Artificial intelligence is a latest buzzword in the technology industry. Suddenly all the startups and companies who operate in technology field are claiming to use artificial intelligence or machine learning in one way or the other to solve their customer's problems. But like all other buzzwords, is artificial intelligence also just a hype cycle which will fizzle. Artificial intelligence specifically is notoriously famous for reaching a crescendo of hype every decade and then going through a period of non activity and non investment which is known as the AI winter.
So first of all let's try to understand what exactly is artificial intelligence. Artificial intelligence or AI, simply put, allows machines or software to learn from its past experiences so that it can adjust dynamically according to the new inputs and produce accurate outputs. This flexibility allows artificial intelligent software to perform tasks which are not possible by hard coded programs. The computer can be trained to perform specific tasks, once it is allowed to ingest large amounts of data from which it can derive different patterns.
Artificial intelligence is a very broad field and consists of various different approaches, technologies and theories. The field of AI have multiple subfields such as machine learning, neural networks, deep learning, computer vision, cognitive computing, natural language processing et cetera. Let's take a closer look at each one of them.
Machine learning is nothing but a subfield of artificial intelligence and this is the core statistical technique in artificial intelligence which gives computers the superpower to consistently improve their performance on a specific given task based on the data provided. Machine learning reduces the dependence of software on static hardcoded instructions and allows it to make models from the sample data and based on that model make predictions about the decisions. Machine learning is increasingly being applied in areas where it is unfeasible to write if-then kind of hard coded rules for example in computer vision, spam detection, cyber security, fraud detection et cetera. Machine learning can both be supervised and unsupervised. Machine learning simply means that a software is able to learn from the experience if it is subjected to perform certain task and it's performance is measured and fed back into the system as a reinforcement.
Neural networks are computing systems which resemble organic animal brains. These systems are able to figure out patterns and learn from examples without being fed any specific instructions. An artificial neural network resembles a human neuron and they transmit signals to each other like how organic brains work through synapses. Artificial neurons have something called weight which helps to dynamically adjust the learning at every layer of network. The input travels from first layer to the end layer and every layer helps to build an abstraction based on the data which is being fed. For example in artificial neural network can identify different species of animals based on the image data which is labelled. Once an artificial neural network has been trained then it can be extrapolated to identify animals in other images without them being labelled.
You must have used voice assistant such as Siri, Cortana, Alexa you would have realised how efficient they have become in understanding your voice instructions. This technology which allows computers to understand our natural language is called natural language processing and it is a very important field of artificial intelligence. There are multiple domains within the broader field of Natural Language Processing such as natural language understanding, natural language generation and speech recognition.
Machine learning has really advanced the natural language processing field as earlier it used to depend upon direct hardcoded rules which is not very optimum since the variation in natural language is way too high. This major advancement in natural language processing has come about in last few years only and has led to huge commercial boom in smart assistants and smart speakers such as Google home and Amazon Alexa. It has really opened up voice as another modality for human computer interface which is highly intuitive and requires the very low learning curve, thus making computing even more ubiquitous.
Self Driving Car
Computer vision is a technology which is a subfield of artificial intelligence as well as a highly interdisciplinary field which seeks to build computer systems which are capable of doing task which human visual systems can do. Computer vision involves building models through digital images and videos by acquiring, processing and analysing those digital images. This technology gives computers an intricate and textured understanding of the real world and it is applicable in some breakthrough technological inventions such as self driving cars, self serve autonomous retail stores, augmented reality etc. There are multiple subdomains of computer vision such as object recognition, video tracking, event detection, distance measurement, motion estimation, image reconstruction etc.
The artificial intelligence was coined in the year 1956, and since then the technology has evolved and matured a lot due to multiple advancements in different overlapping fields such as decision tree algorithms, neural networks, computing power which allows large processing of large amounts of data at low cost and high speed.
Artificial intelligence is the intelligent behaviour is shown by the machine and this is in contrast to the natural intelligence which is shown by organic life forms such as humans and other animals. Intelligence is nothing but when an agent is able to perceive its environment and is able to take the appropriate actions which increases the probability of it achieving its goals. These agents are called intelligent agents and the study of their behaviour is called AI research. Higher forms of artificial intelligence involves a machine which begins to mimic the cognitive functions of human mind such as learning and problem solving.
The field of AI research draws upon multiple academic fields such as linguistics, mathematics, psychology, computer science et cetera. Central problems that AI research intends to solve deal with perception, natural Language Processing, learning, planning, knowledge representation etc. These are some really hard problems from a technical perspective and requires approaching the problem from multiple perspective and Fields.
Us department of Defence and specifically defence advanced research projects agency or DARPA started multiple projects to train computers to make basic human reasoning in the early 1960s. This initial foundation of work in the automation and formal reasoning led to the technologies that we see every day in a life such as decision support systems and smart search systems which augment the human abilities.
Some people such as late scientist Stephen Hawkings and billionaire entrepreneur Elon Musk have often talked about how AI can be a threat to humanity. This notion of artificial intelligence becoming a threat to human civilisation is a topic which has been taken up a lot of times in popular media movies like The Terminator and others Si-Fi movies.
