It would be quite difficult \u2013 some might say impossible \u2013 to design broadly adopted ethical AI systems. A share of the experts responding noted that ethics are hard to define, implement and enforce. They said context matters when it comes to ethical considerations. Any attempt to fashion ethical rules generates countless varying scenarios in which applications of those rules can be messy. The nature and relative power of the actors in any given scenario also matter. Social standards and norms evolve and can become wholly different as cultures change. Few people have much education or training in ethics. Additionally, good and bad actors exploit loopholes and gray areas where ethical rules aren\u2019t crisp, so workarounds, patches or other remedies are often created with varying levels of success.<\/p>\n\n\n\n
The experts who expressed worries also invoked governance concerns. They asked: Whose ethical systems should be applied? Who gets to make that decision? Who has responsibility to care about implementing ethical AI? Who might enforce ethical regimes once they are established? How?<\/p>\n\n\n\n
A large number of respondents argued that geopolitical and economic competition are the main drivers for AI developers, while moral concerns take a back seat. A share of these experts said creators of AI tools work in groups that have little or no incentive to design systems that address ethical concerns.<\/p>\n\n\n\n
Some respondents noted that, even if workable ethics requirements might be established, they could not be applied or governed because most AI design is proprietary, hidden and complex. How can harmful AI \u201coutcomes\u201d be diagnosed and addressed if the basis for AI \u201cdecisions\u201d cannot be discerned? Some of these experts also note that existing AI systems and databases are often used to build new AI applications. That means the biases and ethically troubling aspects of current systems are being designed into the new systems. They say diagnosing and unwinding the pre-existing problems may be difficult if not impossible to achieve.<\/p>\n\n\n\n
A portion of these experts infused their answers with questions that amount to this overarching question: How can ethical standards be defined and applied for a global, cross-cultural, ever-evolving, ever-expanding universe of diverse black-box systems in which bad actors and misinformation thrive?<\/p>\n\n\n\n
A selection of respondents\u2019 comments on this broad topic is organized over the next 20 pages under these subheadings: 1) It can be hard to agree as to what constitutes ethical behavior. 2) Humans are the problem: Whose ethics? Who decides? Who cares? Who enforces? 3) Like all tools, AI can be used for good or ill, which makes standards-setting a challenge. 4) Further AI evolution itself raises questions and complications.<\/p>\n\n\n\n
Stephen Downes,<\/strong> senior research officer for digital technologies with the National Research Council of Canada, observed, \u201cThe problem with the application of ethical principles to artificial intelligence is that there is no common agreement about what those are. While it is common to assume there is some sort of unanimity about ethical principles, this unanimity is rarely broader than a single culture, profession or social group. This is made manifest by the ease with which we perpetuate unfairness, injustice and even violence and death to other people. No nation is immune.<\/p>\n\n\n\n “The problem with the application of ethical principles to artificial intelligence is that there is no common agreement about what those are. While it is common to assume there is some sort of unanimity about ethical principles, this unanimity is rarely broader than a single culture, profession or social group.”<\/p>STEphen Downes, senior research officer for digital technologies with the National Research Council of Canada, <\/cite><\/blockquote><\/figure>\n\n\n\n \u201cCompounding this is the fact that contemporary artificial intelligence is not based on principles or rules. Modern AI is based on applying mathematical functions on large collections of data. This type of processing is not easily shaped by ethical principles; there aren\u2019t \u2018good\u2019 or \u2018evil\u2019 mathematical functions, and the biases and prejudices in the data are not easily identified nor prevented. Meanwhile, the application of AI is underdetermined by the outcome; the same prediction, for example, can be used to provide social support and assistance to a needy person or to prevent that person from obtaining employment, insurance or financial services.<\/p>\n\n\n\n \u201cUltimately, our AI will be an extension of ourselves, and the ethics of our AI will be an extension of our own ethics. To the extent that we can build a more ethical society, whatever that means, we will build more ethical AI, even if only by providing our AI with the models and examples it needs in order to be able to distinguish right from wrong. I am hopeful that the magnification of the ethical consequences of our actions may lead us to be more mindful of them; I am fearful that they may not.\u201d<\/p>\n\n\n\n Kenneth A. Grady<\/strong>,<\/strong> adjunct professor at Michigan State University College of Law and editor of The Algorithmic Society<\/a> on Medium, said, \u201cGetting those creating AI to use it in an \u2018ethical\u2019 way faces many hurdles that society is unlikely to overcome in the foreseeable future. In some key ways, regulating AI ethics is akin to regulating ethics in society at large. AI is a distributed and relatively inexpensive technology. I can create and use AI in my company, my research lab or my home with minimal resources. That AI may be quite powerful. I can unleash it on the world at no cost.<\/p>\n\n\n\n \u201cAssuming that we could effectively regulate it, we face another major hurdle: What do we mean by \u2018ethical?\u2019 Putting aside philosophical debates, we face practical problems in defining ethical AI. We do not have to look far to see similar challenges. During the past few years, what is or is not ethical behavior in U.S. politics has been up for debate. Other countries have faced similar problems.<\/p>\n\n\n\n \u201cEven if we could decide on a definition [for ethics] in the U.S., it would likely vary from the definitions used in other countries. Given AI\u2019s ability to fluidly cross borders, regulating AI would prove troublesome. We also will find that ethical constraints may be at odds with other self-interests. Situational ethics could easily arise when we face military or intelligence threats, economic competitive threats, and even political threats.<\/p>\n\n\n\n \u201cFurther, AI itself presents some challenges. Today, much of what happens in some AI systems is not known to the creators of the systems. This is the black-box problem. Regulating what happens in the black box may be difficult. Alternatively, banning black boxes may hinder AI development, putting our economic, military or political interests at risk.\u201d<\/p>\n\n\n\n Ryan Sweeney<\/strong>, <\/strong>director of analytics for Ignite Social Media, commented, \u201cThe definition of \u2018public good\u2019 is important here. How much does intent versus execution matter? Take Facebook, for instance. They might argue that their AI content review platform is in the interest of \u2018public good,\u2019 but it continues to fail. AI is only as ethical and wise as those who program it. One person\u2019s racism is another\u2019s free speech. What might be an offensive word to someone might not even be in the programmer\u2019s lexicon.