What Is Artificial Intelligence and How Does It Work?

"I think that AI will lead to a low cost and better quality life for millions of people. Like electricity, it’s a possibility to build a wonderful society."

                                                                           - Andrew Ng, Co-founder and lead of Google Brain

In the year 1950, Alan Turing, in his landmark paper, “Computing Machinery and Intelligence”, speculated the possibility of developing machines that could think [1]. The very popular Turing Test followed this work. It was the first serious proposal towards the philosophy of artificial intelligence.

Shortly after this, John McCarthy in 1956, coined the term “Artificial Intelligence”, which he would define as “the science and engineering of making intelligent machines” [2]. Hence AI was brought to public attention as the branch of computer science that strives to enhance a computer’s ability to imitate human intelligence [3]. To put it more simply, AI deals with building systems capable of performing tasks that are generally attributed to human intelligence.

How Artificial Intelligence (AI) Works?

Artificial Intelligence involves the training of computing machines to act in a way that a human does. This article will help you understand various aspects of the logical design and utilisation of Artificial Intelligence.

Machine Learning

Traditionally, computers are programmed to perform specific tasks, but with the advent of machine learning (ML), computers can now be equipped to learn and improve from such task experiences. ML is concerned with developing computer programs that can access data and use it to learn on their own. Tom Mitchell formally describes Machine Learning as:

A computer program is said to learn from Experience E with respect to some class of Tasks T and Performance Measure P, if its performance in a task inT, as measured by P, improves with experience E [4].

Deep Learning

Machine learning as a subset of artificial intelligence has further subtypes. One such subtype of machine learning is known as Deep Learning. It runs on algorithms that use Artificial Neural Networks [5]. These are computing systems inspired by the human brain’s logical functioning. A neural network hence forms the fundamental component of such deep learning algorithms. Deep Learning allows us to use a hierarchical system of artificial neural networks to perform machine learning. This structure enables a computer to process data in a non-linear fashion instead of the linear approach followed in standard machine learning algorithms. Deep learning allows us to use newer methods of training for bigger neural networks.

Evolution of artificial intelligence

 

Neural Networks

The human brain is made up of a network of billions of neurons that store information. Neural networks are multilayer networks of neurons inspired by the functioning of the human brain. These deploy a network of functions that convert a data input of one form into the desired output, generally in another form [5]. Most artificial intelligence techniques require customisations to the algorithm to handle different kinds of information. But, neural networks are versatile because they don’t need to be explicitly programmed for inputs like images, videos, files, and more. Their fundamental goal is to identify the underlying pattern and relationships between data points in the same way as a human brain does [6].

Natural Language Processing

In the field of linguistics, natural language refers to any language that has evolved naturally through continuous use by humans and by repetition. The processing of this natural language gives rise to a separate field of artificial intelligence. Natural Language Processing (NLP) is concerned with programming computers to process large amounts of natural language data. This data is typically collected from an unstructured language that needs to be converted to a computer understandable form [7]. Some key applications of NLP include named entity recognition, sentiment analysis, text summarisation, and many more.

Computer Vision

We, as humans, make a lot of deductions based on the things we see around us. Consequently, vision is one of the most important senses for us. Computer Vision (CV) is the field of study concerned with developing methods to enable computers to gain a high-level understanding of visual data like images or videos. The key goal of CV is to provide computers with the ability to understand and perform tasks automatically based on visual perception, in the same way as a human vision system does [8]. The field of computer vision gained immense popularity after AlexNet, a type of neural network developed by Alex Krizhevsky, won the ImageNet challenge in 2012 [9]. The key areas of applications of CV include driverless cars, face recognition, medical image analysis, and many more.

Autonomous cars using computer vision

 

Cognitive Computing

Most AI systems are tasked with something, and we judge them on their ability to perform the assigned task successfully. A different and exciting approach to building AI systems is Cognitive Computing, which brings humans into the loop. Traditional AI-based methods take complete control of a process and optimally complete the tasks. On the other hand, cognitive computing serves as an assistant to humans, where it studies the patterns and suggests a relevant course of action to a human-based on the data. This helps to increase further precision, where final decision-making still lies in the hands of humans [10].

