COVER STORY

Shobhna Kunwar

Humans, in their search for more and more intelligent machines, are not just trying to solve day-to-day problems but are also Of course, machines have not yet reached the level of the coming closer to solving the big mystery Robot, the problematic colleague of Dilbert. But the comic that we ourselves are. The quest for Artifi cial fi ctional episode is signifi cant in the light of the fast developing area of Artifi cial Intelligence (AI). The depiction in the comic Intelligence powered super beings may as refl ects human fears that intelligent machines may become too well end up in pushing the frontiers of our intelligent to allow themselves to be manipulated by humans. own thinking. Some time back, there was a temporary shutdown at Facebook’s AI project. The project was testing intelligence N one episode of the comic series Dilbert by Scott Adams, of two machines that had been programmed to bargain with Dilbert tells his colleague, “Our Robot was a good worker each other. But the machines seemed to outsmart their human I until we gave it Artifi cial Intelligence. As soon as it realised developers when they were detected speaking in their own it had immense strength and no soul, it started delegating.” The coded language. comic strip then shows the robot in another scene threatening Are machines then on their way to outperforming human? a co-worker to crush his head if he refuses to do his (Robot’s) This question puts AI and how it is evolving as a fi eld at the work. centre of a raging debate worldwide.

14 | Science Reporter | April 2018 What is Artifi cial Intelligence? a machine would not exhibit selective fragments of human Artifi cial Intelligence generally implies machines exhibiting intelligence but would rather behave like a human being in traits of human intelligence. The important traits of human terms of consciousness. intelligence include observing, taking inputs from the Full AI is also known as Artifi cial General Intelligence. observations, reasoning and rationalising, decision making, It is diffi cult to set a standard test to see whether a machine manipulations, learning from behaviour and patterns, has acquired the level of intelligence that a human possesses. communicating, analysing complex situations and performing Human beings reason, use strategy, solve problems, make tasks to achieve goals and targets. judgments, use commonsense, plan, learn, communicate, and At present, the subfi elds of AI like machine learning, integrate all these to achieve their goals. But apart from all autonomous systems, natural language processing, robotics and these, they are also autonomous in having their consciousness artifi cial creativity are popular areas of research. which has been a result of millions of years of evolution. To test the Artifi cial General Intelligence of machines, Strong AI and Weak AI some tests have been suggested. These are Turing Test, Coffee There are two types of artifi cial intelligence: Strong Artifi cial Test, College Admission Test and Employment Test. Intelligence (SAI) and Weak Artifi cial Intelligence (WAI). What has been actually called Strong AI is actually ‘full AI’. A full AI means a machine becoming capable of performing all intellectual tasks that a human being can perform. In this,

(Source: https://towardsdatascience.com/)

Turing Test: This was devised by Alan Turing, the famous mathematician, to deal with the question: ‘Can Machines Think?’ In this test, a machine and a human talk to a second human who must then judge which of the two is a machine. If the judge is unable to distinguish between the human and the machine, the machine is said to pass the Turing Test.

Table 1 : Thrust Areas of Contemporary AI Research

Machine Learning This is the fi eld in which computers are given the ability to learn without being explicitly programmed

Autonomous Systems A system that learns on its own to perform various tasks

Language Processing The fi eld of interaction between machines and human languages

Robotics Robotics in entirety is not just AI but there is a fi eld of AI powered robots

Artifi cial Creativity Creativity using computer. E.g. Aiva is an AI powered system that can compose soundtracks. Google’s Project Magenta is all about AI applications in arts.

April 2018 | Science Reporter | 15 College Admission Test: Ben Goertzel, the mathematician who has worked extensively on Artifi cial General Intelligence, suggested this test. The test sees whether a machine can enroll itself in a university, take the same classes that human students take and successfully obtain a degree.

