Introduction:
My two favorite things in the whole world, Artificial Intelligence and Government! What is their value, and what sort of interdisciplinary cooperation is necessary to make the world a better place for everybody? I decided to write this now because of a talk Elon Musk gave to a set of America’s Governors where he discussed his business’s plans, the global energy crisis, and the dangers of AI, here is the video. He mentioned again the need for a regulatory agency, or something similar, to ensure the creation of benevolent human-value aligned AI. Which is something I firmly believe is necessary and even want to make a career out of. In this piece I will highlight major advancements in technology and propose significant changes to the way we humans view public policy and governance. There are four sections: AI, Government, AI Helping Government, Government Helping AI. Some definitions and clarifications before I begin:
- AI is any intelligent software and/or hardware combination that operates somewhere on the scale of cognition between dust and humans but is not a living being.
- Government is in most cases meant to be the USA federal government, unless further specified. State institutions are often very similar in structure and thus can stand to benefit from the same recommendations.
- I’m not an expert in either of these fields yet, so read with scrutinous eyes. My opinion is not just my own, but is mine to own.
Let’s dive in!
AI:
Known to many as HAL 9000, Skynet, or Watson, artificial agents are forced to live with the cheesy names we humans give them long after their heyday. Now, two of those AI’s mentioned are fictional, the same two are also malevolent. The third, Watson, is real, and has already started saving lives, literally. The research and development of intelligent systems and algorithms has always been shrouded by public misunderstanding, mostly due to Hollywood’s sensationalism of terror inducing AI. Granted, many scenarios portrayed in film and television are not incredibly far fetched, but they are still too shaky a forecast of what is to come. One thing is for sure, the general populous needs to learn more about the improvements and breakthroughs happening today in artificial intelligence. Their livelihood could potentially be effected by a sweeping wave of new innovations targeting the most automatable parts of civilian life. As a paranoid silicon valley engineer once quoted, “I don’t know of a single job that doesn’t have some Y Combinator startup desperately trying to code it out of existence.” To understand the state of affairs, it’s important to break down the large field of AI and examine each subset’s value and momentum.
A good breakdown of the major components that make up today’s AI is provided by an article by Deloitte on AI’s potential impact on government, posted here. They are:
- Expert/Rule-Based Systems
- Natural Language Processing (Speech Recognition/Machine Translation)
- Computer Vision
- Robotics
- Machine Learning
Of course, there is often multiple pieces used at once by any one project, but these are the pieces themselves. I’ll mention briefly the importance of each and what is happening to develop them further.
Also, this has nothing to do with what is commonly referred to as Artificial General Intelligence or Artificial Super Intelligence, since those are theoretical states of intelligence. Hypothesizing the future state of a budding technology is both naive and dangerous. I present only information that is either in development currently or already created. However, this is not to say that the development of intelligent, potentially self-replicating software isn’t an incredible technology, it is and may be a top 5 human invention. Every powerful technology has its even share of potential harm and benefit. You just have to try to realize the full impact an enabling development like that has, then try to guarantee the maximum positive impact in reality.
So let’s talk about modern day AI.
Expert Systems:
Our knowledge of automated systems is almost entirely due to expert systems. In the early days of AI research many scientists thought they could create algorithms that worked in a specific domain by coding in a large breadth of expert knowledge. This was just before the second AI winter, a period of time where funding dropped off due to a lack of commercial viability. At the time (late 80’s) over 1 billion dollars was being invested into the promise that AI would be “solved” within a generation. These systems were not a complete bust, many just couldn’t process data quick enough to give a valuable answer. Others simply didn’t have enough knowledge programmed in. Most chat bots or automated telephone system are an example of these expert systems in service. Not many people enjoy a robot picking up the phone instead of a human being, but the fact of the matter is that millions of man hours and company dollars are saved every year because of these programs. So however unaesthetic these systems are, they demand respect because as much as you hate having to click through to a get a human assistant, you probably hate more having to wait on hold for hours.