Types of AI

Types of AI based on capabilities

Artificial Narrow Intelligence (ANI)

As human beings, we are versatile creatures capable of doing many tasks, some better than others. However, the same is not true for artificial intelligence based computers. Artificial Narrow Intelligence (ANI), also known as Weak AI,  refers to a computer’s ability to excel at one specific task and perform it even better than humans. Since it is easier to define a set of rules for one task, weak AI is preferred over strong AI because of its ability to perform well at a specific task. This system is bounded by a set of rules and works accordingly. Hence, it is exceedingly well at narrow tasks and performs them optimally [11]. An example of a Weak AI system is the Google Assistant, which works on the internet’s underlying database. Although it seems extremely intelligent for holding up conversations, the narrowness is reflected when it is tasked with something it is not designed to do, like, finding items in your house. This is the only type of artificial intelligence that has been realised to date to perform singular tasks like face recognition, self-driving car, voice recognition, etc.

Artificial General Intelligence (AGI)

Artificial intelligence, which is closer to human intelligence, does exist, although hypothetically. Artificial General Intelligence (AGI), also known as Strong AI, refers to a system’s ability to exhibit general intelligence, i.e., perform a large variety of tasks and mimic human behaviour. Machines equipped with AGI learn from experience and can transfer knowledge about solving one task to more tasks, precisely like humans [12]. The fact that the working of the human brain inspires strong AI, and there is a lack of comprehensive knowledge about the human brain poses a significant challenge to the advancement of AGI. Fujitsu built K is one of the fastest supercomputers, and it took 40 minutes to simulate a single second of neural activity [13], which indicates that there is a long way to go in the realisation of AGI. This certainly makes us think, is there a form of intelligence that even supersede human intelligence?

Artificial Super Intelligence (ASI)

Artificial Super Intelligence (ASI), or simply superintelligence, is a hypothetical AI system that can solve problems beyond the scope of even the smartest human beings. ASI would outperform humans in all life trades, including but not limited to science, math, art, emotions, etc. Billions of neurons limit the human brain’s thinking capability, and it is believed that ASI would overshadow it [14]. This kind of superintelligence is perceived as a threat by some, including Elon Musk, who believes that the rise of ASI could lead to the extinction of the human race.

Types of AI based on functionality

Reactive Machines

Humans “react” to the actions around them. The most simple and basic AI systems are developed on the same principle and are called reactive machines. These perceive information from the world around them and take appropriate action to complete the task. In such a system, the computer does not have any memory of past data nor any experience of performing the task. The machine understands the environmental constraints and the rules to complete the task successfully [15]. A key example of reactive machine AI is Deep Blue; a chess-playing AI developed by IBM that managed to beat grandmaster Garry Kasparov in 1997 [16]. To think about it, humans do consider their experience while reacting to their environment.

Limited Memory

Limited memory type AI uses past data and its predictions on that data to make future predictions. However, the term limited memory indicates that a machine can use past experiences only for a limited period of time. Autonomous vehicles are a classic example of a limited memory system. These cars store information about dynamic objects in their surroundings like the speed of the cars, state of the traffic lights only for a short period of time to make future predictions. The ultimate goal of artificial intelligence agents is to be indistinguishable from humans in every sense.

Theory of Mind

Theory of Mind AI systems should interact socially like humans, but such systems are not yet developed. These systems need to express human emotions, have beliefs, and need to understand fellow human beings [17]. There is active research ongoing to create such machines.

Self-awareness

A very ambitious goal and possibly the future of modern AI is building self-aware AI systems which are currently hypothetical. These systems will have a consciousness just like humans, which will enable them to be self-aware. This means that self-aware AI systems will be able to assess their capabilities and competency to complete a task and will also be able to communicate this knowledge in a human-understandable form.

Strong and Weak AI

Strong AI, also known as Full AI or Artificial General Intelligence, mimics the human brain more broadly. This type of AI exhibits a more remarkable resemblance to the human brain since it can understand a situation and perform actions exactly the way humans would do [12]. Since it is difficult to define a set of rules for a strong AI, weak AI is more preferred because of its ability to perform well at a specific task. Weak AI, also known as a Narrow AI system or Artificial Narrow Intelligence, can accomplish only one task [11]. This system is bounded by a set of rules and works accordingly. Hence, it is exceedingly well at narrow tasks and performs them optimally. An example of a Weak AI system is the Google Assistant, which works on the internet’s underlying database. Although it seems extremely intelligent for holding up conversations, the narrowness is reflected when it is tasked with something it is not designed to do, like, finding items in your house.

