Articles’ “we need huge amounts of annotated data”.

Articles’ summary


Artificial intelligence (AI) can be defined as a series of software technologies that work differently from

the way that a normal human being’s nervous system is used for the actions of sensing, learning, reasoning

and for taking action. The development of AI has been inconsistent, but there has been considerable

progress in the last 60 years. For example, present day video games are programmed by using computer

vision and AI planning. Deep learning is a type of machine learning built on layered demonstration of

variables which are called neural networks. This is used in the speech recognition feature in modern

phones and in various kitchen and home appliances. Such algorithms have a wide variety of applications

which depend upon recognition of patterns.

These technologies are adapted to suit various tasks such as: –

1. Transportation: – Self-driving cars.

2. Home service robots: – Taking care of cleaning, security and delivery jobs.

3. Healthcare: – Assisting as medical staff and caring for patients and elderly.

4. Education: – Teaching kids and military personnel, helping children with learning disabilities.

5. Public safety and security: – Will support in law enforcement.

6. Low resource communities: – Machine learning can assist in dealing with health risks.

7. Employment and work place: – AI can create jobs and make services more affordable.

8. Entertainment: – AI will maintain the data about the customers in the media companies.

Deep learning has permitted numerous technological companies and industries to meet business and

consumer needs in the present-day society. Researchers have decided that using deep-learning models to

process vast amounts of data has benefitted artificial intelligence by amounts unimagined before.

Although deep learning is a very efficient way to enhance AI, it is a piece of technology that is not very

advanced. But experts and scholars like Mohak Shah thinks that “acceleration (in deep learning) is going to

come when we put all the pieces together.” However, Adam Coates, director of Baidu Research Silicon

Valley AI Lab, thinks that to advance deep learning “we need huge amounts of annotated data”. To acquire

this for a feature that uses deep learning, for example the speech recognition feature, he thinks “We don’t

just need the audio for the speech, we need a human to give us a transcription. That can be very expensive.”

But after all, this system of accelerated and more enhanced machine learning can be carried out

unsupervised which is a big advantage over normal machine learning. But there are of course limitations of

this new type of learning. Li Deng, chief AI officer, Citadel explores some off these limits. The “big problem”

according to her, is that “how to have an agent, an intelligent agent, who is able to converse with a human.

That is the kind of problem that not just big data can solve and not just the deep depth of the neural network

can solve.” This is a problem that cannot be solved by even the most cutting-edge technology, deep

learning. Another expert focuses on a way of learning invented upon the principles of deep learning, deep

reinforcement learning. Gary Bradski, chief technology officer, Arraiy, explains deep reinforcement

learning: – “If you’re doing a maze, you went left-right- left-right- left-right, found food. That’s when you get

your signal, “Hey, I got food.” You have to remember and propagate that back in time so that you know,

“Oh, I should take a left here and a right here.” So that’s what reinforcement learning is—it’s learning from

these end rewards, and being able to assign that signal, that learning signal, back through that trace in time.”


Some of the most important applications of quantum computing will be in the realm of complex logistics

and optimization problems-including everything from scheduling, to cybersecurity, supply chain

management and financial services. Quantum computers can simplify problems that are typically too

complex, or have too many variables, for classical computers to solve.

Quantum computers will help investment professionals and financial institutions optimize returns while

minimizing risk by analysing an unimaginable range of scenarios. Quantum computers could also help


businesses with massive supply chains to optimize their processes by more efficiently organizing logistical

variables. Enhanced processing power will also help cybersecurity companies detect threats earlier and

respond more quickly.

Quantum computers also have the ability to develop Artificial Intelligence (AI) and machine learning

systems with higher efficacy and efficiency than conventional computers. Optimized computing abilities

will drive exponential improvements in Artificial Intelligence and machine learning. D-Wave Systems has

already built quantum hardware for Google’s researchers studying machine learning powered image

recognition. This technology could dramatically improve the performance of many technologies, such as

autonomous vehicles.


Alan Turing was one of the greatest mathematicians and scientists in British history. He was one of the

fathers of computer science and he helped defeat Germany in World War II by breaking the German codes.

In 1936 Turing did some of the basic work on today’s computers. Machines were then becoming

increasingly important in British industrial life, such as those in mass production factories or calculators or

typewriters. But they were operating separately from each other. Turing said that it would be possible (in

today’s language) a system to “programme” machines so that one machine (a computer) could do a variety

of different tasks, such as write documents, calculate numbers or play games. This was the “Universal

Turing machine” because one machine could do a variety of tasks. He was describing a computer well

before the technology of his day could build one. He also wrote about artificial intelligence-the way that

computers could ‘think’. He was also faced with the question, “Would it be possible to create a computer

that could be as smart as humans?”. He predicted that such a machine could exist by the year 2000. His

‘Turing Test’ will assess when a machine has become as smart as a human. A human being is in one room

and communicates with a machine in another room, such as by playing chess or asking questions. If the

human cannot from the responses whether the other room contains a human or a machine, then the

machine is as smart as a human. However, Turing was too optimistic about the rate of computer progress.

The world’s best chess player is now a computer (“Big Blue” created by IBM) but in many other respects

humans are still smarter than computers. The ‘Matrix’ movies are based on the era when the computers

are much closer to Turing’s prediction of closing the gap.

Artificial Neural Networks (ANNs) were invented by keeping a real-life example in mind: – the biological

human neural network. ANNs are normally drawn on paper as separate systems, much like the human

biological neural network, which are interconnected as neurons and can send messages between each

system. Actually, ANNs contain layers of neurons- a neuron, in computational terms, is a computational

unit which calculates some piece of information based on weighted input parameters. The bonds between

each system have weights as the variable ‘W’ that changes according to the experience, creating in the

process, neural nets that are adapted to numerous inputs and can continue learning.

Each layer or system of neurons detects some additional information that forms the analysis of whole

picture, for example one layer of neurons detects edges of things in a picture (e.g. recognise tumours in a

brain scan). This technology will help many professions such as doctors, to analyse scans to recognise

problems at an early stage. There can be multiple layers of neurons to detect extra additional information

about the input parameters. Recurrent neural networks were invented based upon this technology and

they create a reverse input loop to assist in certain cases, for example in sentence recognition- because the

meaning of a word is dependent on the prior word or set of words. This helps the computational device to

analyse the situation the word or set of words is placed in and decide on its right meaning to use for the

sentence recognition.