Progress and Perils of Artificial Intelligence (AI)

Dr. RGS Asthana

Senior Member IEEE

Figure 1: Perils of AI [23]. Is it true?

Summary

It is possible today to program robots or, in general, we refer them generally as the machines or automation to do a lot of repetitive works for us but for doing this they need not be super intelligent. This is an issue of management and we must do it well. However, after a few decades and looking at current pace of progress the intelligence imparted to machines by human beings is going stronger enough to a level that it may be of concern. This level of intelligence is referred to as the super intelligence which may someday take on human race and this thought is bothering people like Elon Musk and many others.

Big Data [27, 36], Cloud Computing [33], InfoGraphic [34], Internet of Things [27, 35] and AI technologies when used together may show very interesting results [27, 32]. In fact, humans, are limited by slow biological evolution, and couldn’t compete with ever evolving AI and would in all likelihood be superseded one day but when nobody knows for now.

Keywords

Artificial Intelligence (AI), Weak AI, Strong AI, Deep Learning, Reinforced Learning or (RL), Generative Adversarial Network (GANs), Machine Learning, Neural Networks, Natural Language Processing (NLP), Big Data, Cloud Computing, InfoGraphic, Internet of Things

Prelude [1-5]

Elon Musk is a very future-oriented tech entrepreneur, as evidenced by Tesla and SpaceX. Even so, Musk has always been vocal about his fear of a world take-over by artificial intelligence (AI). Curiously enough, his solution to keep the AI doomsday away is to merge the human brain with machines [3]. Some of the most popular sci-fi movies — 2001: A Space Odyssey, The Terminator, The Matrix, Transcendence, Ex Machina, Transformers, Robot and many others — were created assuming AI will progress in time to a level at which humanity will lose control on its own creations, leading to the end of entire civilization. AI [37] has recently attracted attention of even high profile figures including Stephen Hawking, Bill Gates and Elon Musk and they have voiced concern about the perils it may bring with its advancement.

A few quotes given below illustrate the perils of AI as well as comments of an AI expert saying that they are not correct:

‘If AI power is broadly distributed to the degree that we can link AI power to each individual’s will — you would have your AI agent, everybody would have their AI agent — then if somebody did try to something really terrible, then the collective will of others could overcome that bad actor.’

—

‘I think we should be very careful about artificial intelligence. If I had to guess at what our biggest existential threat is, it’s probably that. So we need to be very careful.’

—

‘I think we need to be very careful in how we adopt artificial intelligence and that we make sure that researchers don’t get carried away,” he stated. “Sometimes what will happen is a scientist will get so engrossed in their work that they don’t really realize the ramifications of what they’re doing.’

—

Elon Musk

‘I think the development of full artificial intelligence could spell the end of the human race.’

—

‘Once humans develop artificial intelligence, it will take off on its own and redesign itself at an ever-increasing rate.’

—

‘A super intelligent AI will be extremely good at accomplishing its goals, and if those goals aren't aligned with ours, we're in trouble.’

—

Stephen Hawking

“I don’t think it’s very helpful for other people who are incredible in their domains commenting on something they actually know very little about, but because they are quite big celebrities now, more than just scientists or businessmen, it gets picked up a lot,”

—

Hassabis, CEO of Deep Mind bought by Google in 2014

Hassabis, CEO of Deep Mind says further that he has criticized the likes of Microsoft founder Bill Gates and SpaceX and Tesla founder Elon Musk for their comments on AI, the Times of London reported on Mar. 27, 2017. The main comment was that they are not an AI expert and hence may not be right. During an "ask me anything" question and answer session on Reddit in Jan, 2015, Mr. Gates wrote: "I am in the camp that is concerned about super intelligence. First the machines will do a lot of jobs for us and not be super intelligent. That should be positive if we manage it well [12].

AI is the theory and development of computer systems that can perform tasks that take human intelligence. Facebook uses AI for targeted advertising, photo tagging and news feeds. Microsoft and Apple use AI to power their virtual intelligent assistants’ viz. Cortana and SIRI respectively. Google has always used it for its search engine.

Sci-fi novelist David Brin feels that these dire predictions are often simplistic and unreasonable [23] (see figure 1).

What is AI [7, 8]?

AI is embedded in Cortana and SIRI from Microsoft Corp. and Apple Inc. respectively and AI is also in self-driving cars technology.

