Figure 4.3- Graph showing the milestones for each estimates ............................................ 28
Figure 4.4- Graph showing the milestones for each estimates with funding ...................... 28
Figure 4.5- Graph showing when will AI emerge ............................................................... 31
Figure 4.6- The probability of who will develop AI machines first ................................... 32
Figure 4.7- Boxplot of the results showing the creation of human-level intelligence ........ 33
Figure 4.8- Displaying the probability of how AI will look like in the future .................... 33
6
CHAPTER 1-INTRODUCTION
1.1| INTRODUCTION
This project aims to critically analyse and evaluate the process of artificial intelligence and
compare the type of methods that are currently being used with artificial intelligence
programming. The analysis will be undertaken by the student to see the future outcome of
artificial intelligence and its method. To see what direction we are heading in this field and
the philosophical foundations of artificial intelligence, with previous literature done in this
area. The primary purpose of the study is to evaluate the likelihood of whether artificial
intelligence will surpass human intelligence.
Background information relating to the study will be provided in this chapter to provide an
overall understanding of what the study concerns and why it has been chosen. The aim and
objectives of this study are also presented in this chapter.
Chapter two will contain research into previous literature surrounding the types of artificial
intelligence currently being used, sourced from academic books, journal articles, newspaper
articles, etc. This chapter will provide insight into the factors and concerns surrounding
artificial intelligence and will also form a structure for the study to build upon.
The methods that will be taken to complete the study will be discussed and justified in
chapter three. Methodology design and approaches will be established, alongside research
analysis methods. This purpose of this chapter is to produce a plan of how the researcher
intends on achieving the objectives of this study.
The results and findings from data gained via secondary research will be presented in chapter
four, followed by a discussion of the findings. Limitations of the study and future predictions
for artificial intelligence progressing with time as technology becomes more advanced.
Chapter five will conclude the study with a reflection and evaluation of the research
completed. It will also revisit the aim and objectives of the study to ensure they have been
met.
7
1.2 | BACKGROUND
Merriam-webster.com provides a precise definition of artificial intelligence it is “an area of
computer science that deals with giving machines the ability to seem like they have human
intelligence” (Merriam-webster.com, 2010)
Technology has rapidly advanced over the last century and is now used in every aspect of dayto-day life such as the increase of smartphones and digital devices becoming an essential part
of our lives. It has now become the norm for this generation with growing up with the internet
and mobile devices providing a quick response at their fingertips and as a result, now it’s made
us connected to the world around us. The implementation of artificial intelligence in our
devices and the internet is ever present such as when using the internet search giant Google
they use sophisticated artificial intelligence programs to calculate you search return. (Cook,
2016)
The field of artificial intelligence came into existence in the 1940s, the intention was that to
get a computer able to do tasks as a human can which being regarded as intelligent. Which can
mimic humans and then do it themselves or better than the human. 20 years later an open
discussion was made on the philosophical aspects of how close we are on how a computer
machine can be close to a human mind. Also, what types of differences are there between the
two? During this period, it was classed the classical AI. (Warwick, 2012). During the 1980s
and 1990s, a new approach was made in artificial intelligence to build AI to tackle problems
human minds cannot calculate. Real world problems for example such as in medicine, finance
and aeronautics. But this AI has the potential to be faster and better than a human mind and
philosophically speaking the Consequence of this was now the artificial system could
potentially outperform a human brain. (Garrido, 2010) pg1
Modern AI and classical AI approaches work both together synergistically as the classical term
is the foundation to this field of research. Modern AI is the evolvement of classical AI to
improve its potential (Warwick, 2012).What this raises are issues with AI for the future. The
aim of this dissertation is to look at the field of AI in its entirety and to evaluate how close we
are to replicate a human brain to having true artificial intelligence. And what’s the potential
risks this can occur if this happens in the future.
The purpose of this dissertation is to research the question ‘could AI surpass human
intelligence’? Could a computer ever display more intelligence than the humans who program
its system? Would a computer act of its free will and show consciousness to us.
8
It’s imperative before delving into artificial intelligence we look at computers and how a human
perceives them in the real life, computers are programmed in a language that it can understand
what you are trying to do. (Copeland, 1993) The Programmers write in such a way which they
can develop applications and systems for virtually anything. Such as developing algorithms for
applications to predict weather forecasting with high accuracy as an example. Right now
computers have no knowledge themselves of understanding the data this particular limitation
has always been here from the earliest calculating machines to our supercomputers now
performing vast calculations very quickly. (Okandan, 2010)
In summary of this, we can say that during the early times of AI in the 1960s we have attempted
to create complex thinking processes solving various types of problems. But this day and age
AI has changed significantly to now it’s evolved to become different types of knowledge-based
systems which will be discussed further. (Müller, 2013)
1.3 | PURPOSE OF STUDY
The aim of this project is to investigate and critically assess artificial intelligent systems to
identify whether computers are capable of surpassing human intelligence in the future and its
possible implications.
Objectives
• To investigate the current technologies being used in Artificial intelligence field to see how
close we are to human-level intelligence and the outcome of this in the future research will be
undertaken through the use of secondary research methods such as reviewing academic books
and journal articles.
• To investigate how far we have reached in AI and how far can we go in terms of developing
intelligence systems, through the process of referencing secondary methods.
• To critically assess the pros and cons of this happening in the future and what safe guards are
in place for negative developments.
9
CHAPTER 2 - LITERATURE REVIEW
2.1| INTRODUCTION
The initial approaches to AI focus mainly on classical AI. These are the bases for systems such
as knowledge-based systems and expert systems using the importance of rules such as an IF…
THEN… statement. (Lawrence, Palacios-González and Harris, 2016). To develop complex
applications. A characteristic of humanity is that we compare ourselves to others in which we
find certain things we are better than others in different ways. In the words of Marvin Minsky,
a famous scientist in the field of AI has said: “Artificial intelligence is the science of making
machines do things that would require intelligence if done by men.” (Warwick, 2012)
p1.Which we are now trying to develop machines that would do a lot of complex work for us.
To understand how to implement human level AI we have to understand fully how our brains
work to cope with many decisions and to act them out in the real world (De Kamps, 2012).
Right now we still don’t know the engineering difficulties of the brain so what we do is model
this system using AI of what we know right now. This field of technology models and simulate
many parts of the human brain as what an AI would be able to do.
Many researchers believe in Apocalyptic AI, which they generalize from current technologies
and trends to claim that first half of this century will be populated by intelligent machines and
they think by the end of this century they very well could be the only intelligence life on this
planet (Geraci, 2010). These famous authors who said this are Ray Kurzweil, Kevin Warwick
and Marvin Minsky, which believe this will happen. However, others remain sceptical of this
idea ever happening. What this research will show that the systems that are now being used in
AI and what’s the further developments for this in the future. Bianchini (2016) report shows
that AI has evolved into many different types of areas and researching them all would be
inconclusive to this study.
10
2.2| INTELLIGENT SYSTEMS
(Brézillon, 2011) The idea of an intelligent system is that a machine being able to reason with
given knowledge about a particular task and to work in a similar way an expert’s brain would
work. For this to work, the machine carrying out this task would need to be given knowledge
about this also. When new information is being inputted new set of rules will be given to be
able to communicate with the user. This intelligent system is called a rule-based system. The
ones that will be mentioned in this literature review will be neural networks, rule-based
systems, case-based reasoning and knowledge engineering (AUDA and KAMEL, 1999)
A neural network is an AI that was created to mimic accurately what the brain does concerning
our neurons. In our brain neurons pass electric signals from part to another to respond to things.
In a neural network a task is fed through the system to learn quicker and better. They are then
organized in layers which are the nodes (neurons) connected to each other for it pass
information through it and learn from it. The system has one input layer which sends signals to
the hidden layers where actual processing is done via the weighted knowledge of the nodes
then the hidden layers link to an output layer where an answer is shown. (Stafford, 2011)
Most neural networks have a learning rule given, which changes the weights of connections to
the input patterns. This form of AI actually learns by more data and time it takes to calculate
it. For example, a child learns to recognize cats from an example of cats it sees (AUDA and
KAMEL, 1999). What neural networks do is very much the same. Although the various types
of learning rules used by neural networks they use a common form of neural networks which
is backpropagation what this means that it calculates the loss of a node that isn’t the answer to
preventing from looping in the neural network. Thus learning quicker and better.