<\/p>\n\n\n\n \u201cI\u2019m sure AI will be used with ethical intent, but ethics require empathy. In order to program ethics, there has to be a definitive right and wrong, but situations likely aren\u2019t that simple and require some form of emotional and contextual human analysis. The success of ethical AI execution comes down to whether or not the programmers literally thought of every possible scenario. In other words, AI will likely be developed and used with ethical intent, but it will likely fall short of what we, as humans, can do. We should use AI as a tool to help guide our decisions, but not rely on it entirely to make those decisions. Otherwise, the opportunity for abuse or unintended consequences will show its face. I\u2019m also sure that AI will be used with questionable intent, as technology is neither inherently good nor bad. Since technology is neutral, I\u2019m sure we will see cases of AI abused for selfish gains or other questionable means and privacy violations. Ethical standards are complicated to design and hard to program.\u201d<\/p>\n\n\n\n Below is a sampling of expert answers that speak to the broad concerns that ethical behaviors can be hard to define and even more difficult to build into AI systems.<\/p>\n\n\n\n Mark Lemley<\/strong>,<\/strong> director of Stanford University\u2019s Program in Law, Science and Technology, observed, \u201cPeople will use AI for both good and bad purposes. Most companies will try to design the technology to make good decisions, but many of those decisions are hard moral choices with no great answer. AI offers the most promise in replacing very poor human judgment in things like facial recognition and police stops.\u201d<\/p>\n\n\n\n Marc Brenman,<\/strong> managing member at IDARE, a transformational training and leadership development consultancy based in Washington, D.C., wrote, \u201cAs societies, we are very weak on morality and ethics generally. There is no particular reason to think that our machines or systems will do better than we do. Faulty people create faulty systems. In general, engineers and IT people and developers have no idea what ethics are. How could they possibly program systems to have what they do not? As systems learn and develop themselves, they will look around at society and repeat its errors, biases, stereotypes and prejudices. We already see this in facial recognition.<\/p>\n\n\n\n \u201cAI will make certain transactions faster, such as predicting what I will buy online. AI systems may get out of control as they become autonomous. Of what use are humans to them? They may permit mistakes to be made very fast, but the systems may not recognize the consequences of their actions as \u2018mistakes.\u2019 For example, if they maximize efficiency, then the Chinese example of social control may dominate.<\/p>\n\n\n\n \u201cWhen AI systems are paired with punishment or kinetic feedback systems, they will be able to control our behavior. Imagine a pandemic where a \u2018recommendation\u2019 is made to shelter in place or wear a mask or stay six feet away from other people. If people are hooked up to AI systems, the system may give an electrical shock to a person who does not implement the recommendation. This will be like all of us wearing shock collars that some of us use on our misbehaving dogs.\u201d<\/p>\n\n\n\n June Anne English-Lueck<\/strong>,<\/strong> professor of anthropology at San Jose State University and a distinguished fellow at the Institute for the Future, said, \u201cAI systems employ algorithms that are only as sound as the premises on which they are built and the accuracy of the data with which they learn. Human ethical systems are complex and contradictory. Such nuances as good for whom and bad for whom are difficult to parse. Smart cities, drawing on systems of surveillance and automated government need mechanisms of human oversight. Oversight has not been our strong suit in the last few decades, and there is little reason to believe it will be instituted in human-automation interactions.\u201d<\/p>\n\n\n\n Amali De Silva-Mitchell<\/strong>,<\/strong> a futurist and consultant participating in multistakeholder, global internet governance processes, wrote, \u201cAlthough there are lots of discussions, there are few standards or those that exist are at a high level or came too late for the hundreds of AI applications already rolled out. These base AI applications will not be reinvented, so there is embedded risk. However, the more discussion there is, the greater the understanding of the existing ethical issues, and that can be seen to be developing, especially as societal norms and expectations change. AI applications have the potential to be beneficial, but the applications have to be managed so as not to cause unintended harms. For global delivery and integrated service, there needs to be common standards, transparency and collaboration. Duplication of efforts is a waste of resources.\u201d<\/p>\n\n\n\n Glenn Edens<\/strong>, <\/strong>professor at Thunderbird School of Global Management, Arizona State University, previously a vice president at PARC, observed, \u201cThe promise: AI and ML could create a world that is more efficient, wasting less energy or resources providing health care, education, entertainment, food and shelter to more people at lower costs. Being legally blind, I look forward to the day of safe and widely available self-driving cars, for example. Just like the steam engine, electricity, bicycles and personal computers (especially laptops) amplify human capacity AI, and ML hopefully will do the same.<\/p>\n\n\n\n \u201cThe concerns: AI and its cousin ML are still in their infancy \u2013 and while the technology progress is somewhat predictable, the actual human consequences are murky. The promise is great \u2013 so was our naive imagination of what the internet would do for humankind. Commercial interests (and thus their deployment of AI and ML) are far more agile and adaptable than either the humans they supposedly serve or the governance systems. Regulation is largely reactionary, rarely proactive \u2013 typically, bad things have to happen before frameworks to guide responsible and equitable behavior are written into laws, standards emerge or usage is codified into acceptable norms. It is great that the conversation has started; however, there is a lot of ongoing development in the boring world of enterprise software development that is largely invisible.<\/p>\n\n\n\n \u201cCredit scoring comes to mind as a major potential area of concern \u2013 while the credit-scoring firms always position their work as providing consumers more access to financial products, the reality is that we\u2019ve created a system that unfairly penalizes the poor and dramatically limits fair access to financial products at equitable prices. AI and ML will be used by corporations to evaluate everything they do and every transaction, rate every customer and their potential (value), predict demand, pricing, targeting as well as their own employees and partners \u2013 while this can lead to efficiency, productivity and creation of economic value, a lot of it will lead to segmenting, segregation, discrimination, profiling and inequity. Imagine a world where pricing is different for everyone from one moment to the next, and these predictive systems can transfer huge sums of value in an instant, especially from the most vulnerable.\u201d<\/p>\n\n\n\n A<\/strong> strategy and planning expert<\/strong> responded, \u201cWhile I say and believe that, yes, ethical boundaries will be put in place for AI by 2030, I also realize that doing this is going to be incredibly difficult. The understanding of what an AI is doing as it builds and adapts its understandings and approaches rather quickly gets to a point where human knowing and keeping up gets left behind. The how and why something was done or recommended can be unknowable. Also, life and the understanding of right and wrong or good-ish and bad-ish can be fluid for people, as things swing to accommodate the impacts on the human existence and condition as well as livable life on our planet. Setting bounds and limitations has strong value, but being able to understand when things are shifting out of areas that are comfortable or have introduced a new realization for a need to correct for unintended consequences is needed. But bounds around bias need to be considered and worked through before setting ethical limitations in place.