As per Wikipedia [7], Weak AI (also known as narrow AI) is focused on a narrow task only (e.g. only playing chess, solving equations, facial recognition, only internet searches or only driving a car or an automatic language translator that converts only one language say English to Russian). Narrow AI is designed to analyze data and form conclusions far more efficiently than humans can. It, presently, can outperform humans at whatever its specific task. AI ultimately would outperform humans at nearly every cognitive task

SIRI from Apple Inc. is a good example of narrow intelligence. SIRI operates within a limited pre-defined range. It has no genuine intelligence, no self-awareness and also no life despite this being a sophisticated example of weak AI.

The purpose of general AI is to mimic human behavior as much as possible. A Developer’s goal is to make a system as life-like as possible [48]. Weak AI is defined in contrast to either strong AI (a machine with consciousness, sentience and mind) or artificial general intelligence (a machine with the ability to apply intelligence to any problem, rather than just one specific problem). All currently existing systems considered based on AI of any sort are weak AI at most.

Any suitably programmed computer with the right Input/ Output or I/O can simulate a mind in exactly the same sense human beings have minds. According to Strong AI or General AI, the correct simulation really is a mind. According to Weak AI, the correct simulation is a model of the mind [11]. A strong AI automation or machine will, therefore, be capable of doing all tasks instead of one as in case of a weak AI machine. It will outperform human mind in all cognition tasks or in other words, will be super intelligent. Such an automation once accomplished may be a cause of risk to human race in time to come.

While a sci-fi movie often depicts AI as robots with human-like characteristics, AI is included in almost anything from Google’s search algorithms to IBM’s Watson (can understand all forms of data, interact naturally with people, and learn and reason, at scale [13].) to autonomous weapons (US has put AI at the center of its defense strategy, with weapons that can identify targets and make decisions [24, 27].) Such weapons in form of drone - with six whirring rotors and a camera - are freely available at Amazon. However, these are equipped with advanced AI software to transform them into a weapon that could find and identify men carrying replicas of AK-47s around say, a village and posing to be rioters but are activists or insurgents or extremists. The autonomy may (or not) be shrewdly compromised while taking an eliminate decision. The Pentagon’s latest budget is $18 billion for the years 2016-18 on technologies that included those needed for autonomous weapons. In fact, Big Data and AI combined will bring new research projects.

Searle [10] does not disagree that AI research can create machines that are capable of super intelligent behavior. The Chinese room argument literally says that a digital machine could be built that could behave in more intelligent manner than a human, but does not have a mind so it cannot behave in the same way as human brains do. Indeed, Searle writes that "the Chinese room [9] argument ... assumes complete success on the part of artificial intelligence in simulating human cognition."

AI vs. Machine Learning vs. Deep Learning [40]

· AI: The area includes automated “knowledge” tasks, Robotics, Internet of Things, 3D Printing technology, and self-driving cars. It is expected that the total commercial effect of these technologies will be USD 50 - 99.5 trillion by 2025.

· Machine Intelligence (MI): Many experts believe that Machine Intelligence and AI are interchangeable terms. “Machine Intelligence” is generally used in Europe, while the term “AI” is used in the US. MI usually shows a contribution of a biological neuron in the research process with deployment of advanced methods than used in Simple Neural Networks.

· Machine Learning (ML): It is inherent part of AI. Machine Learning refers to the software research area that enables algorithms to improve through self-learning from data without any human intervention i.e. unsupervised learning.

· Deep Learning (DL): Deep Learning is really a branch of ML. It pertains to the study of “deep neural networks” in the human brain. Deep Learning tries to emulate the functions of inner layers of the human brain, and its successful applications are found in the areas such as image recognition, language translation, or email security. The Deep Learning technology is modeled after the human brain.

Such systems learn and improve performance over time, and adjust the weight of their nodes and also data sets to remain efficient. To reduce requirement of processing power most training features in commercial AI are set to very simple levels only [48].

Computer - Brain Interface [3]

Efforts to develop human-machine or Brain–Computer interface are also on the way. Neuralink - a US startup company developing implantable human-computer interfaces - was founded in 2016 by Elon Musk and first publicly reported on in March 2017. The U.S. Defense Department’s research arm, DARPA, is also working on similar technologies. “We are giving our physiology the opportunity to work with machines in a different way,” said Justin Sanchez, director of DARPA’s Biological Technologies Office.

The effort is to merge human intelligence with machines? This may also help to avoid an AI doomsday scenario and hence Elon Musk’s interest in such a venture.