(Neurosolutions.com, 2016)
11
Figure 2.1 – This shows how neural network feeds information through this process Source:
http://www.neurosolutions.com/products/ns/whatisNN.html
Rule-based systems are designed to help the user to find certain answers which the machine
will ask you a certain amount of questions in which you answer, once the computer knows the
answer it will display it to you. (Chen and Chang, 2011) for example IF (start button pressed)
THEN (start engine) what this shows that it tells you how to start an engine and telling itself
that when the start button is pressed then launch the engine.
This type of AI has its advantages as it’s relatively easy to code in uniform lines of IF-THEN
functions. If a new rule is made it can be added to the system quite quickly but problems with
this system can be gathering information because this can be cumbersome and awkward. When
the experts input this information onto the systems each person can have different ways of
coming to a solution (Baum etal, 2011) also human experts can be expensive to acquire. Some
systems could have thousands of rules which may not make it ideal for the user as it can be
time-consuming. This type of AI is not entirely autonomous as there is a programmer always
inputting new information or updating the rules to this. (Alternative Mindsets, 2015)
(López, 2013) Case-based reasoning system is used for gathering broad types of data for a
particular item. What this system would do is that you would have a certain number of products
and when the user wants to find the best one out of them, they input certain features about the
product. It will attain the best product for you and also be able to give you review data. How
this is done is shown in the image below.
12
Figure 2.2: CBR cycle Source: (Aamodt and Plaza, 1994)
The cycle shows that the user can retrieve case of a product and then reuse that information to
solve their problem and then revise proposed solutions and retain the parts they want from this
system. But of this, you will need large types of data to be acquired either from a website or
from knowledge which can be costly can and time consuming. (RICHTER and AAMODT,
2005)
In artificial intelligence, there are many types of expert systems which aid a human in help
decision making. For an example, these types of programmes are in navigation, health systems
and forecasting financial markets. For an expert system to be efficient, a knowledge engineer
is required to analyse how the human experts make decisions and then translate that into code.
It is vital that the knowledge engineer understands the problems and the information to solve
the problem. In conventional case based reasoning systems a user may not need to provide a
full description of the problem the user can enter text and can describe their problem the
systems will then assist by saying it needs further information to address the solution. Narooei
and Ramli, 2014)
One final aspect of these systems is that it’s just one field of AI and is an active part of
intelligence in general (Warwick, 2012). These systems attempt to replicate the human brain
regarding how decisions are made given a certain number of rules and exceptions based on the
knowledge of an expert. What it does with this knowledge is indicative to see these types of
programmes performed as we expect based on our given knowledge into the system. In many
ways, expert systems are not true intelligence which they do not provide us with the full
autonomous of a thinking AI. Expert systems are not benevolent superintelligence that we
13
would expect from the word itself “AI” nor are they the same as human making decisive
thinking skills. For us to understand what true AI is then we must look at ourselves to replicate
this (Pantic, 2016).
One final aspect of these systems is that it’s just one field of AI and is an active part of
intelligence in general (Warwick, 2012). These systems attempt to replicate the human brain
in terms of how decisions are made given a certain number of rules and exceptions based on
the knowledge of an expert. What it does with these knowledge is indicative to see these types
of programmes performed as we expect based on our given knowledge into the system. In many
ways expert systems are not true intelligence which they do not give us the full autonomous of
a thinking AI. Expert systems are not human level intelligent systems that we would expect
from the word itself “AI” nor are they the same as human making decisive thinking skills. For
us to understand what true AI is then we must look at ourselves to replicate this (Pantic, 2016).
2.3| APPLICATIONS OF INTELLIGENT SYSTEMS
The applications of this type of expert systems in AI is everywhere, in most of our digital
devices from mobile phones to now the first self-driving cars, the first types of AI applications
were used in the 1990s for computer speech recognition which the United Airlines replaced a
keyboard for flight information by using speech recognition. But most users have gone back to
keyboard and mouse as its more convenient. Understanding natural language for a computer to
learn is complex, the computer has to give words a weighted value for it to understand either
positive or negative number this can be used for review data in the case base system mentioned
earlier. (Www-formal.stanford.edu, 2016)
(Dan and Dudeck, 1992) Medical applications the early systems used are a rule base system to
calculate the problem you have, when given a set of answers. One of the first expert systems
was the MYCIN in 1974 which only diagnosed infections of the blood and treatments it did
better than the student and practitioner while being very limited to its information but still the
knowledge has to be inputted by the expert anyway. (Russell and Norvig, 2010)
Google translate is another AI system which is good at a narrow task which records your voice
and have the translation given out in another language this is a very narrow system as it is
designed to do one thing only translation. Another would be is the best board (Wallace,
McCartney and Russell, 2010) game players are now AI systems which they can deduce the
best ways to win every time. And what’s recently happened this year is that a game called Go
was played by google deep mind and beaten the best human go champion through neural
14
network understanding. (BBC News, 2016) The hugely popular Google search uses narrow
intelligence with sophisticated method’s to show you the specific pages searched, another
aspect would be AI traders can now make a decision much quicker than a human and now
certain trends by using calculations much faster than a human. (Sheu and Wei, 2011)
In this day and age, artificial intelligence is growing exponentially everywhere in all sorts of
areas from ranging to everyday lifestyle to space programs. Why this types of expert systems
are being developed further because we as humans can tell the machines what to do, tell them
a set of rules and then will be able to calculate it at much faster rate than we ever could. Giving
us an answer what this is now is just artificial narrow intelligence (ANI) designed only to find
specific tasks and not to take multiple tasks such a human would do. (Müller, 2013)
2.4| CAN A MACHINE THINK?
What’s been researched so far was the expert systems and their uses in the real world right
now, these systems cannot think for themselves as we the user of the programme has to input
the information or knowledge ourselves. Right now we are at ANI systems or weak AI what
this means that that these systems are designed to be only good at one specific thing at any
given time and cannot produce other decisions, such as if its playing a board game it would
only be good at that. We cannot tell it to find a better way to autonomous driving. Müller,
2014)
(Hern, 2014) Alan Turing developed the Turing test in the 1950s the test was designed to see
whether a computer was intelligent or not. The test was created to determine whether the
answers given to a human, was either by another human or another computer, if the human
conducting the test was unable to determine either a computer is answering or a human. It has
deemed passed this test of intelligence. Showing that a human cannot differentiate between the
two shows the computer is intelligent itself. (Kurzweil, 2012) Ray Kurzweil have stated that
no robot would pass this barrier before 2029 the robot must be able to convince two of the three
judges that it’s human to pass. Right now this competition has been held for twenty years, and
still no machine has passed this test. This is due to the complex task of being a human in terms
of language processing and mimicking this behaviour. (Warwick, 2012)
Artificial general intelligence or referred to as strong AI means that it’s a smart as a human in
all aspects such as performing any intellectual task any human can at the moment. Creating
AGI is a very difficult task than creating ANI as were yet to do this. Professor Linda
15
Gottfredson (Warwick, 2012) p6 describes intelligence as “a very general mental capability
that, among other things, involves the ability to reason, plan, solve problems, think abstractly,
comprehend complex ideas, learn quickly, and learn from experience.” AGI would be able to
do all of those things as easily as you can. (Wait But Why, 2015) however it’s even debatable
whether human intelligence is truly general, although we are better at some cognitive task than
others (Hirschfeld and Gelman 1995) human intelligence is more advanced than non-human
species intelligence (animals). (Goertzel, 2014)
But as of now machines are capable of doing things such as communications, but they do not
understand this process. Only being told what to do (Bostrom and Yudkowsky, 2011), so this
doesn’t display intelligence at the moment. But another concept that Warwick says that animal
voices can interact one to another. But this shows intelligence although we cannot
communicate with them perhaps what Warwick is saying that computers may understand
things and interact in different way to us. As of now this advanced technology doesn’t exist
right now. Strong AI term is machine thinking in a similar way to us displaying consciousness
like how humans do. We still don’t fully understand ourselves to know how a machine like us
would think. (Bostrom, 2012)
(Wait But Why, 2015) The road from ANI to AGI is an uncertain one what needs to happen
for AGI to be a possibility is an exponential increase of computer hardware and processing
power. If this system is going to match a human’s brain, it will need to have the computing
power as human’s brain to do that. A possibility to do that is how many calculations the brain
could manage and figure out the structure of the calculations per second (CPS) and total them
together. Ray Kurzweil made a shortcut of this by asking someone knowledgeable of the CPS
and the structure weight compared to the brain and then multiplied. Which he deduced the
number of 10 quadrillion calculations per second. (Telegraph.co.uk, 2013) The world’s fastest
supercomputer which is china’s Tianhe 2 has beaten that number by clocking at 34 quadrillion
cps but the size of this supercomputer is huge and taking up to a warehouse in size (Wait But
Why, 2015) Ray Kurzweil (Kurzweil, 2012) suggests we think of average computers that can
reach 10 quadrillion cps which could mean that AGI could be a real world application and part
of life
(Webopedia.com, 2016). With Moore laws states that computing power doubles every two
years this exponential growth is shown in the graph below showing the trajectory of its growth
significantly through the years. (Wait But Why, 2015)
16
Figure 2.4. Moore’s law progressing through the years. Source:
http://waitbutwhy.com/2015/01/artificial-intelligence-revolution-1.html
(Wait But Why, 2015) Currently, the world’s £1000 computers are now beating a mouse brain
in terms of (CPS) while this doesn’t sound like a lot, but we were at a trillionth of human level
in 1985 and then at 2015 we are a thousandth of human brain level. This forecast puts us on
affordable computer by 2025 that can rival the power of the human brain shown above
(Goertzel, 2014)
(Russell and Norvig, 2010) While this is the hardware problem being sorted in a few decades.
