Intelligent

Systems

Intelligent Systems

Foreword

Artificial Intelligence is at its peak hype right now. According to Lynsey Barber, who has been named Technology Journalist of the Year at the UK Tech Awards, “deep learning and machine learning are at what it calls the 'peak of inflated expectation', but are just two to five years away from mainstream adoption”[1]. If everything you hear about latest AI advancement leaves you slightly confused, do not worry - you are not alone. The term “artificial intelligence”, much like many other buzz words, has been adopted widely by marketers and advertising copyrighters. We have asked a few experts to help us actually define what intelligent systems are, and we got slightly different answers from everyone. This led us to believe that it is extremely important for people to start learning about the methods and prospects of intelligent systems and what it means for our future.

Most of the researches we have communicated with and spoke tended to think that the latest advancements of AI are strictly connected to deepening our understanding of intelligence in general, more specifically human intelligence. Therefore, we define artificial intelligence with respect to its ability to replicate the way human beings think.

Sub-areas

Artificial intelligence is a vast and wide-ranging field with many facets and areas of focus, including:

Robotics
Robotics is the most famous and widely implemented sub area of artificial intelligence. The applications of intelligent robotics are extremely vast such as in military, medicine and agriculture to name a few. The main aim of robotics is to develop machines that can substitute humans to perform a given task. At this point researchers are nowhere close to making robots that can substitute a human brain completely through artificial intelligence but they are still able to make many tasks much easier to perform without human intervention.

Natural Language Processing
Ever wondered how Siri or Cortana understand and react to your speech fairly accurately? These make use of very sophisticated Natural Language Processing techniques to give such a result. NLPs make use of artificial intelligence to understand and interpret human language. They are used in many fields and are extremely useful for visually impaired people[2].

Expert Systems
Expert systems make use of artificial intelligence to become proficient in a particular domain and provide solutions like a human expert in that field would. It is an intelligent computer program that makes use of a knowledge base which has facts and data collected from various credible sources and an inference engine which makes decisions and provides an answer. An expert system is a branch of artificial intelligence introduced by researchers in the Stanford Heuristic Programming Project.

[Machine learning is the] field of study that gives computers the ability to learn without being explicitly programmed.

——   Arthur Samuel[1]   1959

1920 —

Robot beginnings

Czech author Karel Capek coined the term robot, the first instance of which was in his play Rossum's Universal Robots.

Still from Universal Robots  —  University of Michigan

1940 —

Codes of Conduct

Isaav Asimov devises the 3 laws of robotics[2]:

1) "A robot may not injure a human being or, through inaction, allow a human being to come to harm."

2) "A robot must obey the orders given it by human beings except where such orders would conflict with the First Law."

3) “A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.”

Alan Turing aged 16  —  Wikimedia Commons

1950 —

The Turing Test

Alan Turing, one of the pioneers of computer science, developed a procedure to gauge a machine's intelligence. His concept, called The Imitation Game (now better known as the Turing Test), was proposed in his paper Computing Machinery and Intelligence. In the Turing test, a human evaluator is made to have a natural language text-based conversation between another human and a computer that is designed to respond like a human. The evaluator is aware that one of their conversation partners is a computer, but is not explicitly told which one. The computer passes the Turing test if the evaluator cannot confidently discern (with over 50% accuracy) between the human and the computer conversation partner. While considered not the most definitive way of measuring machine intelligence, the Turing test was still one of the pioneering ideas that helped create conversation around intelligent systems and machines that can "think."

WIRED UK
The Turing test, and how to pass it  →

1950 —

The Turing Test

Alan Turing, one of the pioneers of computer science, developed a procedure to gauge a machine's intelligence. His concept, called The Imitation Game (now better known as the Turing Test), was proposed in his paper Computing Machinery and Intelligence. In the Turing test, a human evaluator is made to have a natural language text-based conversation between another human and a computer that is designed to respond like a human. The evaluator is aware that one of their conversation partners is a computer, but is not explicitly told which one. The computer passes the Turing test if the evaluator cannot confidently discern (with over 50% accuracy) between the human and the computer conversation partner. While considered not the most definitive way of measuring machine intelligence, the Turing test was still one of the pioneering ideas that helped create conversation around intelligent systems and machines that can "think."