A quote on Brain–Computer interface follows:

‘Such a massive interconnection will lead to the emergence of a new global consciousness, and a new organism I call the Meta-Intelligence.’

—

Peter Diamandis

Progress in AI [14 - 17]

Neural Information Processing System (NIPS) 2016 is an annual event. “Reinforcement learning” (RL) [25] and “Generative Adversarial Networks” (GANs) are the in thing in AI now. RL is the process to emulate the animal learning. The idea is to interpret how certain behaviors tend to result in a positive or negative outcome. Using this method, a automation can navigate a maze by trial and error and then associate the positive outcome—exiting the maze—with the actions that led up to it. This lets a machine learn without instruction or even explicit examples. As per definition, RL allows machines and software agents to spontaneously fix the ideal behavior within a specific context, in order to maximize its performance. Simple reward feedback is used as reinforcement signal for the agent to learn its behavior.

The idea has been around for decades, but combining it with large neural networks provides the power needed to make it work on really complex problems (like the game of Go — originated in China more than 2,500 years ago). AlphaGo [16] with hit and trial came out for itself how to play the game at an expert level.

Work on GANs was initiated by Ian Goodfellow [15] — a system of network consisting of two independent networks. One network referred to as ‘D’ generates new data after learning from a given (real) training set. Other Network referred to as ‘G’ tries to differentiate between real and fake data. This approach could be used to generate video-game scenery, de-blur pixelated video footage, or apply stylistic changes to computer-generated designs. In fact, both GANs and RLs lead to improving performance of unsupervised machines (neural networks [17, 26 and 29]).

Goodfellow [15] had shown that network ‘G’ performs a form of unsupervised learning on the original dataset. Further, Yann LeCun [31] - Director of AI Research, Facebook and Founding Director of the NYU Center for Data Science - stated, unsupervised learning is the “cake” of true AI. This powerful technique (called GAN) is claimed to be easily programmed using PyTorch [30], in under 50 lines of code.

The other area of research in AI is language is to parse and generate language effectively. This is a long-standing goal in artificial intelligence, and the prospect of computers communicating and interacting with us using language is interesting. Better language understanding would make machines a whole lot more useful. Natural Language Processing (NLP) [22] solutions can parse regulatory text and pattern match with a cluster of keywords to identify the changes relevant to particular organization [20].

Contextual analysis of vast volumes of conversations from phone recordings, chats and emails conversations can help identify potential market manipulation and collusion activities instead of traditional text and voice processing techniques. Similarly, deep learning techniques on the transactions or the business rules can become more refined and significantly reduce activities flagged for investigation instead of using traditional methods.

Don’t expect to get into deep and meaningful conversation with your smartphone as some remarkable changes can be expected in 2017 due to further advances.

Perils of AI [19, 20]

Some of perils of using AI to solve problems are well known, e.g. herding behavior, out-of-sample extrapolation, and false correlations, to name a few. Besides the above, some additional risks do occur as compliance teams become more technology driven and become reliant on AI based machines to do their jobs.
In 2013, a single fake tweet from a verified account about an explosion in the White House briefly wiped out about $140 billion in US market value. Do we have a full proof AI solution yet? The answer to this question is probably no even today but it may become yes tomorrow.

AI may learn gender bias discrimination through historical data, which can often be twisted for decision making. Take for instance word embedding algorithms in NLP [22]. These algorithms do carry historical categorizes (sexism for example) into the future simply by learning patterns regarding such words that often appear together in historical data. We can automate decision making process but we need to ensure that it may not convey false level of confidence in people in machine performance, e.g., an NLP solution that tracks thousands of regulatory changes.

Similarly, AI also creates a false level of confidence amongst regulators in establishing whether due process is being followed for the different business practices such as sanctions screening or suspicious activity reporting.

Mercedes-Benz’s self-driving car program assigns passenger safety over pedestrian safety by design. It’s a question of Ethics. However, such risks can undo the whole institutions performance, wiping out the edge on its revenue generation side. Companies like Microsoft, Amazon and Facebook have acknowledged these risks and use ethical best practices for AI development. The volume of regulatory changes, increased difficulties posed by of cheats as well as enhanced scrutiny from regulators will predictably make compliance teams turn to AI for solution.

The other risk of AI super intelligence is to survival of humanity as once we have cognitive power of brain and emotions imparted to machines (human like robots as often depicted in movies) then there is a risk of takeover of humanity by machines. Such risks have been talked by celebrities and experts as also mentioned in prelude to this article.