\u201d<\/p>\n\n\n\n A<\/strong> vice president at a major global company<\/strong> wrote, \u201cAI is too distributed a technology to be effectively governed. It is too easily accessible to any individual, company or organization with reasonably modest resources. That means that unlike, say, nuclear or bioweapons, it will be almost impossible to govern, and there always will be someone willing to develop the technology without regard to ethical consequences.\u201d<\/p>\n\n\n\n Wendy M. Grossman,<\/strong> a UK-based science writer, author of \u201cnet.<\/em>wars<\/a>\u201d and founder of the magazine The Skeptic, predicted, \u201cThe distribution of this will be uneven. I\u2019ve just read Jane Mayer\u2019s piece in The New Yorker on poultry-packing plants, and it provides a great example of why it\u2019s not enough to have<\/em> laws and ethics; you must enforce<\/em> them and give the people you\u2019re trying to protect sufficient autonomy to participate in enforcing them. I think ethical\/unethical AI will be unevenly distributed. It will all depend on what the society into which the technology is being injected will accept and who is speaking. At the moment, we have two divergent examples:<\/p>\n\n\n\n But then what we have today isn\u2019t AI as originally imagined by the Dartmouth group. We are still a very long way from any sort of artificial general intelligence with any kind of independent autonomy. The systems we have depend for their ethics on two things: access to the data necessary to build them and the ethics of the owner. It isn\u2019t AI that needs ethics, it\u2019s the owners<\/em>.\u201d<\/p>\n\n\n\n Glynn Rogers<\/strong>, <\/strong>retired, previously senior principal engineer and a founding member at the CSIRO Centre for Complex Systems Science, said, \u201cAI and its successors are potentially so powerful that we have no choice but to ensure attention to ethics. The alternative would be to hand over control of our way of life to a class of developers and implementors that are either focused on short-term and shortsighted interests or who have some form of political agenda particularly \u2018state actors.\u2019 The big question is how to ensure this. A regulatory framework is part of the answer, but I suspect that a major requirement is to change the culture of the AI industry. Rather than developing technologies simply for the sake of it, or to publish clever papers, there needs to be a cultural environment in which developers see as an inherent part of their task to consider the potential social and economic impacts of their activities and an employment framework that does not seek to repress this. Perhaps moral and political philosophy should be part of the education of AI developers.\u201d<\/p>\n\n\n\n Alexandra Samuel<\/strong>, <\/strong>technology writer, researcher, speaker and regular contributor to the Wall Street Journal and Harvard Business Review, wrote, \u201cWithout serious, enforceable international agreements on the appropriate use and principles for AI, we face an almost inevitable race to the bottom. The business value of AI has no intrinsic dependence on ethical principles; if you can make more money with AIs that prioritize the user over other people, or that prioritize business needs over end users, then companies will build AIs that maximize profits over people. The only possible way of preventing that trajectory is with national policies that mandate or proscribe basic AI principles, and those kinds of national policies are only possible with international cooperation; otherwise, governments will be too worried about putting their own countries\u2019 businesses at a disadvantage.\u201d<\/p>\n\n\n\n Valerie Bock, <\/strong>VCB Consulting, former Technical Services Lead at Q2 Learning, commented, \u201cI don\u2019t think we\u2019ve developed the philosophical sophistication in the humans who design AI sufficiently to expect them to be able to build ethical sophistication into their software. Again and again, we are faced with the ways our own unconscious biases pop up in our creations. It is turning out that we do not understand ourselves or our motivations as well as we would like to imagine we might. Work in AI helps lay some of this out for us, aiding us in a quest [that] humanity has pursued for millennia. A little humility based on what we are learning is in order.\u201d<\/p>\n\n\n\n “I don\u2019t think we\u2019ve developed the philosophical sophistication in the humans who design AI sufficiently to expect them to be able to build ethical sophistication into their software.”<\/p>Valerie Bock, VCB Consulting, former Technical Services Lead at Q2 Learning<\/cite><\/blockquote><\/figure>\n\n\n\n The<\/strong> director of a military center for strategy and technology<\/strong> said, \u201cMost AI will attempt to embed ethical concerns at some level. It is not clear how \u2018unbiased\u2019 AI can be created. Perfectly unbiased training datasets don\u2019t exist, and, due to human biases being an inherent part of interactions, such a goal may be unobtainable. As such, we may see gender or racial biases in some training datasets, which will spill over into operational AI systems, in spite of our efforts to combat this.\u201d<\/p>\n\n\n\n Alan S. Inouye<\/strong>, <\/strong>director of the Office for Information Technology Policy at the American Library Association, responded, \u201cI don\u2019t see people or organizations setting out in a nefarious path in their use of AI. But of course, they will use it to advance their missions and goals and, in some sense, employ \u2018local\u2019 ethics. But ethics is neither standardized nor additive across domains. What is ethics across AI systems? It is like asking, \u2018What is cybersecurity across society?\u2019\u201d<\/p>\n\n\n\n Maggie Jackson<\/strong>, <\/strong>former Boston Globe columnist and author of \u201cDistracted: Reclaiming Our Focus in a World of Lost Attention<\/a>,\u201d wrote, \u201cI am deeply concerned by how little we understand of what AI algorithms know or how they know it. This black-box effect is real and leads to unintended impact. Most importantly, in the absence of true understanding, assumptions are held up as the foundation of current and future goals. There should be far greater attention paid to the hidden and implicit value systems that are inherent in the design and development of AI in all forms. An example: robot caregivers, assistants and tutors are being increasingly used in caring for the most vulnerable members of society despite known misgivings by both scientist\u2013roboticists, ethicists and users, potential and current. It\u2019s highly alarming that the robots\u2019 morally dubious fa\u00e7ade of care is increasingly seen as a good-enough substitute for the blemished yet reciprocal care carried out by humans.<\/p>\n\n\n\n \u201cNew ethical AI guidelines that emphasize transparency are a good first step in trying to ensure that care recipients and others understand who\/what they are dealing with. But profit-driven systems, the hubris of inventors, humans\u2019 innate tendency to try to relate to any objects that seem to have agency, and other forces combine to work against the human skepticism that is needed if we are to create assistive robots that preserve the freedom and dignity of the humans who receive their care.\u201d<\/p>\n\n\n\n Alan D. Mutter<\/strong>,<\/strong> a consultant and former Silicon Valley CEO, said, \u201cAI is only as smart and positive as the people who train it. We need to spend as much time on the moral and ethical implementation of AI as we do on hardware, software and business models. Last time I checked, there was no code of ethics in Silicon Valley. We need a better moral barometer than the NASDAQ index.\u201d<\/p>\n\n\n\n Fred Baker<\/strong>,<\/strong> board member of the Internet Systems Consortium and longtime IETF leader, commented, \u201cI would like to see AI be far more ethical than it is. That said, human nature hasn\u2019t changed, and the purposes to which AI is applied have not fundamentally changed. We may talk about it more, but I don\u2019t think AI ethics will ultimately change.