Natural Language Processing (NLP):
Without a doubt, the most important human invention of all time was one of the first, language. When you consider language a man-made technology, it’s obvious how much you take it for granted. Being able to efficiently communicate ideas to each other is what fostered the growth of communities and enabled us as a species to speed up our own evolution. Written and spoken language set us apart from any other animal. All technological advancement ever would not have been possible without the verbalization of ideas, and then the passing along of those ideas to others in the form of writing. Now, thousands of years later, it seems we are finally going full circle with the mission of teaching computers how to communicate like humans. We are trying to teach a man-made technology how to use another man-made technology. NLP, generally, is the attempt to understand the rules and norms of natural human language. There are two main sub-fields: speech recognition and machine translation.
Speech recognition is the transcribing of spoken language to written language. The other way around isn’t as hard since almost all the rules for speech are written down for different languages. Speech recognition is difficult because humans often speak with ambiguity intentionally, meaning of words are implied with context. Context is never easy for computers to understand. Take sarcasm for instance, software is super great at understanding sarcasm. (lol jk)
Machine translation is the science of taking a set of text in one language and turning it into the same text in a different language. For small string lengths this is not as challenging, but holding context after translation of a large document is near impossible. Methods for these NLP tasks has changed over the years, originally using a bottom up approach of breaking down language into sounds and smaller sub-word structures then building sentences and paragraphs from those sounds or words. Nowadays, a lot of this is done with the help of machine learning off of human transcribed/translated data. This research is fairly dominated by Google, since a Google search really is just understanding a natural language query and returning the most desired information. Plus Google Translate is the most used translation service on the web. Recent developments are only a few months old, and their algorithms are even creating a new language to help understand them all?
Computer Vision:
Similar to how NLP is focused on hearing and understanding human ideas, computer vision is focused on seeing and understanding the world around us. This, also like NLP, relies heavily on machine learning for progress. Some older techniques utilized multiple types of cameras and sensors, but there is a big move in industry towards only using regular 3-color photographs. The idea that many researchers have in their head is, “If the brain and eyes can do it with lenses, rods, cones, and some back-end processing, so can we.” (sorta) Take for instance autonomous vehicles, in the beginning of their development, many companies achieved the feat with an array of LIDAR and motion tracking cameras. Now, many of the forerunners are using only cameras and learning off that data alone. Facebook has one of the largest steaks in this technology because it’s widely used as a image sharing platform (especially with the purchase of Instagram). They implement facial recognition software to auto tag your friends and you in pictures, and now are moving toward even more robust capabilities. Soon, labeling all sorts of objects in moving images will be possible. Thanks to the incredible ground work done at Stanford on a project called ImageNet, Facebook is leading a computer vision revolution.
Robotics:
This is truly where AI becomes visual. Advances in robot manufacturing to create human-like bodies for these intelligent systems may be a large step in the wrong direction, but not many people really want that anyway. Currently the cutting edge of AI and robotics is taking place on the eastern hemisphere. As western cultures continue to be timid accepting robots into daily life, countries like Japan are welcoming them. All over the developed world more people are getting older and living longer. The average age of many countries is trending up, this means that there is going to be a higher demand for care-givers for old people. In places where immigration is not an option, the only answer is robotics. Autonomous cars can also be considered robots. I don’t pay much attention to robotics because I think the most important AI will be almost entirely software based and on devices whose primary use is shared with the AI. Like phones.
Machine Learning:
Without a doubt machine learning will revolutionize the way we view intelligence and the way we think about learning. This sub-field can be broken down into smaller subsets: Supervised, Unsupervised, and Reinforcement Learning. Each of these has their applications, whether it’s grouping similar things (clustering), or classifying objects (classification), machine learning is all about taking data and recognizing patterns. If the data is labeled, the algorithm has some insight to learn off of. If there are no labels, the algorithm seeks to gain insight from realizing all examples. What I think is the most potentially groundbreaking sub-field is deep reinforcement learning. This is where an artificial neural network is trained on live streaming data and is tasked with some optimization function. When you think about human learning abstractly, this seems nearly exactly the same. Advancements in this field are shared by almost all tech companies since the boom of deep learning in 2016. But the company with the most gusto right now would probably be Deepmind, since they started the craze with AlphaGo. These people are pushing to create AGI, artificial general intelligence, and they are very clearly on the right path.
