Once again, instead of new writing I’m posting an old essay from undergrad. This time the focus was AI [buzzword gang]. Once again, there is a real sense of naiveness to the thoughts expressed here, but my caveat continues to be that I had zero experience in the field at the time of writing and also I was 19. Anyway, here is my slightly naive and silly take on AI in (what was then) the present and the future.

Abstract

Artificial Intelligence is an emerging and disruptive technology that has been identified as the next great shift in society and production. Society for the most part treats this new field with fear and apprehension. Both the general public and the world’s top minds are divided over the direction artificial intelligence will take and whether it will change the world or end it. Although there is no consensus on the direction this technology will take, the key to landing on the same page as a population will be to educate and spread information in order to help society come to terms with this new stage of innovation and enable leaders to make considerate and deliberate decisions. The purpose of this paper is to educate and enhance the understanding of artificial intelligence and analyze its potential for positive and negative consequences in society.

Diego Veras 12/7/17 ENT4934

Artificial Intelligence and Society: Today and in The Future “The development of full artificial intelligence could spell the end of the human race…. It would take off on its own, and re-design itself at an ever increasing rate. Humans, who are limited by slow biological evolution, couldn’t compete, and would be superseded.” This is not a line from a sci-fi thriller or a post-apocalyptic novel but a quote from Dr. Stephen Hawking, who is generally considered a pretty smart guy. This remark is just one of many from several of the world’s greatest minds who are concerned with the potential threat that artificial intelligence poses on the world and this sentiment is mirrored in the general public, where most people’s first thoughts on AI and its development always revolve around the danger it poses to individuals and society. This reaction stems from the fact that humanity has always feared what it doesn’t understand. However, every emerging technology always has potential for both positive and negative consequences, and the purpose of this paper is to expand on what AI actually is while enhancing the analysis towards its potential for positive impacts and negative effects on society. To fully understand the potential that AI poses, it is important to examine its history . The idea of intelligent and autonomous machines has existed in fiction for decades but the term “artificial intelligence” itself was officially stated by John McCarthy in 1956 at the Dartmouth Conference. That same year, two scientists revealed the first running AI program, which was named the “Logical Theorist”. By 1997, the Deep Blue Chess program defeated the world chess champion at the time, Garry Kasparove (AI Topics, n.d.). It was around this time period that the general public began to truly acknowledge the deep power that machines possessed due to the mass realization that a person at the top of their field had had been bested in a human game by a program. History and current events tend to show a trend in relation to innovation, particularly in terms of technology. This trend (known in the technical world as Moore’s Law) is generally stated as an exponential growth in computing power with a dramatic decrease in costs (Intel, n.d.). The technology surrounding artificial intelligence is no exception to this trend and that can be easily seen when looking at the speed at which this field has expanded. In today’s world, artificial intelligence has grown so developed that it can actually be broken down into 3 sub-categories, which are hierarchical in terms of power and autonomy. These categories are artificial narrow intelligence (ANI), artificial general intelligence (AGI), and artificial Super Intelligence (ASI) (briefly described in Figure 1.1)