\u201d<\/p>\n\n\n\n Randall Mayes<\/strong>,<\/strong> a technology analyst at TechCast Global, observed, \u201cThe standardization of AI ethics concerns me because the American, European and Chinese governments and Silicon Valley companies have different ideas about what is ethical. How AI is used will depend on your government\u2019s hierarchy of values among economic development, international competitiveness and social impacts.\u201d<\/p>\n\n\n\n Jim Witte<\/strong>,<\/strong> director of the Center for Social Science Research at George Mason University, responded, \u201cThe question assumes that ethics and morals are static systems. With developments in AI, there may also be an evolution of these systems such that what is moral and ethical tomorrow may be very different from what we see as moral and ethical today.\u201d<\/p>\n\n\n\n Yves Mathieu<\/strong>,<\/strong> co-director at Missions Publiques, based in Paris, France, wrote, \u201cEthical AI will require legislation like the European [GDPR] legislation to protect privacy rights on the internet. Some governments will take measures but not all will, as is the case today in regard to the production, marketing and usage of guns. There might be an initiative by some corporations, but there will be a need for engagement of the global chain of production of AI, which will be a challenge if some of the production is coming from countries not committed in the same ethical principles. Strong economic sanctions on nonethical AI production and use may be effective.\u201d<\/p>\n\n\n\n Amy Sample Ward<\/strong>,<\/strong> CEO of NTEN: The Nonprofit Technology Network, said, \u201cThere\u2019s no question whether AI will be used in questionable ways. Humans do not share a consistent and collective commitment to ethical standards of any technology, especially not with artificial intelligence. Creating standards is not difficult, but accountability to them is very difficult, especially as government, military and commercial interests regularly find ways around systems of accountability. What systems will be adopted on a large scale to enforce ethical standards and protections for users? How will users have power over their data? How will user education be invested in for all products and services? These questions should guide us in our decision-making today so that we have more hope of AI being used to improve or benefit lives in the years to come.\u201d<\/p>\n\n\n\n Dan McGarry<\/strong>,<\/strong> an independent journalist based in Vanuatu, noted, \u201cJust like every other algorithm ever deployed, AI will be a manifestation of human bias and the perspective of its creator. Facebook\u2019s facial-recognition algorithm performs abysmally when asked to identify Black faces. AIs programmed in the affluent West will share its strengths and weaknesses. Likewise, AIs developed elsewhere will share the assumptions and the environment of their creators. They will not be images of them; they will be products of them and recognisable as such.\u201d<\/p>\n\n\n\n Abigail De Kosnik<\/strong>,<\/strong> associate professor and director of the Center for New Media at the University of California-Berkeley, said, \u201cI don\u2019t see nearly enough understanding in the general public, tech workers or in STEM students about the possible dangers of AI \u2013 the ways that AI can harm and fail society. I am part of a wave of educators trying to introduce more ethics training and courses into our instruction, and I am hopeful that will shift the tide, but I am not optimistic about our chances. AI that is geared toward generating revenue for corporations will nearly always work against the interests of society.\u201d<\/p>\n\n\n\n Irina Raicu, <\/strong>director of the Internet Ethics program at the Markkula Center for Applied Ethics, observed, \u201cThe conversation around AI ethics has been going on for several years now. However, what seems to be obvious among those who have been a part of it for some time has not trickled down into the curricula of many universities who are training the next generation of AI experts. Given that, it looks like it will take more than 10 years for \u2018most of the AI systems being used by organizations of all sorts to employ ethical principles focused primarily on the public good.\u2019 Also, many organizations are simply focused primarily on other goals \u2013 not on protecting or promoting the public good.\u201d<\/p>\n\n\n\n A lawyer and former law school dean who specializes in technology issues<\/strong> wrote, \u201cAI is an exciting new space, but it is unregulated and, at least in early stages, will evolve as investment and monetary considerations direct. It is sufficiently known that there are no acknowledged ethical standards and probably won\u2019t be until beyond the time horizon you mention (2030). During that time, there will be an accumulation of \u2018worst-case scenarios,\u2019 major scandals on its use, a growth in pernicious use that will offend common sense and community moral and ethical standards. Those occasions and situations will lead to a gradual and increasing demand for regulation, oversight and ethical policies on use and misuse. But by whom (or what)? Who gets to impose those ethical prescriptions \u2013 the industries themselves? The government?\u201d<\/p>\n\n\n\n The<\/strong> director of a public policy center<\/strong> responded, \u201cI see a positive future for AI in the areas of health and education. However, there are ethical challenges here, too. Will the corporation who access and hold this data use it responsibly? What will be the role of government? Perhaps AI can help the developing water deal with climate change and water resources, but again, I see a real risk in the areas of equitable distribution, justice and privacy protections.\u201d<\/p>\n\n\n\n A number of the experts who have concerns about the future of ethical AI raised issues around the fundamental nature of people. Flawed humans necessarily will be in the thick of these issues. Moreover, some experts argued that humans will have to create the governance systems overseeing the application of AI and judging how applications are affecting societies. These experts also asserted that there will always be fundamentally unethical people and organizations that will not adopt such principles. Further, some experts mentioned the fact that in a globally networked age even lone wolves can cause massive problems.<\/p>\n\n\n\n Leslie Daigle<\/strong>,<\/strong> a longtime leader in the organizations building the internet and making it secure, noted, \u201cMy biggest concern with respect to AI and its ethical use has nothing to do with AI as a technology and everything to do with people. Nothing about the 21st century convinces me that we, as a society, understand that we are interdependent and need to think of something beyond our own immediate interests. Do we even have a common view of what is ethical?<\/p>\n\n\n\n \u201cTaking one step back from the brink of despair, the things I\u2019d like to see AI successfully applied to, by 2030, include things like medical diagnoses (reading x-rays, etc.). Advances there could be monumental. I still don\u2019t want my fridge to order my groceries by 2030, but maybe that just makes me old? :-)\u201d<\/p>\n\n\n\n Tracey P. Lauriault<\/strong>,<\/strong> a professor expert in critical media studies and big data based at Carleton University, Ottawa, Canada, commented, \u201cAutomation, AI and machine learning (ML) used in traffic management as in changing the lights to improve the flow of traffic, or to search protein databases in big biochemistry analytics, or to help me sort out ideas on what show to watch next or books to read next, or to do land-classification of satellite images, or even to achieve organic and fair precision agriculture, or to detect seismic activity, the melting of polar ice caps, or to predict ocean issues are not that problematic (and its use, goodness forbid, to detect white-collar crime in a fintech context is not a problem).<\/p>\n\n\n\n \u201cIf, however, the question is about social welfare intake systems, biometric sorting, predictive policing and border control, etc., then we are getting into quite a different scenario. How will these be governed, scrutinized, and who will be accountable for decisions and will those decisions about the procurement and use of these technologies or the intelligence derived from them?