It is expected that advancement in robotics and use of AI [21] could affect nearly 30 percent of U.K. jobs by the 2030s, compared to 38 percent in the U.S., 35 percent in Germany and 21 percent in Japan. Robots are likely to take lot of existing jobs in near future.

Bill Gates says that Machines should also get taxed just like Human Labor. Jobs where you've got more of a human touch, like health and education would be safer but automating more manual and repetitive tasks will eliminate some existing jobs, but could also enable some workers to focus on higher value, more rewarding and creative work, removing the monotony from our day jobs.

How AI will shape in future [5]

The future use of AI is in the fields like healthcare, smartphone assistants, and robotics. Deepmind is in partnership with the NHS where machine learning is to be used in healthcare. IBM, Watson is in AI, where expert system which can help in medical diagnosis of images and helping people have healthier lifestyles, are developed. However, for NHS the first stage is to develop helpful tools for visualizations and basic stats on all sorts of twenty first century platforms mainly smart phones and tabs.

Deepmind will make smartphone assistant smart and context sensitive and have a deeper appreciation of what owner of phone is trying to do. Presently, most of these systems are extremely weak as they are mainly template based. Once you go off the templates given in the program then these systems prove pretty useless.

Robotics is used mainly in two ways. You have companies like Fanuc [6] making industrial robots that do incredible sounding things for a very specific purpose and these tasks do not need intelligence, and then one has these concierge-style robots like Softbank’s Pepper and so on with some intelligence and are basically similar to smartphone assistants. In summary, you have robots which only use pre-programmed templates and other type which are little intelligent.

Further, the concierge robots may also be pre-programmed with template responses and it is easy to confuse them. A lot of research is being done in the area of machine learning [18] to inculcate cognition in the machines.

In the month of Feb. 2017 [38] Google released, “Tensorflow 1.0” – an open source library for machine learning or AI – which is a collection of latest AI set of tools. This library, as per Google, is production-ready by way of its application programing interface (API). Artificial neural networks (ANN) [26] with a few enhancements are called Deep Neural Network (DNN). DNN is an ANN with multiple hidden layers of units between the input and output layers [44]. ANN and DNN can be trained on data. Tensorflow includes tools for computing K-means and support vector machines (SVMs). Deep Learning [39] involves a cascade of many layers of nonlinear processing units for feature extraction and transformation. Each successive layer uses the output from the previous layer as input. The algorithms may be supervised or unsupervised and its applications include pattern analysis (unsupervised) and classification (supervised). Convolutional neural networks (CNNs) have now become the de facto method for visual and other two-dimensional data. A CNN is composed of one or more convolutional layers with fully connected layers (matching those in typical ANNs) on top. These can also be trained with standard back propagation techniques. CNNs are easier to train than other regular, deep, feed-forward neural networks and have many fewer parameters to estimate, making them a highly attractive architecture to use.

What is Hadoop? As per definition of Whatis.com [47], “It is an open source, Java-based programming framework that supports the processing and storage of extremely large data sets in a distributed computing environment. It is part of the Apache project sponsored by the Apache Software Foundation.” Forrester Research predicted that the Hadoop revolution (very useful once upon a time and a de facto platform for handling Big Data) will be taken over by the AI revolution by declaring this in a March 2017 report that “artificial intelligence will eclipse Hadoop [46].” Forrester argues that “Hadoop may have sparked a revolution in analytics, but the most recent revolution is the deep learning.” First hit to Hadoop came from companies offering cloud services and giving modern management services in their platform. One of the authors of the Forrester report, Brian Hopkins, has already answered this claim:

“Just like Hadoop wasn’t designed for the cloud, it wasn’t designed to do the matrix math that deep learning requires. And the cloud crew is busy creating specialized AI-friendly environments, which means Hadoop vendors have even more work to do to keep their software relevant. Will they make Hadoop a platform for AI?” The answer is probably negative.

Way forward

AI in the area of Marketing will improve mobile search with relevant results to help marketers in better conversions justifying higher ad rates. With the help of AI search engines are likely to bring better search results that are custom-made for a specific point in time – say, a micro-moment when users might be vulnerable and buy if certain set of results are delivered by a search.

One may expect to see better language understanding and an AI boom in China, among other things [14]. The sole idea is to enhance productivity and maintain quality thus save billions of dollars.

AI will be very useful in processing of large text say coming via e-mails or voice recorded data due to progress in NLP area.