The software problem is a difficult task because it is how the machines can think like us which
is the primary goal of artificial intelligence. Building an AI brain just as powerful as ours is a
complex task. Evolution could be better a method the term genetic algorithms is beneficial for
this. For an example a group of computers would do certain tasks given and the most successful
would be bred with each other producing a better new computer that learns more and over time
more versions of this. Matyszczyk (2014) says that Stephan Hawking fears once this happens
AI could be potentially life threatening to us. The advantages could be our forced evolution on
computers can take mere decades instead of us taking billions of years. The concept is that we
17
build a machine that can change its code and architecture to make it better and to allow it to
learn over a period of time by itself. However we could teach them the fundamentals of
computing to make themselves smarter and faster. (Bostrom and Shulman, 2016)
2.5|THE FUTURE OF AI
The road from AGI to ASI (Artificial Super Intelligence) may be short timed, at some point we
reach this level of AI, AGI would not only be on the same level with us, but it will have
significant advantages than us such their speed will run faster than our brains neurons already
does (Müller, 2013). Microprocessors can run at 2GHz and computers can, in theory, reach
speeds of light which make them already faster than ours will ever be. The storage of these
AGI systems will have space far greater than ours and can be stored externally virtually
anywhere. A setback can be that just like us computers tend to deteriorate over time and need
to be replaced and humans brains also get fatigued easily but computers peak performance can
run almost all the time (Okandan, 2010).
(Wait But Why, 2015) Software for a computer receives regular update and fixes that can be
experimented on itself where as a human brain it’s impossible for this to happen like that. AI
machines could be able to create its body and sensors along with-it. AGI would vastly improve
itself in a short amount time. The progression of ASI systems would be rather quick but from
ANI systems to AGI systems would be considerably slower because this milestone is achieved
yet since it needs to understand the world and itself. Because an ANI system that doesn’t think
like humans it only performs specific tasks. What this is called an intelligence explosion where
its growth exponentially raises in the least amount of time. (Baum, etal, 2011)
In the future if this is the possibility what are safeguards in place to stop from AI from killing
the human race because it could see us as humans as insignificant. Or that AI could have moral
status within it by researching humans and see what we are like. (Hardawar, 2016) Stephan
Hawking states that evolved AI will have similar goals to living organisms but not reproducing
like us only to rather collect more resources to understand more about the world. Elon Musk
and Bill Gates founder Microsoft are concerned that with the rise of AI there are potential
threats to humans (Sainato, 2015). The term AI refers to a broad field of technology now it’s
hard to say what will happen if AI becomes super intelligent as we are not at that stage yet but
many philosophers and scientists are questioning what will happen if this ever occurs. (Potapov
and Rodionov, 2014)
18
2.6|SUMMARY
The use of AI in our lives plays a vital role in how we able to do things now. This ever
expanding range of AI technology and innovation is becoming the dominant factor in how we
interact with computers. (Warwick, 2012) For AI to develop further, more research and more
innovation is needed in this field for true AGI to emerge. Widespread AI in our lives has
changed for the better, but this just show’s our dependency on this type of technology now and
relying upon this no doubt as this becomes more advanced in the future more of our lives will
change forever. (Goertzel, 2014) However, this is just the beginning of AI as we are at the stage
ANI and can do simple tasks using the intelligent systems that has been mentioned above this
is merely a foundation of what is left to come. (Baum, Goertzel and Goertzel, 2011)
The shift from philosophical AI in the 1950s to now, AI has dramatically increased to where
we understand AI concepts and trying to break the barrier to creating new intelligent systems.
(Okandan, 2010) Drawing upon all the existing literature discussed its vital that the continuance
of AI remains for us to learn about machines and what they ultimately mean for us.
Nevertheless, there’s also other considerations to revise such what is the direction we are
heading to in the future of AI and the potential barriers to this (Potapov and Rodionov, 2014)
Despite the breakthroughs in this field of AI, many scientists and experts believe that more of
the advancements in this area is yet to come (Phys.org, 2016) which could mean we are decades
away from this. Researchers now are creating safeguards to ensure AI will perform in a safe
manner even in situations unforeseen by us. (Müller, 2014) The managing director of
Microsoft, Horvitz says "We have to stay vigilant, be proactive and make good decisions,
especially as we build more powerful intelligence, including systems that might be able to
outthink us or rethink things in ways that weren't planned by the creators.”
19
CHAPTER 3 – METHODOLOGY
3.1 | INTRODUCTION
Chapter one outlines the purpose of this study to gain background knowledge and explore
artificial intelligence history. With comparative views on AI exploring the types within this
field of study. It outlined the aims and objectives of this study and how it would be reached.
The main aim of this study is to gain knowledge of artificial intelligence and will it ever surpass
human intelligence in the future. Which could potentially affect us all. By gathering sufficient
relevant data from books, journals, etc. thus enabling a critical evaluation and assessment of
views and the standpoint of where we are with artificial intelligence which will give a
viewpoint of how this field has grown over the years.
Data collections usually take place after the research problem has been clearly defined and then
its designs plan clearly outline, this helps to order the method of data in a hierarchy. To obtain
useful information to acquire relevant information the suitable approaches has to be outlined
in regards to research and data collection methodologies. With research into various types of
methods and the drawing advantages and disadvantages discussion to the specific research.
Approaches were to gather historic knowledge and early stages of AI from secondary research
and then researching new types of AI and the future.
It’s essential to researchers to consider the suitable methodology required for this study before
undertaking any research to determine a result. Which gives a start point and direction to make
sure the reach is adequate and on the right track while also giving a reason in for the choices
made acquiring the data and manipulation of it and analysing it. During this chapter expresses
and substantiates the research philosophy chosen, approach to the design and the methods
preferred of a collection of data and the analysis.
20
3.2 | RESEARCH PHILOSOPHY
Ontology is the philosophy that relies on knowledge of formal naming and definition of types
and the theories of and relationships and entities that are fundamentally existing (UOC), 2013).
This kind of study the for AI systems is what ‘exists’ is that what can be presented.
Computational ontologies are meant to model a structure of a system such as the relevant
entities and relations. The ontology engineer analyses the data and entities and organizes them
into concepts and relationships then being represented. The backbone of ontology philosophy
is that consists of hierarchy and specialization of concepts (Saunders etal, 2012).
Another philosophy that’s relies on knowledge would be epistemology it shows what we learn
from our senses when we cannot gain understating of knowledge (Wheeler and Pereira, 2004)
what this means is that how machines will understand us without senses. There are others who
debate with emergence of AI and its matters. To be familiar with this subject you have to
understand where we are with AI.
(Woods 2006). Two types of data collection can be used these are primary data and secondary
data. Primary data is the data which has been conducted by a person for new knowledge.
Whereas secondary data has been acquired from existing work. Qualitative data is data that is
ideally new data that can be analysed in understanding new ground, and other people’s opinions
on the subject matter, the examples of qualitative data are open interviews and can be very
precise and in-depth. Quantitative data is a method to gain knowledge about a problem through
acquiring either long texts of data or mathematical equations which can be used for statistics
and analysis. Quantitative data in in knowledge of the program and the scale of it and showing
a result (Slevitch, 2011).