WIRED UK
The Turing test, and how to pass it  →

Alan Turing aged 16  —  Wikimedia Commons

1951 —

Snarky Systems

Marvin Minsky built the first neural network, called the Stochastic Neural Analog Reinforcement Calculator (SNARC), which is regarded as one of the principal developments in the field of intelligent systems. The structure of a neural network tries to mimic that of the brain, using silicon "neurons" that receive and process information.

1956 —

The Birth of AI

It wasn't until 1956 at the Dartmouth Conference, lead by John McCarthy and Marvin Minsky, that artificial intelligence, both the term and the field, was officially introduced and established. Here is a section of their project proposal[3]:

The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves.

——   John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon,   1955

1997 —

Checkmate

The Deep Blue supercomputer, developed by IBM, defeated renowned chess champion Garry Kasparov. Although Deep Blue depended mainly on brute force techniques, the win was seen as one of the groundbreaking moments that put AI in the spotlight.

NYU
How Intelligent is Deep Blue?  →

WIRED
What Deep Blue Tells Us About AI in 2017  →

Gary Kasparov losing to Deep Blue  —  Quartz

2009 —

A Driving Force

Google debuts its self-driving car project, now called Waymo, which was headed by Sebastian Thrun. The project aims[4] to make fully autonomous cars that help increase road safety and to make getting around simpler and more seamless. Currently, Waymo uses deep learning technology to "teach" cars how to steer clear of obstacles as well as to anticipate the actions of other motorists and pedestrians on the road in order to avoid accidents.

THE VERGE
Waymo is first to put fully self-driving cars on US roads without a safety driver  →

2011 —

Who is Watson?

IBM's Watson system, which was conceived in 2006, competed in American game show Jeopardy! and beat two former winners of the show, Brad Rutter and Ken Jennings. IBM has opened up the Watson platform to businesses and corporations so that its artifical intelligence technology can be used throughout various industries, including medical, science, and even culinary. Currently, Watson is being used by a slew of companies in a slew of industries: for example, Wimbledon utilises Watson to automatically create video reels of tennis matches[5].

2011 —

Hey Siri

Apple unveils Siri, which was one of the very first mobile virtual assistants that brought together areas of intelligent systems, like natural language processing and speech recognition with machine learning, and rolled it all in one package. Over the years, Apple has honed Siri, making the technology faster more accurate, as well as letting third-parties integrate with the assistant. Siri helped usher in a new era of digital assistants, with Google releasing Google Now and, subsequently Google Assistant, Amazon's Echo, and Microsoft's Cortana.

2011 —

Google's Brain

Around the same time, Google was also helping jump-start the adoption of speech-recognition by incorporating it into a number of its core products, including Search. Google started its Google Brain project, which focuses on researching numerous techniques for machine learning, including image recognition, data analysis and voice recognition, and making their findings available to Google services such as Search, Maps, and Android.

2016 —

DeepMind

The AlphaGo system created by DeepMind (which Google acquired in 2014[6]) beats Go world champion Lee Sedol in a breakthrough event for artificially intelligent systems. The game of Go is deemed one of the most, if not the most complex board games especially for a computer system to crack. AlphaGo uses a slew of artificial intelligence techniques like neural networks and deep learning. The system used a combination of supervised learning (in which it is given data to learn from) and reinforcement learning (in which the system plays against itself) to help it progress and eventually defeat Lee.

2017 —

AlphaGo Zero

DeepMind recently published a paper on the journal Nature, detailing how its new system, AlphaGo Zero, was able to defeat the original AlphaGo 100 games to none[7]. AlphaGo Zero is based solely on reinforcement learning techniques, essentially meaning that it learned the game entirely from scratch, without any data about human strategies or previous gameplays. This research was groundbreaking and helped cement the power of reinforcement learning.