AI scientists are feverishly innovating, trying to build on past inventions to create new marketable algorithms. Different companies are interested in similar tools, so there is some communication, but not much. Some companies are starting to request patents on certain machine learning technologies. I’m all for science and the discovery of new tools and technologies, but there seems to be something missing, right . . . ethics. There are many engineers working very hard to create incredible systems, so many that the philosophers are having a hard time keeping up.
AI Ethics:
There are many non-profit organisations working toward the goal of formalizing AI research and creating a set of standards to govern development. There are a lot more thorny problems when it comes to the development of AI software, many of them centering around what sort of minimizing/maximizing function is morally right and aligned with human values. But much of the AI ethics discussion is also dominated by the idea of mass human unemployment. I’ll talk more about this is a later section, but let me just say this now, we as a society need to talk about what is right and wrong when it comes to introducing a new technology to this world. AI has the potential to both destroy and beautify this universe, the more people dedicated to finding how each scenario plays out the better.
Government:
Never before has American government been so broken, corrupt, and fraudulent. It’s time for a new new deal. There needs to be a huge focus shift. Right now, the administration and congress are more focused on increasing the private sector’s good, than the public sector. Our president would rather guarantee the success of a wealthy coal CEO than the livelihood of millions of unborn children. And our congress would rather fill the pockets of their allies, than develop law for the public good. Money speaks many volumes higher than sense does in DC. And all trust in the system and its contributors has been dropping significantly since January 2017. The american people were desperate for change, sick of politician’s lies and failed promises, but in the end were all conned. Both parties are to blame, the DNC ruined their chances when they unjustly snuffed out Bernie, and the GOP should have never given Trump the light of day. Now everyone suffers, everyone but the already rich. Wealth disparity will only get worse in a Trump America.
Alright, rant aside government has a couple main functions that everybody agrees on. Government is needed to defend and protect its citizens, create fair laws and just punishments, and foster the community of a nation by investing in transportation, education, agriculture, science, and many more. The reason why every kid growing up in the 60’s and 70’s wanted to be an astronaut was because we put a man on the moon. Government should fund inspiring sciences, it’s odd to think about, but wouldn’t it be nice if our government also provided motivation for young people? Kids nowadays don’t dream as big because they are not exposed on a grand scale to anything other than celebrities and pro atheletes. But why does everything seem so broken anyway? Lets explore that question.
Misrepresentation:
Probably the most obvious reason for the apparent disconnect between government officials and their constituents is misrepresentation. Our 115th US congress, the 535 elected officials that are supposed to represent the over 330 million people living in America, are not like us. Lets just look at gender, race, and age for a moment. Here are Congress’ stats: 80% male, 20% female, 7% Hispanic, 9% African-American, 3% Asian-American, 81% white, average age: 58 years old. Here are America’s stats: 50% male, 50% female, 16% Hispanic, 12% African-American, 5% Asian-American, 64% white, average age: 38 years old. If those numbers aren’t enough here are the numbers on wealth: members of congress average net worth is just over 1 million dollars when the average for american households is around 60 thousand dollars. These are only a few statistics showing the lack of diversity and misrepresentation, others such as religion and education draw a larger divide. I think it is important to have wise individuals creating laws for a society, but those wise individuals need to be nested in reality. Perfect democratic representation is probably impossible, but what we have is nowhere even close.
Agency Expansion:
The framers of the constitution would probably not be super stoked about how the government is looking nowadays. Separation of powers is shot. Everybody complains about bureaucratic red-tape and how slow things move in government, but it’s no surprise when you look at how complicated the agency landscape is. I mean, have you ever even heard of Federal Retirement Thrift Investment Board? The president has more power than ever before, having the capability to nominate/replace individuals for all of these major government rolls. If a president were truly corrupt, they would just nominate all their friends to these rolls once elected, or even worse the people who donated the most during the election. Oh wait, shit. I’m not saving that all government agencies are a bad thing, but I do think that there is a lot of fat to be trimmed in the administration and its agencies. The power that they have is derived from the executive branch, but the executive branch should have the least power.