Employment Test: Nils John Nilsson, the American computer scientist, suggested this test in which the machine is given an economically important job in which it must perform as well as or better than the level at which humans can do the same work. The tests suggested point to the disagreement as to what constitutes full AI. The tests are also challenging and there have Robots manning factory fl oors been reports of only Turing Test having been passed. Artifi cial General Intelligence therefore looks like a tough nut to crack. So far, there have been two reports of machines having Weak AI or Applied AI on the other hand, is a subset of passed the Turing Test. One was in 2014 at the University Strong AI as it involves machines being able to solve selective of Reading in the United Kingdom. In this a machine named actions that a human can perform. This requires more human “Eugene Goostman” simulated a thirteen-year-old boy at an intervention and the machine performs according to the program event. The Turing Test is said to be passed if the machine encoded in the machine. Weak AI is therefore application of manages to fool humans more than 30% of times that they are AI to selective functions. The present direction of AI research talking to a human and not a machine. The success of Eugene is towards this application aspect. however has been disputed (BBC June 9, 2014). The second machine that has been said to have cleared From Artifi cial General Intelligence to Applied AI the test is a machine developed by MIT’s Computer Science The history of AI research indicates that what is being touted and Artifi cial Intelligence Lab (CSAIL). It too was reported in as new actually began with much anticipation and focus on 2016 to have made humans believe that the sounds it produced the grand goals of developing a super intelligent machine. AI had been produced not by machines but humans (Robohub. research began in the United States with particular interest org, June 13, 2016). in developing machines that could translate from Russian to English and vice-versa. This was in the context of the Cold Coffee Test: This was suggested by Apple Inc. co-founder War with the then Soviet Union. Steve Wozniak. He said that machines could never be as intelligent as humans. He believed that a machine could not So, earlier researchers were interested in creating machines make coffee. So in this test, a robot has to go to an average that could do just any work that humans are capable of. Herbert American home, fi gure out how to make coffee. It has to fi nd Simon who is regarded as the pioneer of AI even said that the coffee machine, fi nd the coffee, mug and brew the coffee. machines will be capable within twenty years of doing any work a man can. The enthusiasm for AI with the ambitious goal of making machines as intelligent as humans lasted for nearly more than two decades among the government agencies in USA. However, the funding tap was turned off after some Strong Artifi cial disappointments in terms of realisation of grand ideas that had Intelligence fuelled the initial research (Table 2). By 1974, AI had entered its winter, a period when its research received cuts in funding and enthusiasm for research had reached its nadir. The revival of AI has taken place now with strides having been made in the applied aspect of Artifi cial Intelligence. The Weak focus now is on developing AI for solving particular problems. Artifi cial Scientists are optimistic that this approach to AI is better than the top-down approach in which the research initially was Intelligence focused on developing an artifi cial brain that would behave like a human brain. The applications that are attracting maximum attention are machine learning, robotics, autonomous systems and natural language processing. Autonomous systems are systems with Relation between Strong AI and Weak AI the ability to acquire or create content, interpret and develop

16 | Science Reporter | April 2018 Table 2 : AI Research – From Beginning to AI Winter

1950s Interest in Connectionism, Artifi cial Neural Networks, SNARC System, Machine Translation

1960s Machine translation proves to be inadequate; Automatic Language Processing Advisory Committee (ALPAC) concludes that machine translation project is not worth it due to poor performance of machines vis-à-vis human performance

1970s Connectionism, ANN abandoned; Cutback in funding; Lighthill Report by Prof. Lighthill calls AI a failure in terms of inability to achieve its grandiose aims

1980s Quiet cancellation of funding of AI research by Strategic Computing Initiative; Collapse of LISP machines market

1990s Sidelining of Fifth Generation Computing and Expert Systems Table 3 : Terms Related to History of AI

Artifi cial Neural Computing systems based on biological neural networks of animals. These systems Network (ANN) are made to progressively improve their performance by analysing examples. e.g. ANN system may learn to identify images of dogs by analysing images that have been labeled as ‘dog’ and ‘not dog’.

Connectionism Mental phenomena can be described by interconnected networks of simple and often uniform units. The form of the connection can vary from model to model. For example: units in the network represent neurons and connections represent synapses like in a human brain.

SNARC Systems Termed as Stochastic Neural Analog Reinforcement (SNARC). This was the fi rst neurocomputer designed by Marvin Minsky and Dean Edmonds in 1951 that simulated a rat trying to fi nd its way through a maze. It was a 40-neuron computer that was successful in modeling the required behaviour of a rat in trying to search for food in a maze. This was an example of ANN.

ALPAC Artifi cial Language Processing Committee was set up by the National Research Council of USA to look into the problem of research in machine translation.

LISP Machine It was a machine based on programming language called LISP. It was developed in 1973 at the Massachusetts Institute of Technology Artifi cial Intelligence Lab.