<\/p>\n\n\n\n \u201cThey will reflect our current forms of governance, and these seem rather biased and unequal. If we can create a more just society then we may be able to have more-just AI\/ML.\u201d<\/p>\n\n\n\n Leiska Evanson<\/strong>,<\/strong> futurist and consultant, wrote, \u201cHumanity has biases. Humans are building the algorithms around the machine learning masquerading as AI. The \u2018AI\u2019 will have biases. It is impossible to have ethical AI (really, ML) if the \u2018parent\u2019 is biased. Companies such as banks are eager to use ML to justify not lending to certain minorities who simply do not create profit for them. Governments want to attend to the needs of the many before the few. The current concepts of AI are all about feeding more data to an electromechanical bureaucrat to rubberstamp, with no oversight from humans with competing biases.\u201d<\/p>\n\n\n\n A director of standards and strategy at a major technology company<\/strong> commented, \u201cI believe that people are mostly good and that the intention will be to create ethical AI. However, an issue that I have become aware of is the fact that we all have intrinsic biases, unintentional biases, that can be exposed in subtle and yet significant ways. Consider that AI systems are built by people, and so they inherently work according to how the people that built them work. Thus, these intrinsic, unintentional biases are present in these systems. Even learning systems will \u2018learn\u2019 in a biased way. So, the interesting research question is whether or not we learn in a way that overcomes our intrinsic biases.\u201d<\/p>\n\n\n\n Jean Seaton,<\/strong> director of the Orwell Foundation and professor of media history at the University of Westminster, responded, \u201cThe ethics question also begs the questions of who would create and police such standards internationally? We need some visionary leaders and some powerful movements. The last big \u2018ethical\u2019 leap came after World War II. The Holocaust and World War II produced a set of institutions that in time led to the notion of human rights. That collective ethical step change (of course compromised but nevertheless immensely significant) was embodied in institutions with some collective authority. So that is what has to happen over AI. People have to be terrified enough, leaders have to be wise enough, people have to be cooperative enough, tech people have to be forward thinking enough, responsibility has to be felt vividly, personally, overwhelmingly enough \u2013 to get a set of rules passed and policed.\u201d<\/p>\n\n\n\n Cliff Lynch,<\/strong> director at the Coalition for Networked Information, wrote, \u201cEfforts will be made to create mostly \u2018ethical\u2019 AI applications by the end of the decade, but please understand that an ethical AI application is really just software that\u2019s embedded in an organization that\u2019s doing something; it\u2019s the organization rather than the software that bears the burden to be ethical. There will be some obvious exceptions for research, some kinds of national security, military and intelligence applications, market trading and economic prediction systems \u2013 many of these things operate under various sorts of \u2018alternative ethical norms\u2019 such as the \u2018laws of war\u2019 or the laws of the marketplace. And many efforts to unleash AI (really machine-learning) on areas like physics or protein-folding will fall outside all of the discussion of \u2018ethical AI.\u2019<\/p>\n\n\n\n \u201cWe should resist the temptation to anthropomorphize these systems. (As the old saying goes, \u2018machines hate that.\u2019) Don\u2019t attribute agency and free will to software. \u2026The problems here are people and organizations, not code! \u2026 A lot of the discussion of ethical AI is really misguided. It\u2019s clear that there\u2019s a huge problem with machine learning and pattern-recognition systems, for example, that are trained on inappropriate, incomplete or biased data (or data that reflect historical social biases) or where the domain of applicability and confidence of the classifiers or predictors aren\u2019t well-demarcated and understood. There\u2019s another huge problem where organizations are relying on (often failure-prone and unreliable, or trained on biased data, or otherwise problematic) pattern recognition or prediction algorithms (again machine-learning-based, usually) and devolving too much decision-making to these. Some of the recent facial-recognition disasters are good examples here. There are horrible organizational and societal practices that appeal to computer-generated decisions that are correct, unbiased, impartial or transparent and that place unjustified faith and authority in this kind of technology. But framing this in terms of AI ethics rather than bad human decision-making, stupidity, ignorance, wishful thinking, organizational failures and attempts to avoid responsibility seems wrong to me. We should be talking instead about the human and organizational ethics of using machine-learning and prediction systems for various purposes, perhaps.<\/p>\n\n\n\n \u201cI think we\u2019ll see various players employ machine learning, pattern recognition and prediction in some really evil ways over the coming decade. Coupling this to social media or other cultural motivation and reward mechanisms is particularly scary. An early example here might be China\u2019s development of its \u2018social capital<\/a>\u2019 rewards and tracking system. I\u2019m also frightened of targeted propaganda\/advertising\/persuasion systems. I\u2019m hopeful we\u2019ll also see organizations and governments in at least a few cases choose not to use these systems or to try to use them very cautiously and wisely and not delegate too much decision-making to them.<\/p>\n\n\n\n \u201cIt\u2019s possible to make good choices here, and I think some will. Genuine AI ethics seems to be part of the thinking about general-purpose AI, and I think we are a very, very, long way from this, though I\u2019ve seen some predictions to the contrary from people perhaps better informed than I am. The (rather more theoretical and speculative) philosophical and research discussions about superintelligence and about how one might design and develop such a general-purpose AI that won\u2019t rapidly decide to exterminate humanity are extremely useful, important and valid, but they have little to do with the rhetorical social justice critiques that confuse algorithms with the organizations that stupidly and inappropriately design, train and enshrine and apply them in today\u2019s world.\u201d<\/p>\n\n\n\n Deirdre Williams,<\/strong> an independent researcher expert in global technology policy, commented, \u201cI can\u2019t be optimistic. We, the \u2018average persons,\u2019 have been schooled in preceding years toward selfishness, individualism, materialism and the ultimate importance of convenience. These values create the \u2018ethos.\u2019 At the very root of AI are databases, and these databases are constructed by human beings who decide which data are to be collected and how that data should be described and categorised. A tiny human error or bias at the very beginning can balloon into an enormous error of truth and\/or justice.