An observation method analysis is the data gathered as results of the person conducting direct
primary research. Therefore it becomes a passive way for the respondent. (Kothari, 2005
researching an observation is methodical which is a logical tool for research when applying
these types of methods one should understand how these observation will be recorded these
methods may very well be structured or unstructured dependant on the conductor of these
observations. The advantage of a direct observation is the one sided views are taken away and
leaves the conductor of the observations deduce their analysis of it and be responsible for their
findings. This also requires the respondent to participate willingly and not taken out of their
will to participate as of being voluntary by themselves (Kothari 2005) but as the shortcomings
of this, it can be costly as it requires more responds voluntary take part in this which are also
21
time-consuming. Also, may render the data unusable as there’s too much of data that can’t be
analysis in a given amount of time which also has detrimental effects on the sample size also.
This study will only be researching secondary data methods attained by other journals that have
been collected for primary data, from two journals that have already been carried out for the
public and this research will analyse and evaluate in this study showing the outcome of artificial
intelligence in the public.
3.3 | RESEARCH APPROACH
(Goertz & Mahoney, 2012) States that quantitative research is the research of statistical
methods and theories. Which primarily focuses on numerical data types and structured
meanings to attain knowledge from. This approach is for challenging theories which are
objective in nature through an examination of the variables. (Creswell, 2013, p4). This type of
research is acquired via closed questions and already defined answers for you to make your
own assumptions and conclusions about. Which then helps you test the statistics for further
research and facts to be made. Although this is only reliant on the questions asked and how it’s
attained. (O'Hara et al. 2011)
(Goertz & Mahoney, 2012), however states that the qualitative research is set in logic and the
theories. Which focuses on human thoughts and emotions rather than statistics and numerical
data this predefines it as understanding humans as group problem. (Creswell, 2013, p4). This
type of data can’t acquire through primary research of surveys questionnaires and open-ended
questions thus then the findings are made by seeing individual responses see patterns and
theories and testing their hypothesis to see if it’s true. (Creswell, 2013,) The final analysis of
qualitative data is made by the conductor of the primary research undertaken. And then will be
evaluated to find a conclusive answer.
As stated above quantitative data methods will be used only in this study as this research does
not require input from the general public. Because it is to see what previous people have said.
But nonetheless researching other findings and their surveys will show a collection of data
from their primary sources. This will be analysed and evaluated in the next chapter to obtain a
more in-depth understanding of what AI means to experts and their viewpoint of this type of
field.
The key concerns of data are accuracy and credibility when attaining this information (Rhodes
et al. 2014) is that that using multiple ways to evaluate and assess a theory greatly increases it
22
reliability of the end findings. This shows that using both will advance the legitimacy of the
study but as mentioned before only one type off information will be gathered using previous
authors work and journals. But as a future note adding multiple data methods to this will
validate the findings further as a much more varied collection to evaluate but as AI is still in
its infancy perhaps it’s not required this early stage.
What’s been established that that deductive reasoning will be linked to quantitative research
was undertaken. Deductive reading is the testing of the hypothesis that already been predefined
(Saunders et al. 2012) as the hypothesis has been developed in the result of the analysed data
and the study question itself. The hypothesis has been made and will be tested with the
quantitative data to find correlations and theories behind it and then further researched using
the same methods to interpret own findings and the explanation of the results. Indicative
reasoning will not be linked to this subject as the qualitative data has not been carried out in
this study rather acquiring others survey findings and making an analysis of it.
3.4| DATA COLLECTION METHODS
A sufficient amount of data will need collecting to achieve the aims and objectives in chapter
one only one collection method will be used which is quantitative data collection as this will
be sufficient for this study.
The method of data collection will be attained using the survey conducted by (Baum etal, 2011)
this will firstly explore the questions asked by the conductor for the participants to create a
background for the research of the online survey. This will concern 888 participants from all
expert backgrounds of the first question. Types of ages and backgrounds the first question given
is the most important, to see when people think AI will surpass human intelligence. This survey
had a good amount of participants showing very good statistics overall
The survey was online only, because it now gathers more participants than ever before it is also
very time efficient as you done need to conduct the survey yourself rather the participant
themselves would it. So it can be quickly analysed and the participant doesn’t feel empowered
by you to complete it which could prove fairly reliable results. Once these surveys has been
analysed this provide the foundation of the hypothesis written this will show the relationships
and theories behind true AI and explain why such patterns exists.
Another survey conducted by (Sandberg and Bostrom, 2011) wrote questions about AGI and
who will develop it first these two surveys are equally important that outline questions relevant
23
to the study showing where are with AI and how far have we got till AGI. Although this survey
had a response of 32 which is only a small indicative of what experts think of AI.
3.5 | SUMMARY
Using prior authors work for analysis and evaluation to gather sufficient amounts of data for
the study research and analysis of these two papers will be required to have a conclusive
findings and discussions.
The study will use high levels of quantitative data analysis for the best findings and results for
these papers to summarise this the present findings will be displayed in graphs and table
formats to understand the data easily. (Saunders, Lewis and Thornhill, 2012) For this to be
successful through research will be undertaken to see the outcome of the research. Quantitative
methods will vary on the collection data and the method of analysis to summarise this. The
data collector must understand what data is being used and if it should be used in their findings
and be accountable to this when acquiring the data.
The purpose of mining information is to be useful and help towards the research problem. But
there’s various amount of techniques required to do this. As this research is an open field, and
the theory of AI is not known to everyone. Having qualitative data in the study would be
irrelevant however past researchers and authors have conducted primary data which will be
extracted and be analysed in the next chapter.
24
CHAPTER 4 – FINDINGS AND DISSCUSSION
4.1 | INTRODUCTION
The data through secondary research has been used for this study and has been critically
evaluated and analysed to answer the research question. Only quantitative approach has been
used for the collection of the data where others have used to conduct their studies on AI. This
chapter will present a breakdown of the findings occurred and the results in a descriptive
manner. The predictions and the future and the limitations based on quantitative data is
provided in this section. Which will ultimately conclude with a summary of findings to
hypothesise the research question to see if it has been met or whether further research is
required upon this.
4.2. | ANALYSIS OF SURVEY 1
The data obtained from the journal (Baum etal, 2011) which has been analysed by the author
into graphs and tables showing initial participants for this first question as 888 participants to
provide a good analysis and evaluation on the study. This study is a credible source and with
very accurate data as the participants where researchers and scientist in the field of AI.
4.2.1 |RESULTS OBTAINED
The number of participants who completed the first question was 888. Who it was taken by a
futurist Bruce Klein. The question that was asked is “when will AI surpass human level
intelligence?” the distribution of responses is in Fig. 4. (Klein 2007) study shows that a high
percentage (26%) of researchers believe that it will be surpassed within the 45 years which is
well within our lifetime while the next biggest portion (17%) believe that this will happen in
the next 15 years making it much sooner to come. (Baum etal, 2011)
25
Fig. 4.2 Results from the survey taken by Klein [29] in 2007 asking the question “When will
AI surpass human-level intelligence? Source: (Baum etal, 2011)
As this survey was conducted online but it was undertaken by researchers and other
individual’s mainly pursuing a career in this field of work of AGI. Other questions in this
survey were conducted attained far fewer participants to see the milestones of AGI and the
technical approach to achieving it, and the ethics of AGI question will also be shown. The first
question seemed very insightful to AGI as it’s just a simple question. However, the second part
of this survey (Baum etal, 2011) had followed up questions customized for the researchers and
scientists which were distributed to via email. The follow-up questions gave a more an
understanding response and meant from the first questions.
The first set of follow-up questions prompted the researchers and experts beliefs about AI when
they reach four milestones. For each milestone it was various levels of AGI two versions of
questions were asked one was with funding and without funding of AI with a hypothetical £100
billion per year is over the amount needed. This is only to ensure the money wouldn’t be scarce
when developing AGI. The eight milestones questions had probability data presenting in 10%,
25%, 75% and 90% to see when AGI will surpass each milestone.