Expert Systems Systems that can solve complex problems using a knowledgebase that could be facts in a particular domain. An inference rules based system is applied to the facts to solve the problem.

Lighthill Report It is a popular name of the paper titled ‘Artifi cial Intelligence: A General Survey’ authored by Prof. James Lighthill and published in Artifi cial Intelligence: A Paper Symposium in 1973.

Fifth Generation It was a computing initiative launched in Japan in 1982 to make computers using massive Computing parallel computing (conducting parallel operations). The project was meant to provide a strong base for Artifi cial Intelligence. However, the project did not meet with commercial success due to better performance by less specialised hardwares.

knowledge from that content to make a decision on the basis see driverless drones for ferrying people. The passenger just of that interpretation. needs to press a button and it takes off vertically. They are now applied in the development of driverless Robotics is another area experiencing signifi cant changes cars and drones and fraud detection. Prominent examples of that have the potential to knock off many human relations. It has driverless cars include Waymo of Google that was unveiled in the potential to revolutionise areas of medicines, manufacturing, 2014, Tesla’s self driving cars Model S, Model X and Model defense, personal assistants, hospitality industry, personal 3. In February 2017, Dubai announced that its sky would soon uses, etc.

April 2018 | Science Reporter | 17 Intelligent robots

Driverless cars

that a computer program can learn and adapt to new data without human interference. It involves having an inbuilt source code or an algorithm which would create a model that builds predictions around the data it identifi es. Microsoft’s Project Adam and India’s Defence Research and Development Organisation Google’s Google Brain are machine learning projects with a (DRDO) has developed a robot called Daksh which can destroy focus on Deep Learning which in turn is a subfi eld of ML. life-threatening objects. Apart from applications like these, Deep Learning is the fi eld that is trying to make machines AI powered robotics has been developed for entirely different learn using imitation of workings of the human brain in uses. For instance, a robotics workshop in California, USA, processing data and creating patterns for use in decision making. called Abyss Creations has created a female life-size humanoid In Deep Learning projects, Natural Language Processing (NLP) called Harmony. She has been dubbed a sexbot. The main has emerged as an important area that would enable human- breakthrough in this robot is that harmony runs on AI to know machine interaction using speech and pattern recognition, what its owner wants and likes on the basis of AI. Although understanding of machine in reading comprehension. Harmony cannot walk, AI gets a boost from this because the Artifi cial Intelligence in India robot uses AI to understand human needs. India ranks third or fourth (depending on China’s position in However, it is machine learning as a subfi eld that has AI) in the size of AI clusters. India has already taken a lead in received attention for the advancements and research being this regard thanks to several AI startups. In this industry, the undertaken. Machine Learning (ML) is based on the concept

Table 4 : Popular AI Projects/Products

Google Brain Deep Learning AI project at Google

Adam Deep Learning project at Microsoft Inc.

Macie Amazon Inc. Security Service that uses machine learning to discover, classify and protect sensitive data in Amazon Web Service (a secure cloud service platform)

Siri Intelligent personal assistant, part of Apple’s iPhone Operating System (iOS) that relies on speech recognition, natural language processing, advanced machine learning

Google Assistant Google Inc.’s Virtual Personal Assistant AI that uses Natural Language Processing

Alexa Amazon’s intelligent personal assistant connected to smart speakers called Ama- zon Echo. Capable of voice interaction, music playback, making to-do-lists, setting alarms, streaming podcasts, playing audio books, giving real time information about traffi c and weather.

Watson IBM’s question answering computer system that is based on Natural Language Processing and Machine Learning. It is capable of answering questions posed in natural language.

Cortana Microsoft’s Intelligent personal assistant involving Natural Language Processing. Meant for Windows 10.