AI's scientific challenge [28] comprises of providing new heuristic based computational models that accommodate the wide range of capabilities attributed to human brain and considered useful even for nonhuman intelligence. Common logical keystones help in understanding theories of knowledge representation, planning, problem solving, reasoning, and some aspects of NLP whereas economic aspects and the mathematics of Markov decision processes help unify probabilistic forecasting, fault diagnosis and repair, reinforcement learning, robot control, and many aspects of speech recognition and image processing. Many of these include cross-disciplinary boundaries and lead to integration with other fields. AI has also included logic, philosophy, psychology, and linguistics for some time in its domain. Inclusion of economics, decision theory, control theory, and operations research has served as a focus for more recent efforts. Next area is integration of big data results with AI and to see and analyze how these systems can derive new heuristics from large amount of data processed using big data methods instead of heavy mathematical time-consuming computations. AI may be useful even in handling big Data [32].

Experts believe that AI, Big Data [27, 36], Cloud Computing [33], InfoGraphic [34], Internet of Things [27, 35], will continue to trend in 2017, building on Big Data as data is an essential building block of AI. One may also see integration of data from variety of apps in many fields including machine translation and image recognition.

The IBM supercomputer Watson [41] has exploited power of science, technology and culture to develop many AI based systems. 34 Insurance employees in Japan were made redundant to make way for AI based system. This trend is likely to continue and may go viral in certain industries over the coming years. And as IBM Watson's supercomputer gets smarter day by day with more and more experience, soon a time will come when the robots will no more think like humans, but the humans will need to learn to think like robots.

Besides threat to several types of mundane and routine jobs, Watson has made some incredible breakthroughs in areas such as reducing threats of cybercrimes, automated diagnostic or doctor and automated lawyer: Ross Intelligence and IBM Watson joined hands to tackle legal matters [42]. Other apps of AI with high potential for replacement will be in areas such as replacing teachers, auto-grading student papers, for publishing industry: automatically writing high-quality articles in a short amount of time; automated chefs: making new and tasty recipes and replacing expansive chefs in restaurants; and automated policemen or soldiers: algorithms will decide who to kill or who to arrest so this may take time although drones are replacing soldiers in a number of wars [45]. All of this can be achieved by using latest advances in many areas of science, technology and culture. It also establishes the fact that AI and its capabilities are going to improve day by day and thus become very useful to the humans.

Lawyers can do research on The ROSS Legal App or automated Lawyer system by asking questions in natural language, just as they would consult their colleague. Since this app is built upon a cognitive computing system, therefore, it is able to scan over a billion text documents in less than a second and return the exact passage the user needs. Thus, there is no need of endless Internet browsing and database searches to get to desired results.

A future challenge will be that AI based systems will be deployed in many new areas and this may also be a cause to bring unemployment. But it may also create jobs with new specs. The quantum of new jobs created may be small. So while many of these research areas in and around AI, Data Science, ML, MI, and Deep Learning show a lot of potential, however, the execution and fine-tuning route will add main risks if deployed hastily and without suitable preparation.

Can AI writes AI algorithms, or in short, can AI automate AI [45]? Answer to this question will be big yes in time to come.

Can one inculcate emotions in a machine or automation or robot [37? How would you feel about getting therapy from a robot [43]? Emotionally intelligent machines may not be as far away as it seems. Over the last few decades, AI has improved a lot in understanding emotional responses in humans. It’s possible that AI may one day matches humans in recognizing different types of emotions. But whether an AI based machine could ever experience emotions? Further can one give a command through a computer to masses via computer-brain interface? If answer to any of two questions above is in affirmative then the problem of takeover by machines may not stay imaginary.

References

[1] Stephen Hawking, Elon Musk, and Bill Gates Warn About Artificial Intelligence

http://observer.com/2015/08/stephen-hawking-elon-musk-and-bill-gates-warn-about-artificial-intelligence/

[2] What is Artificial Intelligence? Stop worrying about AI destroying humans, expert says

http://www.ibtimes.com/what-artificial-intelligence-stop-worrying-about-ai-destroying-humans-expert-says-2515898

[3] Elon Musk just confirmed that he’s making Tech to merge human brains and computers

https://futurism.com/elon-musk-just-confirmed-that-hes-making-tech-to-merge-human-brains-and-computers/

[4] Here’s what Elon Musk thinks about Brain Implants, AI, And Basic Income

https://futurism.com/4-elon-musk-talks-tesla-ai-and-more-in-dubai/