\u201d<\/p>\n\n\n\n Alexa Raad<\/strong>,<\/strong> co-founder and co-host of the TechSequences podcast and former chief operating officer at Farsight Security, said, \u201cThere is hope for AI in terms of applications in health care that will make a positive difference. But legal\/policy and regulatory frameworks almost always lag behind technical innovations. In order to guard against the negative repercussions of AI, we need a policy governance and risk-mitigation framework that is universally adopted. There needs to be an environment of global collaboration for a greater good. Although globalization led to many of the advances we have today (for example, the internet\u2019s design and architecture as well as its multistakeholder governance model), globalization is under attack. What we see across the world is a trend toward isolationism, separatism as evidenced by political movements such as populism, nationalism and outcomes such as Brexit. In order to come up with and adopt a comprehensive set of guidelines or framework for the use of AI or risk mitigation for abuse of AI, we would need a global current that supports collaboration. I hope I am wrong, but trends like this need longer than 10 years to run their course and for the pendulum to swing back the other way. By then, I am afraid some of the downsides and risks of AI will already be in play.\u201d<\/p>\n\n\n\n Andrea Romaoli Garcia<\/strong>,<\/strong> an international lawyer actively involved with multistakeholder activities of the International Telecommunication Union and Internet Society, said, \u201cI define ethics as all possible and available choices where the conscience establishes the best option. Values and principles are the limiters that guide the conscience into this choice alongside the purposes; thus, ethics is a process. In terms of ethics for AI, the process for discovering what is good and right means choosing among all possible and available applications to find the one that best applies to the human-centred purposes, respecting all the principles and values that make human life possible.<\/p>\n\n\n\n \u201cThe human-centred approach in ethics was first described by the Greek philosopher Socrates in his effort to turn attention from the outside world to the human condition. AI is a cognitive technology that allows greater advances in health, economic, political and social fields. It is impossible to deny how algorithms impact human evolution. Thus, an ethical AI requires that all instruments and applications place humans at the center. Despite the fact that there are some countries building ethical principles for AI, there is a lack of any sort of international instrument that covers all of the fields that guide the development and application of AI in a human-centred approach. AI isn\u2019t model-driven; it has a data-centred approach for highly scalable neural networks. Thus, the data should be selected and classified through human action. Through this human action, sociocultural factors are imprinted on the behavior of the algorithm and machine learning. This justifies the concerns about ethics and also focuses on issues such as freedom of expression, privacy and surveillance, ownership of data and discrimination, manipulation of information and trust, environmental and global warming and also on how the power will be established among society.<\/p>\n\n\n\n \u201cThese are factors that determine human understanding and experience. All instruments that are built for ethical AI have different bases, values and purposes depending on the field to which they apply. The lack of harmony in defining these pillars compromises ethics for AI and affects human survival. It could bring new invisible means of exclusion or deploy threats to social peace that will be invisible to human eyes. Thus, there is a need for joint efforts gathering stakeholders, civil society, scientists, governments and intergovernmental bodies to work toward building a harmonious ethical AI that is human-centred and applicable to all nations. 2030 is 10 years from now. We don\u2019t need to wait 10 years \u2013 we can start working now. 2020 presents several challenges in regard to technology\u2019s impact on people. Human rights violations are being exposed and values are under threat. This scenario should accelerate efforts at international cooperation to establish a harmonious ethical AI that supports human survival and global evolution.\u201d<\/p>\n\n\n\n Olivier MJ Cr\u00e9pin-Leblond<\/strong>,<\/strong> entrepreneur and longtime participant in the activities of ICANN and IGF, said, \u201cWhat worries me the most is that some actors in nondemocratic regimes do not see the same \u2018norm\u2019 when it comes to ethics. These norms are built on a background of culture and ideology, and not all ideologies are the same around the world. It is clear that, today, some nation-states see AI as another means of conquest and establishing their superiority instead of a means to do good.\u201d<\/p>\n\n\n\n A professor emeritus of social science<\/strong> said, \u201cThe algorithms that represent ethics in AI are neither ethical nor intelligent. We are building computer models of social prejudices and structural racism, sexism, ageism, xenophobia and other forms of social inequality. It\u2019s the realization of some of Foucault\u2019s worst nightmares.\u201d<\/p>\n\n\n\n An advocate and activist<\/strong> said, \u201cMost of the large AI convenings to date have been dominated by status quo power elites whose sense of risk, harm, threat are distorted. Largely comprised of elite white men with an excessive faith in technical solutions and a disdain for sociocultural dimensions of risk and remedy. These communities \u2013 homogenous, limited experientially, overly confident \u2013 are made up of people who fail to see themselves as a risk<\/em>. As a result, I believe that most dominant outcomes \u2013 how \u2018ethical\u2019 is defined, how \u2018acceptable risk\u2019 is perceived, how \u2018optimal solutions\u2019 will be determined \u2013 will be limited and almost certainly perpetuate and amplify existing harms. As you can see, I\u2019m all sunshine and joy.\u201d<\/p>\n\n\n\n Glenn Grossman<\/strong>,<\/strong> a consultant of banking analytics at FICO, noted, \u201cIt\u2019s necessary for leaders in all sectors to recognize that AI is just the growth of mathematical models and the application of these techniques. We have model governance in most organizations today. We need to keep the same safeguards in place. The challenge is that many business leaders are not good at math! They cannot understand the basics of predictive analytics, models and such. Therefore, they hear \u2018AI\u2019 and think of it as some new, cool, do-it-all technology. It is simply math at the heart of it. Man governs how they use math. So, we need to apply ethical standards to monitor and calibrate. AI is a tool, not a solution for everything. Just like the PC ushered in automation, AI can usher in automation in the area of decisions. Yet it is humans that use these decisions and design the systems. So, we need to apply ethical standards to any AI-driven system.\u201d<\/p>\n\n\n\n R. \u201cRay\u201d Wang,<\/strong> principal analyst, founder and CEO of Silicon Valley-based Constellation Research, noted, \u201cRight now we have no way of enforcing these principles in play. Totalitarian, Chinese, CCP-style AI is the preferred approach for dictators. The question is: Can we require and can we enforce AI ethics? We can certainly require, but the enforcement may be tough.\u201d<\/p>\n\n\n\n “The question is: Can we require and can we enforce AI ethics? We can certainly require, but the enforcement may be tough.\u201d<\/p>R. “Ray” Wang, PriNCipal Analyst, Founder and CEO of Silicon Valley-based Constellation REsearch<\/cite><\/blockquote><\/figure>\n\n\n\n Maja Vujovic<\/strong>,<\/strong> a consultant for digital and ICT at Compass Communications, noted, \u201cEthical AI might become a generally agreed upon standard, but it will be impossible to enforce it. In a world where media content and production, including fake news, will routinely be AI-generated, it is more likely that our expectations around ethics will need to be lowered. Audiences might develop a \u2018thicker skin\u2019 and become more tolerant toward the overall unreliability of the news. This trend will not render them more skeptical or aloof but rather more active and much more involved in the generation of news, in a range of ways. Certification mechanisms and specialized AI tools will be developed to deal specifically with unethical AI, as humans will prove too gullible. In those sectors where politics don\u2019t have a direct interest, such as health and medicine, transportation, e-commerce and entertainment, AI as an industry might get more leeway to grow organically, including self-regulation.