26
Fig 4.3. Showing the milestones for each estimates when they will pass it what the experts say
about it. Source: (Baum etal, 2011)
Fig 4.4. Showing the milestones for each estimates when they will pass it what the experts say
about it but this time with funding. Source: (Baum etal, 2011)
What this shows that the experts think that with funding to AI we are likely to achieve these
milestones a lot earlier with most experts saying that we could reach superhuman levels in 2020
during that decade but without funding many experts say that this may happen or at a later stage
or perhaps possibly never will. The analogy of human intelligence is that the first level system
will be very close to mimicking human intelligence and that reaching Nobel Prize level of
intelligence will take much longer to achieve. (Baum etal, 2011)
27
4.2.2 | LINKS AND RELATIONSHIPS
Within the data obtained links between the questions were observed and found. The difference
in the milestones with and without funding was only a few years on variances. However few
experts believe with huge amounts funding AGI will be later stages a reason behind this would
be they would primarily focus on monetising this field and becoming dominated. Philosophical
views on the power that AGI has which could just push back on an even later stage. Others
have noted that with increased funding that could produce knowledge of what needs to be done
for AI to progress and this type of knowledge will not be likely known to the company that
distributes funds rather they want to know only the outcome and not the repercussions.
Few experts gave a reason on why funding would not have an impact on AGI this is because
its needs more philosophical and theoretical study rather than investment into it another said,
“I believe the development of AGI’s to be more of a tool and evolutionary problem than simply
a funding problem. AGI’s will be built upon tools that have been developed from previous
tools. This evolution in tools will take time. Even with a crash project and massive funding,
these tools will still need a chance to develop and mature.” (Baum etal, 2011) What this shows
that perhaps funding isn’t a necessity in this area as we are at the early stages, and we have to
understand what the result of human level AI is in the future.
The participants showed a various range of levels of certainty in their estimates. However, few
showed uncertainty which indicates that these milestones are with very narrow timeframe but
the others giving this narrow ranges showed they were optimists in the area of AI. Some of the
individuals in the report showed that the milestones were precise and an effective time scale of
when it’s going to happen. However, one expert stated that the Turing test at least will not be
achieved until 2020, and there was a 10% chance of it ever happening and another person has
estimated that it will not be reached until year 3000. As this research was taken in 2007, one
person has stated that it will be achieved in 2016, but this is yet happen.
Amongst the experts of this survey, the agreement of order in which the milestones would be
achieved was little, and the section of concern was superhuman millstone would be achieved
at the same time as the others. Many experts believe that AGI from this survey do not think the
first systems will mimic human intelligence rather Nobel winning AGO is going to take much
longer and the milestones are just an indication of what the experts believe in this survey and
how these individuals see future of AI.
28
4.3 | ANALYSIS OF SURVEY 2
(Sandberg and Bostrom, 2011) The data collected from this paper addresses different questions
from the first survey of AI this paper focuses on what people think of AI and who develop this
type of technology and why also the repercussions of this happening and its affects. This study
is from the reliable and credible source of Dr. Nick Bostrom and Anders Sandsberg also says
this data was obtained from participants of the field of AI and other various types of professions
which will be mentioned later. This Survey will differ considerably from the last because of
the types of questions asked in this survey.
4.3.1 |RESULTS OBTAINED FROM SURVEY 2
The participants who answered the first question was 35 who was undertaken by (Sandberg
and Bostrom, 2011) and the question was “Time estimates for human-level machine
intelligence (HLMI) 1. Assuming no global catastrophe halts progress, by what year would you
assign a 10% /50% /90% chance of the development of human-level machine intelligence? Feel
free to answer ‘never’ if you believe such a milestone will never be reached”. This question
showed 10% said that (HLMI) is reachable within 2028 and minimum of 2015 while 50% say
that 2050 there’s a change HLMI and by the year 2150 90% of the participants stated that there
will be HLMI.
29
Fig. 4.5. “Boxplot of the time estimate responses. The red line denotes the median, the top and bottom
of the boxes the 3rd and 1st quartiles respectively, and the whiskers extend 1.5 times the median quartile
distance. Outlier points marked by red crosses. The diagram excludes two extreme outliers. Source:
(Sandberg and Bostrom, 2011)
What this survey displays is that majority of the people who have taken this first question think
that HLMI is imminent in the future even though it’s another hundred years away. The
participants who undertook this survey wherein areas of profession such as AI and robotics,
neuroscience, physics, psychology, mathematics and literature these various academic
disciplines is good to display different types of decisions and answers that people make and
not just the field of AI.
The second question is an important one as to who will develop HLMI as the graph shows that
the most common was the military and the academic principles with smaller estimations on
groups such as industry perhaps because AGI at the moment has no uses for the industry.
However one participant noted that the military would use AI from the academic institutes and
change it, however, they may like it. (Sandberg and Bostrom, 2011) and assembly scientific AI
breakthroughs will be covered by state projects if they are an interest the military. The other
category could produce large-scale AI from narrow intelligence as they are already working on
it, and it takes a breakthrough it happens (Sandberg and Bostrom, 2011)
30
.
Fig 4.6 This shows that the probability of who develop AI machines first source: (Sandberg
and Bostrom, 2011)
31
Fig. 4.7 boxplot showing how positive or negative the creation of human level intelligence
likely to be for us. The possible fields were: “extremely good”, “good”, “neutral”, “bad”, and
“extremely bad”. Source: (Sandberg and Bostrom, 2011)
This display shows that the responses from this question is mixed reviews on this subject matter
whether it will be good for us or not the majority think that it’s extremely bad for AGI as they
could see us a threat if look to ourselves our history is paved with domination and control.
However large part think that this will be neutral for us working in ways that can benefit both
of us in ways me may not be able to comprehend yet. Which was given to either side of
extremes to as to AGI can pose extreme benefits and possibility of working together.
Another question that’s important it the study is how similar will machine intelligence be to
human intelligence? Will they be very similar to us as we are creating it? 32 responses to this
question 8 has said that biologically inspired machine is most likely to happens and 12 thought
that it will be brain inspired AGI and 12 thought said entirely de novo but this result is not
distinct from the cases as there’s even amount distributed between the three.
Fig. 4.8 displaying how AI will look like in the future. Source: (Sandberg and Bostrom, 2011)
32
4.3.2 | LINKS AND RELATIONSHIPS
This survey was itself selected group so the results should be cautious and the small number of
views did have a have a few correlation and the survey was limited reliability and validity as
this was in-depth enough. The views were express that AGI was most likely to happen during
mid-century or possibly next century away and that its high outcome form being created by
large organisations rather than smaller companies and that it can be dangerous of AI can
catastrophic to us as it perhaps might want to wipe us out. However, there is uncertainty to this
field from this reach, as others think that positive outcome can be plausible too.
4.4 |SUMMARY OF FINDINGS
The response rate from both questionnaires was high and sufficient for the analysis of
quantitative data attained. The participants for (Baum etal, 2011) paper showed that although
many similarities were found between these two papers, the participants were more optimistic
and pessimistic of the outcome of AI. This evident in the results discussed in section 4.2 and
4.3.The was limitations and exceptions to this as (Baum etal, 2011) had not only addressed the
question of whether AI will surpass human intelligence but the milestones to achieve this, was
also predicted.
From the findings of (Sandberg and Bostrom, 2011) the questions asked was also more the
ethical side of AI and asked participants whether it will be good for humankind or not. This
was a significant difference between the two findings as one was showing the milestones for
AI becoming truly intelligent, and the other was display the findings of what people thought of
AI when it gets to that stage. This was interesting as two papers had very different outcomes.
One being the ethical issues concerning with AI and the other predicting the outcome of AI
and how long it will take.
Both papers have similarities as they are trying to find out the research question of when ‘AI
will surpass human intelligence’. Both have milestones and predictions which were carried by
participants in a professional field to produce the findings. The study discovered that both of
these survey questions had a lot in common but ultimately have different views.
What was discovered in the (Baum etal, 2011) paper was that participants were more
pessimistic of AI passing human intelligence. The results showed that more people have said
that the emergence of AI passing human intelligence is likely to happen within 50 years this is
because the participants were only in the profession of AI and these experts think this is the
33
most predictable timescale. But results from the paper (Sandberg and Bostrom, 2011) suggests
otherwise, which the experts say it’s likely to happen in the next 150 years or more from this
information is displays that the participants have different professions such as philosophy,
cognitive science, machine intelligence, journalism. So that they understand that particular
technology is not advanced enough.
The other underlying factor identified in papers are economic and social issues that have arisen.
The paper (Baum etal, 2011) expresses that with funding of AI few experts have said that it
will only delay the development of AI and because the researchers would focus on making it
more financial and administration instead of on research. This study brought to light that having
a substantial funding amount will not only delay the time of AI, but companies will think to
monopolise on this new type of technology instead.