Spending:
One of the most difficult things that our government has to do is figure out who gets what share of the taxes. Luckily the congress can’t raise their own paychecks easily, but they do need to decide who gets what. Remember when the government shutdown and everybody lost their mind in 2013? That’s because they couldn’t agree to where the money went (house republicans hated ACA). They couldn’t vote on an fair budget. As hard as I know this process to be, I think it could be made significantly easier if one thing changed: de-funding defense. Every year the “defense” budget increases. America is a country that was built off of conflict, the only reason why we are considered a world power is because we didn’t go bankrupt fighting the Nazi’s on our own turf. Ever since the end of WWII and the dissolution of Secretary of War into the Secretary of Defense, money flow has increased. Over half of the US federal discretionary spending is spent on the military, more than education, agriculture, science, technology, energy and transportation combined. NASA and the EPA are being decimated under the Trump Administration. The majority of taxpayer dollars goes into programs like medicare, medicaid and social security, I recognize, but those reapportionment’s deserve another conversation entirely. Seriously, when 17 billion dollars is less than a percent of the US budget, I’m sure the Military can do without its total of 13%.
Reactivity:
The final point I wanted to bring up when talking about the functionality of the american federal system is the way it makes change. For all of it’s history, the government has enacted a majority of reactive policies. For instance, the Federal Aviation Administration didn’t exist until over 50 years after the first manned flight. (However there was small regulation committee created in the late 20s) It’s practically impossible for the federal judiciary to act proactively since they can only make rulings on cases that are brought to them. The congress and executive branches have more potential to pass and influence proactive legislature. The reason why I am bringing this up is simple, for government to stay useful, they need to be adaptable. In this age of incredibly fast technological development the american government has had to start acting proactively to maintain it’s competitive advantage over adversaries. I mean, take even the space race and nuclear arms race for instance. Nowadays most of government proactivity is seen in cyber warfare. Since before NATO added cyber to the domains of war America has been attacking and defending itself from cyber attacks. Many times throughout America’s history there has been a necessity to use or regulate a technology that is in it’s early stages of development. The ability to recognize this before full development has always given America an advantage. Technology isn’t the only thing that is moving faster, all of human life seems to be speeding up. So either the government accelerates its functions tremendously, or they start devoting more time to creating more proactive policy, or both hopefully. Technologies like renewable energy, advanced bio-engineering, and AI need proactive policy, whether that means grants or regulations depends on the benefits and dangers.
AI Helping Government:
Both state and national government could use some work when it comes to efficiency. In the Deloitte article I linked earlier, the one that highlighted the different genre’s of modern day AI, they talk at length of ways we could utilize software to speed up government processes. Not only could many processes be sped up with new technology, but many services could be partially automated with already built technology. There has been a trend in the private sector that sees all sorts of different companies transitioning towards a more tech dense environment modelling the numerous software companies that have utilized their computer science and business expertise to create the most efficient workplace possible. The government should do the same. One of Deloitte’s studies followed a child protection service’s employee and found out that more than a third of their day was spent just on paperwork. Having an AI-augmented government would speed up processes and save the taxpayer money.
This may be the the only positive note coming out of the woodwork of the Trump Administration, the new formation of the American Technology Council. This council is made up of some of the most famous Silicon Valley CEOs with the mission of revolutionizing the government’s information technology and cybersecurity. But it’s going to take a lot more than just the top 20 tech CEOs to take care of such a huge problem. There needs to be a full agency or department dedicated to exploring each federal position and whether or not it would be viable, valuable, and vital to automate a portion of it. This agency would not be viewed as the bad guys, they would not be taking people’s jobs away. In fact, they would be seeking solutions to problems that these workers have voiced for years. They would be giving them their job back in a sense, because most jobs would have more time to dedicate to “mission-aligned” tasks.