\u201d<\/p>\n\n\n\n A number of respondents noted that any attempt at rule-making is complicated by the fact that any technology can be used for noble and<\/em> harmful purposes. It is difficult to design ethical digital tools that privilege the former while keeping the latter in check.<\/p>\n\n\n\n Chris Arkenberg,<\/strong> research manager at Deloitte\u2019s Center for Technology, Media and Telecommunications, noted, \u201cThe answer is both good and bad. Technology doesn\u2019t adopt ethical priorities that humans don\u2019t prioritize themselves. So, a better question could be whether society will pursue a more central role of ethics and values than we\u2019ve seen in the past 40 years or so. Arguably, 2020 has shown a resurgent demand for values and principles for a balanced society. If, for example, education becomes a greater priority for the Western world, AI could amplify our ability to learn more effectively. Likewise, with racial and gender biases. But this trend is strongest only in some Western democracies.<\/p>\n\n\n\n \u201cChina, for example, places a greater value on social stability and enjoys a fairly monochromatic population. With the current trade wars, the geopolitical divide is also becoming a technological divide that could birth entirely different shapes of AI depending on their origin. And it is now a very multipolar world with an abundance of empowered actors.<\/p>\n\n\n\n \u201cSo, these tools lift up many other boats with their own agendas [that] may be less bound by Western liberal notions of ethics and values. The pragmatic assumption might be that many instances of ethical AI will be present where regulations, market development, talent attraction, and societal expectations require them to be so. At the same time, there will likely be innumerable instances of \u2018bad AI,\u2019 weaponized machine intelligence and learning systems designed to exploit weaknesses. Like the internet and globalization, the path forward is likely less about guiding such complex systems toward utopian outcomes and more about adapting to how humans wield them under the same competitive and collaborative drivers that have attended the entirety of human history.\u201d<\/p>\n\n\n\n Kenneth Cukier<\/strong>,<\/strong> senior editor at The Economist and coauthor of \u201cBig Data<\/a>,\u201d said, \u201cFew will set out to use AI in bad ways (though some criminals certainly will). The majority of institutions will apply AI to address real-world problems effectively, and AI will indeed work for that purpose. But if it is facial recognition, it will mean less privacy and risks of being singled out unfairly. If it is targeted advertising, it will be the risk of losing anonymity. In health care, an AI system may identify that some people need more radiation to penetrate the pigment in their skin to get a clearer medical image, but if this means Black people are blasted with higher doses of radiation and are therefore prone to negative side effects, people will believe there is an unfair bias.<\/p>\n\n\n\n \u201cOn global economics, a \u2018neocolonial\u2019 or \u2018imperial\u2019 commercial structure will form, whereby all countries have to become customers of AI from one of the major powers, America, China and, to a lesser extent, perhaps Europe.\u201d<\/p>\n\n\n\n Bruce Mehlman<\/strong>,<\/strong> a futurist and consultant, responded, \u201cAI is powerful and has a huge impact, but it\u2019s only a tool like gunpowder, electricity or aviation. Good people will use it in good ways for the benefit of mankind. Bad people will use it in nefarious ways to the detriment of society. Human nature has not changed and will neither be improved nor worsened by AI. It will be the best of technologies and the worst of technologies.\u201d<\/p>\n\n\n\n Ian Thomson<\/strong>,<\/strong> a pioneer developer of the Pacific Knowledge Hub, observed, \u201cIt will always be the case that new uses of AI will raise ethical issues, but over time, these issues will be addressed so that the majority of uses will be ethical. Good uses of AI will include highlighting trends and developments that we are unhappy with. Bad uses will be around using AI to manipulate our opinions and behaviors for the financial gain of those rich enough to develop the AI and to the disadvantage of those less well-off. I am excited by how AI can help us make better decisions, but I am wary that it can also be used to manipulate us.\u201d<\/p>\n\n\n\n A<\/strong> professor of international affairs and economics at a Washington, D.C.-area university<\/strong> wrote, \u201cAI tends to be murky in the way it operates and the kinds of outcomes that it obtains. Consequently, it can easily be used to both good and bad ends without much practical oversight. AI, as it is currently implemented, tends to reduce the personal agency of individuals and instead creates a parallel agent who anticipates and creates needs in accordance with what others think is right. The individual being aided by AI should be able to fully comprehend what it is doing and easily alter how it works to better align with their own preferences. My concerns are allayed to the extent that the operation of AI, and its potential biases and\/or manipulation remain unclear to the user. I fear its impact. This, of course, is independent from an additional concern for individual privacy. I want the user to be in control of the technology, not the other way around.\u201d<\/p>\n\n\n\n Kate Klonick<\/strong>,<\/strong> a law professor at St. John\u2019s University whose research is focused on law and technology, said, \u201cAI will be used for both good and bad, like most new technologies. I do not see AI as a zero-sum negative of bad or good. I think at net AI has improved people\u2019s lives and will continue to do so, but this is a source of massive contention within the communities that build AI systems and the communities that study their effects on society.\u201d<\/p>\n\n\n\n Stephan G. Humer,<\/strong> professor and director, Internet Sociology Department at Fresenius University of Applied Sciences in Berlin, predicted, \u201cWe will see a dichotomy: Official systems will no longer be designed in such a naive and technology-centered way as in the early days of digitization, and ethics will play a major role in that. \u2018Unofficial\u2019 designs will, of course, take place without any ethical framing, for example, in the area of crime as a service. What worries me the most is lack of knowledge: Those who know little about AI will fear it, and the whole idea of AI will suffer. Spectacular developments will be mainly in the U.S. and China. The rest of the world will not play a significant role for the time being.\u201d<\/p>\n\n\n\n An anonymous respondent<\/strong> wrote, \u201cIt\u2019s an open question. Black Lives Matter and other social justice movements must \u2018shame\u2019 and force profit-focused companies to delve into the inherently biased data and information they\u2019re feeding the AI systems \u2013 the bots and robots \u2013 and try to keep those biased ways of thinking to a minimum. There will need to be checks and balances to ensure the AI systems don\u2019t have the final word, including on hiring, promoting and otherwise rewarding people. I worry that AI systems such as facial recognition will be abused, especially by totalitarian governments, police forces in all countries and even retail stores \u2013 in regard to who is the \u2018best\u2019 and \u2018most-suspicious\u2019 shopper coming in the door. I worry that AI systems will lull people into being okay with giving up their privacy rights. But I also see artists, actors, movie directors and other creatives using AI to give voice to issues that our country needs to confront. I also hope that AI will somehow ease transportation, education and health care inequities.