DEEPMIND
AlphaGo Zero: Learning from scratch  →

QUANTA
Artificial Intelligence Learns to Learn  →

Video by DeepMind explaining AlphaGo's unprecedented feat  —  Youtube

Q

&

A

Q

&

A

Dr Sunny Bains  —  UCL

We sat down with UCL Senior Teaching Fellow Dr. Sunny Bains to ask her about some of her previous work in the artificial intelligence industry, back when many of the concepts and techniques that we take as commonplace today were only just coming into fruition.

Dr. Bains had been interested in AI since she was a child. She fell in love with the science behind holograms and optics, which lead her to study physics. But she later discovered that many of the innovations behind optical computing, back in the late 80s, were related to machine intelligence.

Dr Sunny Bains  —  UCL

We sat down with UCL Senior Teaching Fellow Dr. Sunny Bains to ask her about some of her previous work in the artificial intelligence industry, back when many of the concepts and techniques that we take as commonplace today were only just coming into fruition.

Dr. Bains had been interested in AI since she was a child. She fell in love with the science behind holograms and optics, which lead her to study physics. But she later discovered that many of the innovations behind optical computing, back in the late 80s, were related to machine intelligence.

I've spoken in my lectures about understanding your competitors. In order to understand the competitors to these optical technologies, I needed to start to understand a little bit more about the non-optical technologies that were emerging at that time, and so that got me interested [in artificial intelligence].

——   Dr Sunny Bains  on machine learning

While AI might appear to be a predominantly digital phenomenon, Bains was more interested in the analogue side of it. In 1997, she wrote her PhD thesis on "Physical computation and embodied artificial intelligence," in which she focused on ways to compact the power of artificial intelligent systems into a "human-sized package."

The question to me was, how could you do stuff in real time in a machine that was either not connected or not that tightly connected to bigger networks; and literally trying to build brains - can you build something that could do what a human could do in that kind of space and with that kind of power?

——   Dr Sunny Bains  on small-scale AI

Currently, Bains is no longer in the AI industry, after having moved on to other fields including philosophy and journalism. However, she still looks back fondly at her time in the field, and continues to follow AI closely.

It was awesome, I've had an awesome life, I have no regrets.

——   Dr Sunny Bains  on her AI research

When asked what developments in AI she was most excited about, Bains described the brilliance of self-driving cars, most especially Google's work in the domain and Uber's traction in the industry.

You probably don't remember life before Uber...That to me is amazing, and of course the next step are Google [self-driving] cars. I've lived in London, in Boston, near Seattle, in San Francisco, the Bay Area, and there's not one of those places that would not be a thousand times better if you got rid of most of the cars.

——   Dr Sunny Bains  on the future

Dr John Shawe-Taylor  —  UCL

Dr John Shawe-Taylor  —  UCL

Dr. John Shawe-Taylor is the Head of the Computer Science Department at the University College London. He is also professor of Computational Statistics and Machine Learning, and has a lot of experience in the field of artificial intelligence, having witnessed and worked on some key projects in AI's intricate timeline.

He graciously gave us some of his time to talk about his research in the field of artificial intelligence, where he sees the industry going, and what he identifies as some of the main obstacles that researchers are facing.


Listen below:

Dr. John Shawe-Taylor is the Head of the Computer Science Department at the University College London. He is also professor of Computational Statistics and Machine Learning, and has a lot of experience in the field of artificial intelligence, having witnessed and worked on some key projects in AI's intricate timeline.

He graciously gave us some of his time to talk about his research in the field of artificial intelligence, where he sees the industry going, and what he identifies as some of the main obstacles that researchers are facing.

Listen here  →

Dr Peter Bentley  —  UCL

Dr Peter Bentley  —  UCL

In addition to being a Teaching Fellow at the University College London, Dr. Peter Bentley is also the Chief Technology Officer of Braintree, a company that utilises artificial intelligence to provide services across a variety of industries and clients. We were able to spend some time with Dr. Bentley to learn a bit more about Braintree and the company's vision, as well as ask about his views on the current state of AI and how it is portrayed in the media.