The survey was a quantitative survey with closed-ended questions which was asking an
outcome in binary yes/no. However, the participants were encouraged to write any remarks to
clarify what they said in more in-depth. Most experts have expressed their feelings towards AI
and that they have stated that the milestones in (Baum etal, 2011) will be achieved in a few
decades. From the results imply that there’s agreement between the two papers of the impact
of AI, and the embodiment of AI will take will it be either close to biology, completely artificial
intelligence robot body or something else entirely.
The study has found that both papers have answers from experts on what they believe the
impact AI can have to us in society whether it will be good or not as of now we are unable to
predict the implications this will have based on these findings alone or until AI has been.
From the first survey, there was a high response of participants for the first questions which
was conducted in an online survey. On analysing the quantitative data collected from the first
survey showed various types of links between each question and were found that experts had
different views and standpoints of AI these results imply that these experts have knowledge in
AI and would different aspects rather than the public answering the survey.
From the findings in chapter 2 shows that the applications of AI is still in its infancy and
requires more development in this field to advance further. What research and philosophers are
saying that we have decades of research left to get to the stage of where we want to be.
(Warwick, 2012). From the research gathered in this section, displays the various types of
applications in AI.
34
Applications such as neural networks is a promising field for the growth of AI as many experts
are now creating complex tasks for computers to compute large calculations (Chen and Chang,
2011) which we cannot do. However, these systems show us that they can perform good at one
task only and not many. From the literature findings implies more research is needed although
some aspects would be reached in mere decade or so such as the hardware problem of AI, this
because of Moore’s law stating that computing power doubles every two years meaning the
capabilities of AI is within our reach and doesn’t seem like science fiction
(Neurosolutions.com, 2016).
But the software problem does remain the crucial point of AI at this date and time we are yet
to solve the problem of AI becoming truly intelligent as us. (Müller, 2014) This because we
simply don’t understand what we need to develop regarding software for AI. Right now we
only have programs that can mimic certain parts of our intelligence such as neural networks
they learn over time, and we use natural language processing for the computer to understand
us this doesn’t mimic our intelligence as a whole which doesn’t give it the term AGI
(Alternative Mindsets, 2015).
It’s inevitable that people's knowledge and understanding will increase in time about AI as it
advances and becomes part of everyone’s life for the better. But from the finding’s it shows
the experts remain worried about the nature of AI whether it will be safe or have a different
agenda it’s imperative to understand the methods of AI being used now and the theory behind
if you were to go and research into this field of computing with more undoubtedly innovations
in the future there would be more applications AI ever present in the world around us. For us
to actively have a debate and discuss on this matter.
The idea that AI concept seems alien to people when talking about this they cannot understand
the idea of this that humans creating smarter intelligent being than themselves which at the
moment AI isn’t really developing towards this, rather algorithms and networks that can work
out complex problems for us being only narrow intelligence that it designed for one job only.
Another aspect that’s been researched is that we are unsure of the safe guarding place when AI
becomes sentient and intelligence what does this mean to the human race as computers are very
different to us many in things. Such as they don’t require like what we do also how can we
control it. In one aspect perhaps we can’t compete as with intelligence it grows over time and
understands and maybe we cannot give it rules to obey.
35
4.5 |FUTURE PREDICTIONS AND LIMITATIONS
What’s interesting is to compare both (Baum etal, 2011) and (Sandberg and Bostrom, 2011)
where this survey was conducted in a conference, and the first question was online survey by
Klein just being revised. This survey placed the estimation of 10% of beating the Turing test
and passing the third grad test and doing the work at 2020 and the 50% chance at 2040 for the
Turing test and 2030 for the third-grade test and Nobel at 2045. The participants for this test
were very pessimistic at the upper end of that test too especially, this had a very broad spree
and lack of consensus which estimated a high risk to humanity but the second survey estimates
that the approach of AGI will be most likely to be achieved in the later stage of this century or
the next. The survey listed large number specific methods applied when conducting this as to
survey already being 4 years old from the (g) survey which showed now advancement in AI or
beating the Turing test.
(Baum etal, 2011) Also mentioned that the survey was done by Klein about when AI will
surpass human intelligence (Klein 2007). This survey was mostly dominated by optimistic
participant’s and has the range of up to the year 2050 AI will be intelligent thus both surveys
are similar in many ways, but they remain pessimistic on the timescale as to be a longer time
or even the feasibility of will AI surpass human intelligence.
The survey that was used was very simple, and limitations of the questionnaire contribute to
some reliability and validity for this progress further questions need to be asked in order to gain
a well-rounded conclusion of what experts think about AI. From an insight of the findings this
shows we may possibly be another 40 years away from AGI.
36
CHAPTER 5 – CONCLUSION
5.1 | REVIEW OF AIM AND OBJECTIVES
The overall aim of this study was to critically assess and investigate whether AI will surpass
human intelligence the aim was accomplished through achieving all of the three objectives the
objective are presented below along with the explanations of they were met in the chapters
above.
To investigate the current technologies being used in Artificial intelligence field to see
how close we are to human level intelligence. The outcome of this in the future research will
be undertaken through the use of secondary research methods such as reviewing academic
books and journal articles. This was achieved through explanation of existing literature and
thorough research into existing technologies that we have right now and what the next logical
step towards AI.
To critically understand how far we have reached in AI and how far can we go in terms
of developing intelligence systems, through the process of referencing secondary methods.
This was achieved by looking at the findings which were derived from the analysis of
quantitative data collected via primary reach of another Paper.
To critically assess the pros and cons of this happening in the future and what safe guards
are in place if this ever happens. This was accomplished through also the findings and
mentioned in chapter 4 this was evaluated on chapter 4 and discussed in chapter 5.
37
5.2 | REFLECTION AND FURTHER RESEARCH
The reflection of this study as a whole proved to be sufficient and credible research study it
showed that the ever increasing field of AI is always changing as new theories and studies on
this is being developed every day. Calculating the emergence of AI regarding Human
intelligence is difficult. The data collected and obtained from papers for primary research
showed enough information to show good and valid results about AI at this early stage.
However, more data from primary research such as self-conducting new primary research
would be the better outcome next time as you can control the sample data and precisely
understand the data as this would offer more a perspectives from a various range of participants
showing more results and also different findings from the papers researched. Although the data
that was used in this study was adequate into this subject as not many people, know about AI
or understand fully what it is at this time.
The research provided in this study from the papers shows that there are more investigate on
the subject matter of AI suggested a more in depth into how AI works in companies. At the
moment by researching people in the workplace of AI to understand their viewpoints, more
research into what famous scientists are saying about AI and their perspective into it. The
development of AI with general intelligence remains to be necessary for the AI researchers and
its role within society given that the hardware and software capability are right now
unreachable. But perhaps not for long regarding the future of AI. The study has assessed the
best analysis of AI because of how long AI has been with us. But making incorrect predictions
and results must be cautious. However, this adds to our understanding of the future of AI.
From existing literature in chapter 3, it shows that AI at the human level or beyond would occur
between the middle of this century to the end of this century the survey revealed interesting
information that the disagreement of experts of AGI milestones and the safety of AGI is varied
across the survey. Also, others remain sceptical whether this will ever happen at all. Currently,
AGI experts hold very different views, and standpoint about this and these situations makes
the ability have future accuracy predictions about AGI uncertain. But this is not alarming
because the development of AI has been rather slow in this field. It would be intriguing to carry
out further research in the future to how expert views change over time as the AI advances in
time.
Although current AI has little to none ethical issues which are only present in advanced systems
of cars and power plants, the idea of AI with super intelligence seems unfathomable at the
38
moment because we are at the early stages of AI. The challenges facing AI is that the prediction
scale of AI is unexpected meaning we don’t know when intelligent AI will emerge but from
researching existing literature and prior work, ultimately now it’s not a case of if this will
happen rather it’s the case when this will happen.
39
CHAPTER 6 – REFERENCES
6.1| REFERENCES
(UOC), e. (2013). Research Paradigms:Ontology's, Epistemologies & Methods.
[online]
Slideshare.net.
Available
at:
http://www.slideshare.net/eLearnCenter/research-methods-uoc-2013 [Accessed
11 Feb. 2016].
Aamodt, A. and Plaza, E. (1994). Case-Based Reasoning: Foundational Issues,
Methodological Variations, and System Approaches. AI Communications. IOS
Press, 7(1), p.8.