GovTech Singapore:
Only one nation has an agency dedicated to bringing it’s government into the information age: Singapore. Here’s a quote from the man in charge of the agency, “How do we look at every aspect of our lives and our public sector services and make sure that where there’s an opportunity for that service, product, or experience to be enabled by technology we do so maximally. Where there’s an opportunity to reduce friction and increase efficiency, we do so maximally.” I love this quote, spoken like a true scientist. Singapore is a small city-state-nation near Malaysia and Indonesia, and have only been a country for a little over 50 years. But just in that small amount of time they have transitioned from an undeveloped nation into one of the world leaders in education, health care, and quality of life. They even just passed America on the UN Human Development Index. But how could they do all this in such a short amount of time? Some experts say it’s the pride of the people, others say it’s due to the increasing acceptance and use of technology. While I’m sure it’s a combination of the two, I care more about how they used technology to spring forward past many nations 10 times as old. How did they make this transition in one generation? By caring about the kids. The government has invested large sums into teaching young children the important skills they need for a successful life in the 20th century. They learn sequencing and coding from preschool on, with the curriculum getting more and more complex as they grow older. Even more so, they consider immediate student feedback. If Singapore is different from every other nation because of one thing, it would be their insentient desire to improve the living environment of generations to come. By creating this agency, Singapore hopes to bring positive data analytics and information processing automation to their public sector employees. Not only improving the quality of life of the workers, but also all of the citizens that they interface with. Then they instill their future leaders with the knowledge they need to continue to improve the society. Doesn’t sound that hard right? Well, Singapore only has 5.5 million inhabitants, so there are some scaling problems if adapted to America.
Federal Technology Agency:
I propose that a new agency be created under the president, this agency will be tasked with a similar mission as GovTech in Singapore: to maximize efficiency in government jobs with the proper implementation of technology. This agency would report to the president on the current state of every other department’s information technology landscape, and where improvement can be made. This agency can then recommend, to other agencies, actions that can be taken to reduce friction in operations and alleviate stress on behalf of the workers. More over, the Federal Technology Agency would work alongside other government technology organizations such as the OSTP to research further advancements in technology and promote development competition. This agency will use it’s budget to pay its workers, and auction contracts to companies that can create and maintain technologies devoted to augmenting government processes. Similar to what this company does. We as a society are entering a new digital age, one where information is bountiful but often neglected. If not neglected we as a people can transcend the drudgery of data and work with what we are naturally better at understanding, people. This agency would create a metric calculating potential technological impact to determine which department or agency needs the most improvement. From there, they can start working on how to relieve pressure. This would be a medium to large agency, comprising of executives, managers, automation engineers, economists, human resource managers, and many other fields so that each department scrutinized by the agency would have a fair judgement on what could improve. The short term goal of the agency would be to revamp the government’s IT. The long term goal would be to aide research and development of better, more augmenting technologies. Someday, no one will have to work, but until that day let’s make everybody’s job easier and more fun. Not to mention the amount of money this agency would be saving all of the taxpayers.
Government Helping AI:
For the content you probably skipped all the other stuff to read, you first have to read this: government needs AI to work more efficiently, but government also needs to control the future development of AI. When I say control, I don’t mean that we need another Manhattan Project for AI. First and foremost, artificial intelligence should never be utilized as a weapon, it may augment warfare, but never used directly as a weapon. There are many fundamental problems with artificially intelligent systems, questions concerning who has control and what are good motives, that haven’t been answered correctly yet. Of course this is a new field, but there is a huge problem with this field: money. The problem isn’t a lack of money, but the over incentivization of money. Companies like Google and Amazon are duking it out right now trying to create even more cutting edge software to outperform competitors. As much as I am a capitalist, I’m also a realist, and that sort of fuel for development could lead to unforeseen accidents. You’d think after the 2010 high frequency trading crash everyone would realize that making AI compete against each other is a recipe for disaster. But just like how the mutual funds kept competing, these tech companies will too: because it’s a business. There are two ways the federal government can help the development of benevolent AI: grants and regulations.
Grants:
A lot of the funding for artificial intelligence development comes from the already wealthy silicon valley CEOs. Those who see its value and understand the power that comes with it. This needs to change. As everybody in the industry has noted, research is moving from academia to the private sector. This to me has a lot more negative affects than positive ones. Academia research has always been more transparent than private company research. When the secrets have a dollar sign attacked to it, they are much more important to keep. But that is the opposite of what we need when we are moving forward with the develop of AI. I propose that the american government greatly increase the budget for science and technology research and use it to not only contract companies to build software for the government, but also provide more grants to promising AI research. This could be done under the umbrella of the previously proposed new Federal Technology Agency. With the money also comes an understanding that the government wants to know what you’re up to. These grants would not just be buying a ticket to the magic show, but also a behind the scenes VIP explanation of all the tricks done. The agency should have the expertise available to understand the magic and relay that information to others and make the developments understandable to all. Grants would be given to research groups in both academia and the private sector. These grants will help the research organization hire more talent, and motivate the completion of projects. It helps the government by providing the agency with much needed valuation of progress. The agency will already have a basic road map of where it wants research to go and where not. If any grant recipient starts along the wrong path, the government will know about it sooner rather than later, or at least before it’s too late. Currently the only funding for research given out by the federal government is under the National Institutes of Health and the National Science Foundation. There needs to be more than just grants.gov for us as a nation to improve faster, and we need to keep up.