Listen here  →

In addition to being a Teaching Fellow at the University College London, Dr. Peter Bentley is also the Chief Technology Officer of Braintree, a company that utilises artificial intelligence to provide services across a variety of industries and clients. We were able to spend some time with Dr. Bentley to learn a bit more about Braintree and the company's vision, as well as ask about his views on the current state of AI and how it is portrayed in the media.


Listen below:

The artificial intelligence industry is progressing at a rapid rate, with new technologies being introduced and new papers being published every day.

Here's a look at some of the recent developments in AI...

The artificial intelligence industry is progressing at a rapid rate, with new technologies being introduced and new papers being published every day.

Here's a look at some of the recent developments in AI...

UCL RobERt

Looking At The Stars - Or Exoplanets

In 2016, researchers at UCL, led by Dr. Ingo Waldmann, unveiled the Robotic Exoplanet Recognition system (called RobERt), which uses machine learning to identify the chemical composition of the atmospheres of exoplanets. RobERt is built on a deep-belief nerual network[1], and learns by being given sample data of atmospheric compositions. Typically, it takes astronomers many days to anaylse one exoplanet's atmosphere, especially because of the immense volume of data being gathered. RobERt would significantly speed up this process and help make finding the next habitable planet much easier.
Dr. Waldmann's paper can be found on the American Astronomical Society Astronomy Abstract Service[2]

POPULAR SCIENCE
How scientists will use artificial intelligence to find aliens  →

Backpropagation

The Silicon Brain

Geoffrey Hinton is one of the principal researchers in artificial intelligence. His most notable work involved utilising backpropagation to train neural networks. Neural networks learn by comparing the output it produces to the desired correct output (specified by the human supervisor), and then adapting its "neural connections" accordingly. Such an example could be in attempting to recognise images of cats - if a neural network is shown an image of a cat, and it incorrectly states (outputs) that it is not a cat, this error will be used to update the connections in the network so that it can more accurately identify images of cats.

WIRED
Google's AI Wizard Unveils A New Twist On Neural Networks  →

Geoffrey Hinton at Google campus in Mountain View, California  —  Toronto Star

Generative Adversarial Networks (GANs)

Machine versus Machine

A neural network is a system that is built to model the structure of the human brain and the networks of neurons that make up the nervous system. A Generative Adversarial Network (GAN) consists of two neural networks that are made to compete against each other: a generator and a discriminator[3]. The generator is given a set of randomised "noisy" input, and from it must produce a piece of data, such as an image of a dog. This data is subsequently passed into the discriminator, along with real data not created synthetically by the generator. The role of the discriminator is to determine whether or not the data it is receiving was produced by the generator. The goal of the generator is to develop data that is realistic enough to fool the discriminator, whilst the goal of the discriminator is to be smart enough to recognise the difference between the real and generated pieces of data. This back-and-forth feedback loop between the two networks is an effective method of training both networks.
The concept of GANs was conceived by researcher Ian Goodfellow and a number of partners. Their paper, published in 2014, can be found on arXiv.org[4].

Ok, obviously there's loads of future prospects for AI...

But here are a few examples we've chosen to show what's to come in the field of Intelligent Systems.

Ok, obviously there's loads of future prospects for AI...

But here are a few examples we've chosen to show what's to come in the field of Intelligent Systems.

Self-driving cars

AI is set to rule the roads in the near future. With completely autonomous cars showing up as early as 2021, humans will soon have their own personal drivers available to them 24/7[1].

Waymo self-driving car  —  Documentary Tube

Waymo self-driving car  —  Documentary Tube

Unmanned aerial vehicles

From military drones to pilotless jetliners, AI will take complete control of the skies. All aircraft currently use some form of automation to ease the workload of the pilot. The next step is to make the aircraft completely autonomous.

Intelligent houses

The dream of a smart home has been around for a long time, and is recently becoming more and more of a reality. With the gradual rise of standardised platforms and artificial intelligence, the vision of an interconnected home, in which your lights, temperature control systems, entertainment hubs and kitchen appliciances, can communicate with one another and provide you with a seamless automated experience is now closer than ever[2].