Alternative Mindsets. (2015). Different Types of Artificial Intelligence. [online]
Available
at:
http://alternativemindsets.co.uk/different-types-artificialintelligence/ [Accessed 15 Feb. 2016].
AUDA, G. and KAMEL, M. (1999). MODULAR NEURAL NETWORKS: A
SURVEY. Int. J. Neur. Syst., 09(02), pp.129-151.
Baum, S., Goertzel, B. and Goertzel, T. (2011). How long until human-level AI?
Results from an expert assessment. Technological Forecasting and Social
Change, 78(1), pp.185-195.
BBC News. (2016). Google’s AI beats world Go champion in first of five matches BBC News. [online] Available at: http://www.bbc.co.uk/news/technology35761246 [Accessed 17 Feb. 2016].
Bianchini, F. (2016). Artificial intelligence and synthetic biology: A tri-temporal
contribution.Biosystems.
Bloch, M., Hirschfeld, L. and Gelman, S. (1995). Mapping the Mind: Domain
Specificity in Cognition and Culture. The Journal of the Royal Anthropological
Institute, 1(2), p.424.
Bostrom, N. (2012). The Superintelligent Will: Motivation and Instrumental
Rationality in Advanced Artificial Agents. Minds and Machines, 22(2), pp.7185.
Bostrom, N. and Shulman, C. (2016). How Hard is Artificial Intelligence?
Evolutionary Arguments and Selection Effects. Journal of Consious Studies,
19(7-8), pp.103-130.
40
Bostrom,
N.
and
Yudkowsky,
E.
(2011).
THE ETHICS OF ARTIFICIAL INTELLIGENCE.Draft for Cambridge Hand
book of Artificial Intelligence, eds, 1(1), p.3.
Brézillon, P. (2011). From expert systems to context-based intelligent assistant
systems: a testimony.The Knowledge Engineering Review, 26(01), pp.19-24.
Chen, S. and Chang, Y. (2011). Weighted fuzzy interpolative reasoning for sparse
fuzzy rule-based systems. Expert Systems with Applications, 38(8), pp.95649572.
Cook, J. (2016). Here's how much smarter Google search results have become using
artificial intelligence. [online] Business Insider. Available at:
http://uk.businessinsider.com/google-search-results-using-artificialintelligence-2016-3 [Accessed 1 Feb. 2016].
Copeland, J. (1993). Symbolic computation. artificial intelligence. Oxford:
Blackwell.
Creswell, J. (2013). Research Design: Qualitative, Quantitative, and Mixed
Methods Approaches.. 4th ed. London: Sage, p.4.
Dan, Q. and Dudeck, J. (1992). Certainty factor theory: Its probabilistic
interpretations and problems.Artificial Intelligence in Medicine, 4(1), pp.21-34.
De Kamps, M. (2012). Towards truly human-level intelligence in artificial
applications. Cognitive Systems Research, 14(1), pp.1-9.
Garrido, A. (2010). Connectionism vs. Computational Theory of Mind. Brain :
Broad Research in Artificial Intelligence and Neuroscience, 1(1), p.1.
Geraci, R. (2010). Apocalyptic AI. Oxford: Oxford University Press.
Goertz, G. and Mahoney, J. (2012). A Tale of Two Cultures: Qualitative and
Quantitative Research in the Social Sciences.. USA: Princeton University Press.
Goertzel, T. (2014). The path to more general artificial intelligence. Journal of
Experimental & Theoretical Artificial Intelligence, 26(3), pp.343-354.
Hardawar, D. (2016). Stephen Hawking: 'The real risk with AI isn't malice but
competence'.
[online]
Engadget.
Available
at:
http://www.engadget.com/2015/10/09/stephen-hawking-ai-reddit-ama/
[Accessed 2 Mar. 2016].
41
Hern, A. (2014). What is the Turing test? And are we all doomed now?. [online] the
Guardian.
Available
at:
https://www.theguardian.com/technology/2014/jun/09/what-is-the-alan-turingtest [Accessed 3 Mar. 2016].
Klien, B. (2007). When will AI Surpass Human-Level Intelligence? AGI-World,.
[online]
AGI-World.
Available
at:
http://www.novamente.net/bruce/?p=54
[Accessed 13 Jan. 2016].
Kothari, C. (2004). Research methodology. New Delhi: New Age International (P)
Ltd.
Kurzweil, R. (2012). How to create a mind. New York: Viking.
Lawrence, D., Palacios-González, C. and Harris, J. (2016). Artificial
Intelligence. Cambridge Quarterly of Healthcare Ethics, 25(2), pp.250-261.
López, B. (2013). Case-Based Reasoning: A Concise Introduction. Synthesis
Lectures on Artificial Intelligence and Machine Learning, 7(1), pp.1-103.
Matyszczyk, C. (2014). Stephen Hawking: Humans evolve slowly, AI could stomp
us out. [online] CNET. Available at: http://www.cnet.com/news/stephenhawking-ai-could-end-humanity/ [Accessed 8 Jan Apr. 2016].
Merriam-webster.com. (2010). Definition of ARTIFICIAL INTELLIGENCE.
[online]
Available
at:
http://www.merriamwebster.com/dictionary/artificial%20intelligence [Accessed 3 Jan. 2016].
Müller, V. (2013). Philosophy and theory of artificial intelligence. Berlin: Springer.
Müller, V. (2014). Risks of general artificial intelligence. Journal of Experimental
& Theoretical Artificial Intelligence, 26(3), pp.297-301.
Narooei, K. and Ramli, R. (2014). New Approaches in Tool Path Optimization of
CNC Machining: A Review. AMM, 663, pp.657-661.
Negnevitsky, M. (2005). Artificial intelligence. Harlow, England: Addison-Wesley.
Neurosolutions.com. (2016). What are Neural Networks & Predictive Data
Analytics?.
[online]
Available
at:
http://www.neurosolutions.com/products/ns/whatisNN.html [Accessed 1 Jan.
2016].
42
Okandan, M. (2010). Can Machines Truly Think?. Sandia National Laboratories,
PO Box 5800 MS1080, Albuquerque,, 1, p.1.
pantic, m. (2016). Introduction to Machine Learning & Case-Based
Reasoning. Computing Department, Imperial College London., 1., 1(1), p.2.
Phys.org. (2016). The future of artificial intelligence. [online] Available at:
http://phys.org/news/2015-07-future-artificial-intelligence.html [Accessed 18
Feb. 2016].
Potapov, A. and Rodionov, S. (2014). Universal empathy and ethical bias for
artificial general intelligence. Journal of Experimental & Theoretical Artificial
Intelligence, 26(3), pp.405-416.
RICHTER, M. and AAMODT, A. (2005). Case-based reasoning foundations. The
Knowledge Engineering Review, 20(03), p.203.
Rhodes, T, Guise, A & Green, J.. (2014). Mixed Method and multidisciplinary
approaches. In: Durand, M & Chantler, T. Principles of Social Research.
England: McGraw-Hill. 163-175
Russell, S. and Norvig, P. (2010). Artificial intelligence. Upper Saddle River:
Prentice-Hall.
Sainato, M. (2015). Stephen Hawking, Elon Musk, and Bill Gates Warn About
Artificial
Intelligence.
[online]
Observer.
Available
at:
http://observer.com/2015/08/stephen-hawking-elon-musk-and-bill-gates-warnabout-artificial-intelligence/ [Accessed 26 Jan. 2016].
Sandberg, A. and Bostrom, N. (2011). Machine intelligence survey Technical
Report. published by the Future of Humainity Institute,Oxford University, 1(1).
Saunders, M., Lewis, P. and Thornhill, A. (2012). Research methods for business
students. 6th ed. 6th ed.
Sheu, H. and Wei, Y. (2011). Effective options trading strategies based on volatility
forecasting recruiting investor sentiment. Expert Systems with Applications,
38(1), pp.585-596.
Slevitch, L. (2011). Qualitative and Quantitative Methodologies Compared:
Ontological and Epistemological Perspectives. Journal of Quality Assurance in
Hospitality & Tourism, 12(1), pp.73-81.
43
Stafford, R. (2010). Constraints of Biological Neural Networks and Their
Consideration in AI Applications. Advances in Artificial Intelligence, 2010,
pp.1-6.
Telegraph.co.uk. (2013). Chinese supercomputer is world's fastest at 33,860 trillion
calculations
per
second.