Regulation:
Nobody likes to be controlled, I get that, but nobody wants to die to nuclear fallout or an avoidable plane crash. Consumers expect their products and services to be safe. So when a company lets them down, it really hurts. As many people, perhaps, already have a unfair negative outlook on AI, making sure that companies deploying artificial agents do so safely and wisely is of the government’s utmost concern. No technology or research needs to be stifled when good regulation is done. If anything, regulation should focus organizations on what is the fair way of operating. Regulation is mainly necessary in the field of AI because of the danger it poses to the general public. The danger is not yet realized by the public, and even worse the engineers. Software is biased, because humans are biased. It’s going to take a lot of work to insure the creation of an unbiased, secure system. Organizations like the Future of Life Institute are actively thinking about how we as a society can promote the development of beneficial AI. They recently held a conference of AI research elites that had a goal of creating a list of “must-follow” principles voted yes on by at least 90% of the attendance. This list can be found here. It outlines three areas of principles: research issues, ethics and values, and long-term issues. I encourage you all to read them carefully because they are obviously well thought out. Each is very important, but I choose to focus on numbers 3, 4, and 5. Together these principles say that there should be a positive communication channel between government and researchers as well as between all research organizations. Most importantly it says that these research and development organizations should avoid an AI arms race. Number 5 references the avoidance of cutting corner on safety standards but fails to recognize any agreed upon set of standards. Hmm, sounds like everybody wants a set of standards to hold each other accountable to, I wonder who would make that? Oh right, the government. These scientists want the government’s help formulating a set of AI research and development standards. So we should do it. Who does it? Maybe it be under the Federal Technology Agency as a Technology Standards Committee. Regardless, the first step for regulators of this industry would be to gain useful insights into the current developments happening around the globe and attempt to measure the rate of advancement. Then and only then can a set of standards be made, and not without the help of all the current researchers. Ideally, whoever is creating these standards has had numerous years of experience in industry designing these systems. Alongside the creation of standards is the creation of enforcement policy. Whether it be the pulling of grant money or, even more extreme, the shutting down of a project or company, this regulatory agency must have the power to resign any project that it deems potentially a great hazard to human life. This would be a tough decision to make by any regard, but having some metric for how positively human-aligned a program is without a doubt would be vital to this survey. Many steps need to be taken to insure public protection from security threats and potential accidents brought about by the uncareful development of AI. With a government agency dedicated to this subject, both policy makers and research scientists can be aware of the paths of AI and their entailed dangers/benefits.
Conclusion:
So, AI is advancing without any signs of stopping, government is bogged down with paperwork and focusing on the wrong things to advance, and no one in government seems to be doing anything about either. The Obama administration at least made a report on the subject, but the few steps he took forward towards positive technology-policy relationships have all but been taken back by the new administration. The government needs to re-imagine the way they do their job, and they need to be comfortable allowing technology to augment their professions. Because with the size of our population, these clerical problems are only going to compound. A new agency needs to be created with the task of providing government with viable and valuable answers to the increasing need for more technology in the workplace. Also, the government needs to do more work to progress the cutting edge of AI technology through grants and regulation. With this sort of agency, billions of taxpayer’s dollars would be saved in man hours, and the potential harms of revolutionary technology like AI can be managed securely. These propositions that I’ve outlined could be realized differently in government. Perhaps instead of an entirely new organization being created, a large subset of an already created agency can be devoted to these practices. But I think it’s fair to say that such an organization, one that both seeks to improve government operations and provide grants and regulations to emerging technologies, would be quite large and in need of a good portion of the US federal budget (% 0.5 maybe). Having those qualities make me want to envision it as a full fledged agency.