Artificial intelligence is seeing another resurgence with multiple advances in different technologies such as availability of computing power which gives us ability to process large amounts of data cheaply and in a very short time, advancements in neural networks and also the availability of huge amount of data which is required to train artificial intelligence systems. Lot of people who are sceptics of AI talk about how AI will create mass unemployment. But this fear is far fetched according to experts and this sort of hysteria has been associated even with the previous technological revolutions like Industrial Revolution et cetera.
In the 1980s there was a huge interest in the field of artificial intelligence after some of the expert systems, a program that simulated the knowledge and analytical skills of the human experts, began to gain some traction in the business landscape. These expert systems were really good at emulating the decision making abilities of the human experts by synthesising the processing power of computers with the in depth knowledge in a very specific domain. Most of the logic of these expert systems was written like if - then rules and it was mostly hard coded. Some of the use cases of these expert system included diagnosis of specific diseases, unknown organic molecules. These expert systems were solving some real world business problems and hence were one of the first truly successful Artificial Intelligence Software systems. These expert systems were mainly developed in lisp programming environment as well as prolog.
One of the major problem associated with the expert systems was the knowledge acquisition problem along with that there was a performance problem. These systems were written in interpreted languages and couldn't match the efficiency of fast compiled languages such as C. Also the rigidity of if-then rules restricted the ability of these expert systems to solve problems which are off the standard template, thus limiting the utility significantly.
LISP Symbolic Machine. One of the earliest expert system.
The application of artificial intelligence have led to some of the most important technological breakthroughs such as digital voice assistant like Siri, Cortana ,Alexa, Self driving cars, autonomous drones. Artificial intelligence is increasingly being used in different industries such as Healthcare where AI based applications are providing personalised diagnosis of diseases, reading X rays and medical reports as well as reminding patients to take pills on time. AI in retail is increasingly being used to provide recommendations to the customers based on their previous purchase history and his customer profile. Retail industry is leveraging AI to make smarter predictions. In gaming artificial intelligence is used to make virtual worlds increasingly realistic through advanced interactive experiences. AI is also used in factories which stream IoT data to provide real time recommendation related to saving energy and managing processes autonomously without human input.
One of the most useful application of artificial intelligence is in the medical and Healthcare industry. AI can be used to help doctors to find treatments of cancer for patients and also help in the process of diagnosis by analysing data of millions of medical reports of previous patients. For example there are more than thousand plus medicines and vaccines to treat cancer and it is almost humanly impossible for doctors to figure out the best permutation and combination of medicine applicable to the condition of patient. So AI can analyse both the symptoms, the stage of cancer and based on the previous historical data it can recommend the best possible medicine and treatment route. Artificial Intelligence has significantly performed better than human doctors in the process of identifying specialised types of cancer such a skin cancer.
One of the major use cases where AI is being applied in today's time is in the transportation industry in the creation and development of self driving autonomous vehicles. Some of the companies which are using AI to create autonomous vehicles are Tesla, Google's Waymo, Apple, Uber, Cruise by General Motors et cetera. There are multiple processes involved in building the driverless car and AI has to perform all the tasks simultaneously which are involved in driving a car such as breaking, collision prevention, mapping, navigation, lane changing cetera. Governments in multiple states have relaxed the regulation for technology companies to test out their self driving vehicles in United States,China and Israel.
Artificial intelligence is increasingly being used in financial industry in the field of personal finance, credit scoring, portfolio management through Robo Advisors, algorithmic trading which involves using AI systems to make split second decisions to do millions of trades in a short period of time. Is also used by banks and payment processing software companies to detect and flag potential frauds by analysing behavioral patterns and detecting any abnormal changes or anomalies. Traditional financial industries such as private equity and venture capital industries are also being disrupted by AI where AI systems are analysing data of millions of startups/companies and investing autonomously.
Security breaches and hacking has become a serious threat to companies, government institutions and individuals. Privacy and data protection of data has become one of the most important mandate for 21st century. Artificial intelligence has very strong applicability in cyber security industry. For example companies are using machine learning to understand behaviour analytics of structured and unstructured data to model network behaviour and to improve security threat detection in real time. Cyber security companies are also using artificial intelligence to stitch together a cohesive incident detection narrative by curating, correlating and enriching high volume security alerts and incidents. This highly relevant incident detection system can then be used for prioritisation and incident response protocol improvement. AI is also used by companies to mine large amount of software vulnerabilities data, configuration errors and patches to figure out incidents which require immediate attention and to mobilize the right teams responsible for it.
Artificial intelligence has led to increase annual military spending. Countries are building artificial intelligence to control a fleet of unmanned military drones which are capable of autonomous actions and which can be used in various defence situations.
Ad Tech companies are using artificial Intelligence and machine learning to optimise the conversion of the advertising networks. Also companies are using AI to predict the behaviour of customers so that they can target them in a more efficient manner through personalized promotions. Companies in e-commerce and travel industry are using multivariate testing to increase the ROI of their search engine marketing spend.
Some of the challenges of today's artificial intelligent systems is that they are currently able to do very specific tasks. AI learns from the data it is fed and the quality of data will determine what kind of output the system is able to produce. So these systems for able to do one task extremely well but when it is told to do some other task it fails miserably. For example a system which is designed to drive self driving car cannot detect bank fraud or a system which can detect fractures by observing X-rays cannot do natural language processing. To simply put these systems are currently built in a very specialised manner and focus on a single task and are extremely far from what we call general purpose intelligence or how the humans behave. So the judgement day from the movie Terminator is way far of into the future.