Subfields of AI

There are multiple subfields in the domain of artificial intelligence and can be listed as follows:

Neural Networks

Humans process complex information and make sense of it by passing it through a dense network of neurons. Neural Networks are a biologically inspired computing system that converts a specific set of inputs in one form to the desired set of outputs [5]. These mimic the working of the human brain and enable a computer to find and learn patterns in data [6].

Evolutionary Computations

Evolutionary computation (EC) is an exciting development in the field of AI. A set of candidate solutions is initially generated, and these are iteratively updated. EC generally models some natural phenomena like Darwinian principles. Every new generation of candidate solutions is obtained by dropping the previous generation’s less desired solutions and introducing randomness [18].

Vision

In the field of vision, artificial intelligence systems gain high-level information by processing digital images or videos. This field is inspired by the human’s visual perception systems, which help us to understand the world around us and perform tasks on the visual data[8]. Some key applications of AI-based vision systems include object recognition, scene description, face recognition, etc.

Robotics

Robots are a unique application of AI since these are systems that are equipped with manipulators that enable them to do human-like tasks. Robotic systems work by perceiving the environment, processing the gathered information using AI, and then picking, moving, or changing the physical properties of the objects. Good knowledge of computer science, electrical and mechanical engineering is required to develop robotic systems. It is quite tempting to think that cost reduction in manufacturing is the main motive behind the industrial revolution that robots bring along. However, aside from doing repetitive physically laborious tasks for us, robots also offer high precision machinery for surgical procedures and traverse over dangerous or physically inaccessible for humans during mining and rescue operations.

Expert Systems

An expert system is a type of AI-based computer system that emulates a human’s decision-making capability. These are designed to solve more complex problems using the knowledge defined as if-then rules instead of conventional procedural code. An expert system is made up of two subsystems:

• Knowledge base which represents facts and rules

• Inference engine which applies rules to deduce new facts [19]

Natural Language Processing

With NLP advancements in AI, we can now communicate with computers using the natural language [7]. NLP is used for a large variety of tasks like autocorrect and autocomplete, language translation, chatbots, voice assistants, and many more.

Speech Recognition

The technology that powers voice assistants like Amazon Alexa, Siri, etc., is Speech Recognition (SR), a domain of AI that deals with translating the spoken language into text by computers. This technology helps to increase convenience to users. SR is also capable of identifying humans based on their voice [20]. 

Machine Learning

We, as humans, don’t always have a well-defined procedure for solving tasks but are dependent on our experience for the same. Similarly, machine learning enables computers to learn from data and perform tasks without being explicitly programmed to do so, i.e., without any human assistance. The choice of the algorithm depends on the type of data and the task to be done [4]. 

Applications of AI

Artificial intelligence has various applications in today’s digital age because it can solve complex problems in different industries like healthcare, finance, education, entertainment, etc. Some of the applications of AI are as follows:

• We always love shopping at our favourite e-commerce stores, primarily because of their recommendations. AI technology has a crucial role in providing personalised shopping experiences for such platforms, which help to engage the customers better and, in turn, generate more sales.

• AI finds varied applications in the healthcare sector. It is used for medical imaging diagnosis to assist the doctors and reduce the time in the decision-making process. It also combines historical data and medical knowledge to help discover new drugs.

• Ever wondered what drives the most advanced robots in the world? Indeed, AI drives these robotic systems. It helps robots to plan their paths optimally and perform the maneuvers accurately. Computer Vision which is a crucial subfield of AI enables robotic systems to perceive the environment better.

• With the increase in online transactions and internet banking, the financial industry also reaps the rewards of AI technology. AI helps in fraud detection and management, predictive analytics for trading, and is also used for developing risk assessment strategies.

• AI in education can help to automate monotonous tasks for teachers like grading. Educational Softwares equipped with AI can provide adaptive courses which are custom-tailored to a student’s needs. AI systems can record student data and provide the teachers with feedback on their courses and point out places that need improvement.

Challenges of AI

The impact of artificial intelligence on human lives in this digital age is astonishing. However, there are some challenges associated with the development of AI systems:

• Most of the modern AI we see around us is powered by deep learning techniques. To develop a deep learning algorithm, we need to train it on vast amounts of data, and this training is computationally expensive. Consequently, most AI systems are power-hungry, and some even need the support of supercomputers, which are very expensive.

• For a non-specialist, the working of a deep learning algorithm, i.e., converting a specific set of inputs to the required output, is relatively unknown, which leads to a trust deficit. Moreover, most of us are unaware of AI systems’ usage or existence in our everyday electronic devices like smart TVs, smartphones, etc.