Figure 1.1

Artificial narrow intelligence is the base level of artificial intelligence computing, and utilizes machine learning to focus on performing one particular function at the level of human intelligence (McCelland, 2017). A great example of this is the chess program that beat Kasparove, Deep Blue. This machine was excellent at chess, but not much else. Machine learning is at its core the use of algorithms to analyze large amounts of data, learn, and then find trends and predict outcomes without being explicitly told what to look for (Thompson, Li, Bolen, n.d.) . The easiest way to think of this level of intelligence is to rank it in terms of human input. At this level of intelligence, the machine can learn and anticipate but it still requires higher level of human input to point them in the right direction and provide the data that it needs to extrapolate conclusions and results. More recent examples of advanced systems at this tier include AlphaGo, a program designed to play the the Chinese game Go, one of the oldest games in human history. This game also happens to contain more possible board configurations than the number of atoms in the known universe. In 2016, AlphaGo managed to beat the world Go champion and arguably best player in history, Lee Se-dol. AlphaGo utilizes machine learning in the same logic as Deep Blue, but it uses neural networks and streamlined architecture to teach itself exponentially as it plays (DeepMind, n.d.). Programs like AlphaGo are examples of the power narrow intelligence possesses when compared to the human mind. One of the most overlooked characteristics of artificial narrow intelligence is its wide spread use. Most of the general public does not recognize many of today’s systems and machines as ANI driven, yet some of the most common applications in our day to day life such as search engines, virtual assistants, and even video games all fall into the category of narrow intelligence (Westerheide, 2017). . The underlying foundation of ANI is machine learning, a concept touched on earlier. Machine learning has evolved over the years in accordance to humanity’s increased technological and financial advances. The early days of machine learning can be identified to be in the 1950s, when Arthur Samuel wrote the first computer learning program with a purpose of playing checkers and the system getting better in every iteration (Medium, 2017). For the following decades, machine learning and the algorithms at the heart of the systems continued to increase in complexity. By the 1990s, machine learning was being used for professional applied functions and included a major break through in the logic used to create systems with the ability to learn. This shift took the approach from being knowledge-driven and shifted to the data-driven approach that constitutes modern machine learning. This switch in methodology along with higher computing power has allowed AI systems to advance to applications on a more complex scale. The narrow AI systems of today learn through a process that provides a start state, based on the universe in which the problem exists, and the specific goal, and then letting the learning process system figure out the intermediate states and how to progress from one state to the next. The states and transitions are developed with the use of statistics and big data (Drepper, 2017). Today’s society quietly revolves around the wide spread use of AI in every day life and this helps to balance out the fear and hesitancy that society has attached to artificial intelligence. The next tier of AI is artificial general intelligence, which is cutting-edge today. Most projects close to reaching this level are engines similar to Google’s DeepMind. Artificial general intelligence is defined by the AGI Society as “a general-purpose system with intelligence comparable to that of the human mind (and perhaps ultimately well beyond human general intelligence)”. The key difference between AGI and ANI is the scope and width of their ability to process at human like levels or above. Reaching the true levels of artificial general intelligence is the goal in today’s intelligence community, but it is not exactly here yet. The engines closest to general intelligence use a technique called deep learning to perform cross-domain tasks. Deep learning uses neural networks with layers of processing power to learn complex patterns in large amounts of data. Some examples of the way AGI is currently being used include image and speech recognition, which are the engines that drive the highest level of data analytics and virtual interactions with people. Although true artificial general intelligence does not currently exist to the fullest extent, the concepts that will help us achieve it are being put in place within many of the most prominent AI engines in the industry. Deep learning is the next level as far as machine learning and is integral for today’s systems. It is the logical strategy being utilized in order to approach general intelligence in machines. Deep learning utilizes algorithms that are inspired and derived from the system of neural networks within the human brain. When describing this method of learning, renowned computer scientist Yoshua Bengio described systems being able to “learn feature hierarchies with features from higher levels of the hierarchy formed by the composition of lower level features. Automatically learning features at multiple levels of abstraction allow a system to learn complex functions mapping the input to the output directly from data, without depending completely on human-crafted features.” Using this process, systems with general intelligence will be able to solve issues across all areas that would typically require a human brain. Past the first two categories in this field lies artificial super intelligence, which is exactly what it sounds like. There is currently no system capable of operating at this capacity, although there are some scientists that see it coming within 65 years or so (Müller, Bostrom, 2014.). The most basic way to think of a machine that would be in this category is a system that can perform any function in any domain at a level far higher than even the greatest human mind could attempt. Any ASI system would need to incorporate the concept of “Theory of Mind” which is essentially the comprehension that other entities in the world can have thoughts and emotions that affect their own behavior (https://theconversation.com/understanding-the-four-types-of-ai-from-reactive-robots-to-self-aware-beings-67616). In layman’s terms, a machine would have to achieve self-awareness. This is the level of intelligence that is often depicted as the fall of society in popular culture. There are currently 2 school of thoughts in regards to how this level of technology will be achieved. The first group believes that the key lies in advances in machine learning and projected hardware innovation, they place their beliefs in the idea that a breakthrough in the current methodology will help us reach super intelligence. On the other hand, there is another group who believe that the key lies in the human brain itself, their school of thought is directed at replicating the structure and logic of the human brain using far deeper and more complex neural networks and applying the concept of the binary firing that neurons in the human brain utilize (Harvard Science Review, 2015). Once operational, a system with ASI would essentially become the lead researcher within the field of artificial intelligence because it would be able to create its own AI systems at a far faster rate than humanity could. It remains to be seen exactly how we will reach super intelligence, but it is clear that a system of this scope would be more powerful than anything we have seen before. Artificial super intelligence would be a paradigm shifter due to the fact that it would be able to create on an exponential scale and this would signal a new era in production. Human history can be broken down into periods dominated by a certain type of machine and the technology that powered those machines. The initial period encompasses the 18th century, which was the first Industrial Revolution and used steam-based machines. Following this, the second industrial revolution began in the 19th century and continued through the 20th century and was dominated by electrical and energy-based mass production. Then the first information revolution occurred and this was in the late 20th century. The first information revolution operated using computer and internet-based knowledge, which is how we view the world today. With the introduction of AI however, the next projected stage is the 2nd information revolution and this phase is what we are currently working towards. We are still not confident in exactly what this next stage will consist of or how society will be affected, but we are in the midst of answering those questions. Figure 2.1 depicts a visualization of these stages in society. Figure 2.1