[online]
Available
at:
http://www.telegraph.co.uk/technology/news/10129285/Chinesesupercomputer-is-worlds-fastest-at-33860-trillion-calculations-per-second.html
[Accessed 18 Feb. 2016].
Wait But Why. (2015). The Artificial Intelligence Revolution: Part 1 - Wait But Why.
[online] Available at: http://waitbutwhy.com/2015/01/artificial-intelligencerevolution-1.html [Accessed 1 Apr. 2016].
Wallace, S., McCartney, R. and Russell, I. (2010). Games and machine learning: a
powerful combination in an artificial intelligence course. Computer Science
Education, 20(1), pp.17-36.
Warwick, K. (2012). Artificial intelligence. New York: Routledge.
Webopedia.com. (2016). What is Moore's Law? Webopedia Definition. [online]
Available
at:
http://www.webopedia.com/TERM/M/Moores_Law.html
[Accessed 25 Mar. 2016].
Wheeler, G. and Pereira, L. (2004). Epistemology and artificial intelligence. Journal
of Applied Logic, 2(4), pp.469-493.
Woods, P. (2006). Successful writing for qualitative researchers. London:
Routledge.
Www-formal.stanford.edu. (2016). Applications of AI. [online] Available at:
http://www-formal.stanford.edu/jmc/whatisai/node3.html [Accessed 1 Apr.
2016].
44
6.2 | BIBLIOGRPAHY
BBC Timelines. (2016). AI: 15 key moments in the story of artificial intelligence.
[online] Available at: http://www.bbc.co.uk/timelines/zq376fr [Accessed 3 Jan.
2016].
Greer, K. (2014). Is Intelligence Artificial?
Goertzel, B., Pitt, J., Ikle, M., Pennachin, C. and Rui, L. (2010). Glocal memory: A
critical design principle for artificial brains and minds. Neurocomputing, 74(13), pp.84-94.
Ichise, y., Arakawa,, N. and Yamakawa., H. (2014). Ryutaro Ichise, Naoya
Arakawa, & Hiroshi Yamakawa. (2014). An Overview of Artificial General
Intelligence as of 2014.
Mckinney, L. (2009). Will AI surpass human intelligence by 2020?. [online] The
Daily Galaxy --Great Discoveries Channel: Sci, Space, Tech. Available at:
http://www.dailygalaxy.com/my_weblog/2009/09/a-singular-something-insixty-years.html [Accessed 3 Jan. 2016].
Metz, C. (2016). For Google, Quantum Computing Is Like Learning to Fly. [online]
WIRED. Available at: http://www.wired.com/2015/12/for-google-quantumcomputing-is-like-learning-to-fly/ [Accessed 3 Mar. 2016].
Metz, C. (2016). The Sadness and Beauty of Watching Google’s AI Play Go. [online]
WIRED. Available at: http://www.wired.com/2016/03/sadness-beautywatching-googles-ai-play-go/ [Accessed 18 Mar. 2016].
Özkural, E. (2014). What Is It Like to Be a Brain Simulation?
Pavlacka, B. (2012). Pavlacka, B. (2012). Artificial General Intelligence and the
Future of the Human Race.
Vardi, M. (2012). Artificial intelligence. Communications of the ACM, 55(1), pp.55.
Ven, A., & Schouten, B. (2010). A Minimum Relative Entropy Principle for AGI.
Proceedings of the Third Conference on Artificial General Intelligence,
Advances in Intelligent Systems Research, 2010, 2010.
Weinbaum, D., & Veitas, V. (2015). Open Ended Intelligence: The individuation of
Intelligent Agents.
45
Yampolskiy, R. and Fox, J. (2012). Safety Engineering for Artificial General
Intelligence. Topoi.
46
CHAPTER 7-APPENDICES
APPENDIX A: ETHICS APPROVAL
CARDIFF METROPOLITAN UNIVERSITY
APPLICATION FOR ETHICS APPROVAL
When undertaking a research or enterprise project, Cardiff Met staff and students are obliged to
complete this form in order that the ethics implications of that project may be considered.
If the project requires ethics approval from an external agency (e,g., NHS), you will not need to
seek additional ethics approval from Cardiff Met. You should however complete Part One of this
form and attach a copy of your ethics letter(s) of approval in order that your School has a record of
the project.
The document Ethics application guidance notes will help you complete this form. It is available
from the Cardiff Met website. The School or Unit in which you are based may also have produced
some guidance documents, please consult your supervisor or School Ethics Coordinator.
Once you have completed the form, sign the declaration and forward to the appropriate person(s) in
your School or Unit.
PLEASE NOTE:
Participant recruitment or data collection MUST NOT commence until ethics approval has
been obtained.
PART ONE
Name of applicant:
Riaz Ali
Supervisor (if student project):
Dr Jason Williams
School / Unit:
CSM
Student number (if applicable):
20044055
Programme enrolled on (if applicable):
Software Engineering BSc (Hons)
Project Title:
Will AI ever surpass Human intelligence?
Expected start date of data collection:
9/02/15
Approximate duration of data collection:
2 months
Funding Body (if applicable):
N/A
Other researcher(s) working on the project:
N/A
Will the study involve NHS patients or staff?
No
Will the study involve taking samples of
human origin from participants?
No
47
Does your project fall entirely within one of the following categories:
Paper based, involving only documents in
Yes
the public domain
Laboratory based, not involving human
No
participants or human tissue samples
Practice based not involving human
No
participants (eg curatorial, practice audit)
Compulsory projects in professional practice No
(eg Initial Teacher Education)
A project for which external approval has
No
been obtained (e.g., NHS)
If you have answered YES to any of these questions, expand on your answer in the non-technical
summary. No further information regarding your project is required.
If you have answered NO to all of these questions, you must complete Part 2 of this form
In no more than 150 words, give a non-technical summary of the project
In this research area topic I’m a going to critically investigate and evaluate where we at in true artificial
intelligence and will we be ever surpassed by our own technology in creating a high level of consciousness
in computing. This is the aim of the research paper. The objectives is to evaluate what are current
technologies and what’s the outcome of this in the future. Another objective is to understand how far
artificial intelligence can go in terms of advancement. And the last objective is to discuss the pros and cons
of this happening in the future and what safe guards are in place if this ever happens.
As mine is based on primarily a theoretical dissertation, it will have all the secondary data research applied
to it. My secondary research will be based on mostly journal papers and books from 2011 onwards as this
has the relevant data for my dissertation. Research I will be using is typically, the items will be written or
produced on paper, such as newspaper articles, Government policy records, leaflets and minutes of
meetings. Items in other media can also be the subject of documentary analysis, including films, songs,
websites and photographs.
DECLARATION:
I confirm that this project conforms with the Cardiff Met Research Governance Framework
I confirm that I will abide by the Cardiff Met requirements regarding confidentiality and
anonymity when conducting this project.
STUDENTS: I confirm that I will not disseminate any material produced as a result of this project
without the prior approval of my supervisor.
Signature of the applicant:
Date: 8th Jan 2015
Riaz Ali
FOR STUDENT PROJECTS ONLY
Name of supervisor: Dr Jason Williams
Date: 25th Jan 2015
Signature of supervisor:
48
Research Ethics Committee use only
Decision reached:
Project approved
Project approved in principle
Decision deferred
Project not approved
Project rejected
Project reference number:
Name:
Date:
Signature:
Details of any conditions upon which approval is dependant:
PART TWO
A RESEARCH DESIGN
A1 Will you be using an approved protocol in your
project?
A2 If yes, please state the name and code of the approved protocol to be used1
A3 Describe the research design to be used in your project
A4 Will the project involve deceptive or covert
research?
A5 If yes, give a rationale for the use of deceptive or covert research N/A
A6 Will the project have security sensitive implications?
A7 If yes, please explain what they are and the measures that are proposed to address
them
B PREVIOUS EXPERIENCE
B1 What previous experience of research involving human participants relevant to this
project do you have?
1
An Approved Protocol is one which has been approved by Cardiff Met to be used under supervision of
designated members of staff; a list of approved protocols can be found on the Cardiff Met website here
49
B2 Student project only
What previous experience of research involving human participants relevant to this
project does your supervisor have?
C POTENTIAL RISKS
C1 What potential risks do you foresee?
C2 How will you deal with the potential risks?
Application for ethics approval v4 March 2015
ETHICS NUMBER: 2015D0400
50