• More data implies better and more accurate AI systems. Major companies like Google, Facebook, Apple, and many more collect vast amounts of data. The security of this data and prevention of unethical use of data is very crucial. These companies have previously faced ethical charges against unfair use of data, and thus, data security poses a challenge to the advancement of AI [21].

Conclusion

Artificial Intelligence is an exciting branch of computer science that aims at developing machines that can make rational decisions based on external inputs. The goal of AI is to create systems that can mimic human behaviour and perform tasks that would otherwise require human intelligence. There is virtually no industry that modern AI hasn’t already disrupted, and the technology is growing exponentially. The impact of AI on our daily lives is very evident, even if we don’t notice it.

“AI doesn’t have to be evil to destroy humanity – if AI has a goal and humanity just happens to come in the way, it will destroy humanity as a matter of course without even thinking about it, no hard feelings.”

                                                                          - Elon Musk, Technology Entrepreneur, and Investor.

References

1. Turing, A. M. (2009). Computing machinery and intelligence. In Parsing the turing test (pp. 23-65). Springer, Dordrecht.

2. McCarthy, J., Minsky, M. L., Rochester, N., & Shannon, C. E. (2006). A proposal for the dartmouth summer research project on artificial intelligence, august 31, 1955. AI magazine, 27(4), 12-12.

3. Charniak, E. (1985). Introduction to artificial intelligence. Pearson Education India.

4. Raschka, S. (2015). Python machine learning. Packt publishing ltd.

5. Goodfellow, I., Bengio, Y., Courville, A., & Bengio, Y. (2016). Deep learning (Vol. 1, No. 2). Cambridge: MIT press.

6. Jain, A. K., Mao, J., & Mohiuddin, K. M. (1996). Artificial neural networks: A tutorial. Computer, 29(3), 31-44.

7. Bird, S., Klein, E., & Loper, E. (2009). Natural language processing with Python: analyzing text with the natural language toolkit. " O'Reilly Media, Inc.".

8. Szeliski, R. (2010). Computer vision: algorithms and applications. Springer Science & Business Media.

9. Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25, 1097-1105.

10. Bini, S. A. (2018). Artificial intelligence, machine learning, deep learning, and cognitive computing: what do these terms mean and how will they impact health care?. The Journal of arthroplasty, 33(8), 2358-2361.

11. Meskó, B., Hetényi, G., & Győrffy, Z. (2018). Will artificial intelligence solve the human resource crisis in healthcare?. BMC health services research, 18(1), 1-4.

12. Goertzel, B. (2007). Artificial general intelligence (Vol. 2). C. Pennachin (Ed.). New York: Springer.

13. Hornyak, T. (2013, August 05). Fujitsu supercomputer simulates 1 second of brain activity. Retrieved March 16, 2021, from https://www.cnet.com/news/fujitsu-supercomputer-simulates-1-second-of-brain-activity/

14. Müller, V. C., & Bostrom, N. (2016). Future progress in artificial intelligence: A survey of expert opinion. In Fundamental issues of artificial intelligence (pp. 555-572). Springer, Cham.

15. Nilsson, N. J., & Nilsson, N. J. (1998). Artificial intelligence: a new synthesis. Morgan Kaufmann.

16. Campbell, M., Hoane Jr, A. J., & Hsu, F. H. (2002). Deep blue. Artificial intelligence, 134(1-2), 57-83.

17. Cuzzolin, F., Morelli, A., Cirstea, B., & Sahakian, B. J. (2020). Knowing me, knowing you: Theory of mind in AI. Psychological medicine, 50(7), 1057-1061.

18. Angeline, P. J. (1995). Adaptive and self-adaptive evolutionary computations. In Computational intelligence: a dynamic systems perspective.

19. Waterman, D. (1986). A guide to expert systems.

20. Chiu, C. C., Sainath, T. N., Wu, Y., Prabhavalkar, R., Nguyen, P., Chen, Z., ... & Bacchiani, M. (2018, April). State-of-the-art speech recognition with sequence-to-sequence models. In 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) (pp. 4774-4778). IEEE.

21. Facebook–cambridge analytica data scandal. (2021, March 09). Retrieved March 16, 2021, from https://en.wikipedia.org/wiki/Facebook%E2%80%93Cambridge_Analytica_data_scandal