The scientific community is currently at odds over when the exact arrival time of artificial super intelligence, but they all agree that it is still reasonable enough to start discussing the ethics and morality surrounding this exciting field. Like you’d expect, the idea of ASI systems is very polarizing amongst the general public but also with some of the world’s top minds as shown below by Figure 3.1 Figure 3.1 *Results come from “Future Progress in Artificial Intelligence: A Survey of Expert Opinion” by Müller, Vincent C. and Bostrom, Nick

The most publicized sentiment in this debate is the idea that super intelligence will be the end of humanity and the society we recognize today (the existential catastrophe bit in Figure 2.1). The prevailing sentiment within this camp was summed up nicely by Elon Musk when he said that AI is “our greatest existential threat.” A large part of the reason that this general fear exists stems from pop culture’s depiction of AI in scenarios such as Terminator and countless sci-fi movies and novels. Once a system of this level were to be activated, it is more than likely that we would not get a chance to turn it off due to its potential as a dominant learning machine. Stephen Hawking himself remarked that he thought “the development of full artificial intelligence could spell the end of the human race.” Movements to get ahead of this potential threat have been growing in accordance to the general sense of foreboding and warning messages. When Google purchased the DeepMind system, part of the contract stipulated that a joint AI Ethics board be formed. In a similar vein, tech leader such as Sam Altman, Elon Musk, Peter Thiel, and others have joined efforts to create a research-driven safe AI group, OpenAI. The specific functions of this enterprise are somewhat vague (it is an emerging technology after all), but the purpose is clear; to work for safer artificial intelligence and push for pro active legislation (Dowd, 2017). The current fears in the artificial intelligence community are based on the idea that a super intelligent system would be too powerful to even try to fight. When experts are pressed on their views of how a system could end humanity, the response tends to be quite varied. One researcher, Eliezer Yudkowski elaborated on several possible scenarios of how AI could best humanity, including a possibility of a system being able to “solve the science technology of predicting protein structure from DNA information”. This scenario doesn’t seem too frightening until Yudkowski’s narrative concludes with the system creating “tiny invisible synthetic bacteria made of diamond, with tiny onboard computers, hiding inside your bloodstream and everyone else’s. And then, simultaneously, they release one microgram of botulinum toxin. Everyone just falls over dead.” This is just one of many paths that a malevolent super intelligent machine could take, but the underlying message is consistent; we must get ASI right the first time, because there is only a very slim chance that we will get a second attempt. The other side of this divide sees artificial intelligence as the next logical step in innovation and something to pursued relentlessly. A majority of the world’s leaders in creation and advancement including Google and Facebook are at the heart of this movement. Mark Zuckerberg actually directly downplayed Musk’s concerns by saying “Some people fear-monger about how A.I. is a huge danger, but that seems far-fetched to me and much less likely than disasters due to widespread disease, violence, etc.”  The long-term plan for some in this school of thought is an integration of biological intelligence and artificial intelligence in order to reach a higher level of society. This may sound far-fetched, yet for many experts it is the expected extension of today’s society. When confronted with fears from their opposing peers, the justification that ASG supporters use is essentially that it is the next step in society’s growth whether we like it or not. There is almost a sense of inevitability amongst those immersed in this field, a type of confirmation bias that may cloud their views on the potential negative effects of artificial super intelligence on society. There is a justification for the fervent desire that the supporters of this group have for pushing the envelope when it comes to artificial intelligence. Even with just narrow intelligence, the benefits that AI can provide to general society are hard to argue with. The field of big data has become pivotal to the way businesses run and markets expand and this industry is driven by AI engines that analyze large clusters of consumer data and allow decision makers to create connections and reach conclusions. Other examples of narrow AI being applied towards benefiting general society include fraud detection and cyber security, automated vehicles, and virtual assistants such as Apple’s Siri. Past these common applications, there are even more powerful applications for the higher levels of artificial intelligence. The engines that are approaching AGI in today’s society are being used for visual and auditory recognition, developing mechanical prototypes, and analyzing systems in industries ranging from finance to health care. Looking to the future, there are researchers who hypothesize that the solutions to some of the most difficult challenges facing humanity such as climate change and resource shortages will be resolved using the higher levels of AI in the future. There is no doubt that artificial intelligence has the potential to change the way the world operates for the better due to its sheer power and ability to connect information at the highest speed. Overall, artificial intelligence presents an exciting opportunity for humanity and society in many different ways. There are currently several different levels of AI, with differing levels of computing power and complexity. These levels are recognized as Artificial Narrow Intelligence, Artificial General Intelligence, and Artificial Super Intelligence. We are currently only scratching the surface of these systems, but we are nearing the point when regulations must start coming into the picture. The dilemma that is currently raging is a division between two groups. The first is those who would work slower and proactively regulate in order to mitigate risk, with another faction who believes in stepping forward and pushing the envelope towards reaching the vision of super intelligence at the fastest rate possible. With no clear answers in today’s climate, the only proper approach is to continue to invent and create with AI researcher Altman’s words of caution to temper us, “The next few decades we are either going to head toward self-destruction or toward human descendants eventually colonizing the universe.”