18 | Science Reporter | April 2018 First

” is a social humanoid robot developed by Hong Kong- based company . Able to display more than 62 facial expressions, Sophia has participated in many high-profi le interviews. And, in October 2017, the Sophia even became a Saudi Arabian citizen, the fi rst robot to receive citizenship of any country. The robot is modeled after actress Audrey Hepburn and uses artifi cial intelligence, visual data processing and facial recognition. Cameras within Sophia’s eyes combined with computer algorithms allow her to see. She can follow faces, sustain eye contact, and recognise individuals. She is able to process speech and have conversations using Alphabet’s Google Chrome voice recognition technology and other tools. Sophia also imitates human gestures and facial expressions and is able to answer certain questions and to make simple conversations on predefi ned topics (e.g. on the weather). Hanson designed Sophia to be a suitable companion for the elderly at nursing homes, or to help crowds at large events or parks. Sophia has seven robot humanoid “siblings” who were also created by Hanson Robotics – Alice, Albert Einstein Hubo, Bina48, Han, Jules, Professor Einstein, Philip K. Dick , Zeno, and Joey Chaotic. In December 2017, fellow Hanson robot BINA48 passed a college course on philosophy and love taught by Professor William J. Barry at Notre Dame de Namur University. David Hanson has said that Sophia would ultimately be a good fi t to serve in healthcare, customer service, therapy and education. private sector has been vibrant. The NITI Aayog has even been asked to prepare a roadmap for However, the 2018 Indian budget presented on 1st February a nationwide programme in Artifi cial Intelligence, including 2018 signaled the government’s intention to push for the research and development of its applications. development of Machine Learning and Artifi cial Intelligence. The government has also promised to offer incentives to

April 2018 | Science Reporter | 19 Table 5 : AI Startups in India

Active.ai Providing conversational banking service, aims to bring automation. Natural Language Processing and advanced machine intelligence are focal areas.

Alndra Systems Working on devices that can detect and identify people and objects

Artifacia Deep Learning

Artivatic Data Labs Ingestible Computing, which is used in making very tiny robots inside a capsule that can be ingested to collect data from the body

Arya.ai Neural network that will resemble human brain in functioning

Fluid.ai Facial recognition; present in problem solving for banking, fi nance, web and government sectors

Morph.ai Chatbots for marketing and customer services

Sig Tuple In medical diagnosis, AI powered visual analysis of medical data

Table 6 : CAIR Products Under development

AI Techniques for Net Centric Operations (AINCO) which will be a suite of technologies for creating knowledge base, information reception, handling, inferencing and event correlation Knowledge Resources and Intelligent Decision Analysis (KRIDA) that would manage large scale military moves by using knowledge base and data handling RoboSen, a mobile robot system that would patrol, conduct recce and surveillance for armed forces and paramilitary forces and police. Can negotiate autonomously, avoiding obstacles and can give video feedback. A miniature unmanned Ground Vehicle which is a man-portable robotic system for low intensity confl icts and surveillance in urban setting. Robots meant for logistics support that can walk with six and four legs Robots that can climb walls and fl ap wings for low intensity confl icts CAIR_ICR (Intelligent Character Recognition) system for the automated processing of handwritten entries. Snake Robot has been designed after analysis of gait of snake. It is for use in search and rescue operations during natural calamities. startups and venture funds that undertake application-oriented effort to give quick results. research on Artifi cial Intelligence across key sectors including banking, insurance, education, health, retail and transportation. Is AI Moving Towards Super Intelligent Machines? The Department of Science & Technology is also looking into The question is pertinent since the revival of intelligent setting up of centers of excellence that will push investments machines has not just created euphoria over its potential benefi ts in “research, training and skilling in robotics, artificial but has also given rise to fears that it would replace humans. intelligence, digital manufacturing, big data analysis, quantum Even as robots are being made for performing simple communication”. human tasks that do not require complex problem solving, Apart from these vibrant private sector clusters, India’s there is a worry that planet Earth would turn into a place Centre for Artificial Intelligence and Robotics (CAIR), where humans would be slaves to machines that would be more a laboratory of the Defence Research and Development intelligent than them. Artifi cial Intelligence is defi nitely trying Organisation (DRDO), has developed a number of products that to create machines and systems that will be more intelligent are powered by AI. Many are under development. Most of the than their predecessors. So, this march of AI may lead to a products are in the areas of robotics, autonomous systems and future where machines are extremely intelligent but the fears intelligent systems that would do analysis without any human may turn out to be exaggerated.

Ms Shobhna Kunwar has submitted her Ph.D. thesis on the topic ‘Cyberspace Governance: A Study of Russia’s Approach’, at the School of International Studies, JNU, New Delhi. She also has an article titled Strident Politics and Grey Economics: Debating Net Neutrality in the January 9, 2016 issue of Economic and Political Weekly to her credit. Address: DDA Flat 238, Sector-A-10, Pocket-6, Narela, Delhi-110040; Email: [email protected]

20 | Science Reporter | April 2018