\u201d<\/p>\n\n\n\n Ilana Schoenfeld<\/strong>, <\/strong>an expert in designing online education and knowledge-sharing systems, said, \u201cI am frightened and at the same time excited about the possibilities of the use of AI applications in the lives of more and more people. AI will be used in both ethical and questionable ways, as there will always be people on both sides of the equation trying to find ways to further their agendas. In order to ensure that the ethical use of AI outweighs its questionable use, we need to get our institutional safeguards right \u2013 both in terms of their structures and their enforcement by nonpartisan entities.\u201d<\/p>\n\n\n\n A<\/strong> pioneer in venture philanthropy<\/strong> commented, \u201cWhile many will be ethical in the development and deployment of AI\/ML, one cannot assume \u2018goodness.\u2019 Why will AI\/ML be any different than how:<\/p>\n\n\n\n Greg Shatan<\/strong>,<\/strong> a partner in Moses & Singer LLC\u2019s intellectual property group and a member of its internet and technology practice, wrote, \u201cEthical use will be widespread, but ethically questionable use will be where an ethicist would least want it to be: Oppressive state action in certain jurisdictions; the pursuit of profit leading to the hardening of economic strata; policing, etc.\u201d<\/p>\n\n\n\n Some respondents said that the rise of AI raises new questions about what it means to be ethical. A number of these experts argued that today\u2019s AI is unsophisticated compared with what the future is likely to bring. Acceleration from narrow AI to artificial general intelligence and possibly to artificial superintelligence is expected by some to evolve these tools beyond human control and understanding. Then, too, there\u2019s the problem of misinformation and disinformation (such as deepfakes) and how they might befoul ethics systems.<\/p>\n\n\n\n David Barnhizer,<\/strong> professor of law emeritus and author of \u201cThe Artificial Intelligence Contagion: Can Democracy Withstand the Imminent Transformation of Work, <\/em>Wealth and the Social Order?<\/a>\u201d wrote, \u201cThe pace of AI development has accelerated and is continuing to pick up speed. In considering the fuller range of the \u2018goods\u2019 and \u2018bads\u2019 of artificial intelligence, think of the implications of Masayoshi Son\u2019s warning that: \u2018Supersmart robots will outnumber humans, and more than a trillion objects will be connected to the internet within three decades.\u2019 Researchers are creating systems that are increasingly able to teach themselves and use their new and expanding ability to improve and evolve. The ability to do this is moving ahead with amazing rapidity. They can achieve great feats, like driving cars and predicting diseases, and some of their makers say they aren\u2019t entirely in control of their creations. Consider the implications of a system that can access, store, manipulate, evaluate, integrate and utilize all forms of knowledge. This has the potential to reach levels so far beyond what humans are capable of that it could end up as an omniscient and omnipresent system.<\/p>\n\n\n\n \u201cIs AI humanity\u2019s \u2018last invention\u2019? Oxford\u2019s Nick Bostrom suggests we may lose control of AI systems sooner than we think. He asserts that our increasing inability to understand what such systems are doing, what they are learning and how the \u2018AI Mind\u2019 works as it further develops could inadvertently cause our own destruction. Our challenges are numerous, even if we only had to deal with the expanding capabilities of AI systems based on the best binary technology. The incredible miniaturization and capability shift represented by quantum computers has implications far beyond binary AI.<\/p>\n\n\n\n \u201cThe work on technological breakthroughs such as quantum computers capable of operating at speeds that are multiple orders of magnitude beyond even the fastest current computers is still at a relatively early stage and will take time to develop beyond the laboratory context. If scientists are successful in achieving a reliable quantum computer system, the best exascale system<\/a> will pale in relation to the reduced size and exponentially expanded capacity of the most advanced existing computer systems. This will create AI\/robotics applications and technologies we can now only imagine. \u2026 When fully developed, quantum computers will have data-handling and processing capabilities far beyond those of current binary systems. When this occurs in the commercialized context, predictions about what will happen to humans and their societies are \u2018off the board.\u2019\u201d<\/p>\n\n\n\n An expert in the regulation of risk and the roles of politics within science and science within politics<\/strong> observed, \u201cIn my work, I use cost-benefit analysis. It is an elegant model that is generally recognized to ignore many of the most important aspects of decision-making \u2013 how to \u2018value\u2019 nonmonetary benefits, for example. Good CBA analysts tend to be humble about their techniques, noting that they provide a partial<\/em> view of decision structures. I\u2019ve heard too many AI enthusiasts talk about AI applications with no humility at all. Cathy O\u2019Neil\u2019s book \u2018Weapons of Math Destruction<\/a>\u2019 was perfectly on target: If you can\u2019t count it, it doesn\u2019t exist. The other major problem is widely discussed: the transparency of the algorithms. One problem with AI is that it is self-altering. We almost certainly won\u2019t know what an algorithm has learned, adopted, mal-adopted, etc. This problem already exists, for example, in using AI for hiring decisions. I doubt there will be much hesitancy about grabbing AI as the \u2018neutral, objective, fast, cheap\u2019 way to avoid all those messy human-type complications, such as justice, empathy, etc.\u201d<\/p>\n\n\n\n Neil Davies<\/strong>, <\/strong>co-founder of Predictable Network Solutions and a pioneer of the committee that oversaw the UK\u2019s initial networking developments, commented, \u201cMachine learning (I refuse to call it AI, as the prerequisite intelligence behind such systems is definitely not artificial) is fundamentally about transforming a real-world issue into a numerical value system, the processing (and decisions) being performed entirely in that numerical system. For there to be an ethical dimension to such analysis, there needs to be a means of assessing the ethical outcome as a (function from) such a numerical value space. I know of no such work. \u2026<\/p>\n\n\n\n \u201cThere is a nontrivial possibility of multiple dystopian outcomes. The UK government\u2019s track record, as an example \u2013 but other countries have their examples, too, on universal credit, Windrush<\/a>, EU settled status<\/a>, etc., are all examples of a value-based assessment process in which the notion of assurance against some ethical framework is absent. The global competition aspect is likely to lead to monopolistic tendencies over the \u2018ownership\u2019 of information \u2013 much of which would be seen as a common good today. \u2026<\/p>\n\n\n\n \u201cA cautionary tale: In the mathematics that underpins all modelling of this kind (category theory), there are the notions of \u2018infidelity\u2019 and \u2018junk.\u2019 Infidelity is the failure to capture the \u2018real world\u2019 well enough to even have the appropriate values (and structure of values) in the evaluatable model; this leads to \u2018garbage in, garbage out.\u2019 Junk, on the other hand, are things that come into existence only as artefacts in the model. Such junk artefacts are often difficult to recognise (in that, if they were easy the model would have been adapted to deny their very existence) and can be alluring to the model creator (the human intelligence) and the machine algorithms as they seek their goal. Too many of these systems will create negative (and destructive) value because of the failure to recognise recognize this fundamental limitation; the failure to perform adequate (or even any) assurance on the operation of the system; and, pure hubris driven by the need to show a \u2018return on investment\u2019 for such endeavours.\u201d<\/p>\n\n\n\n Sarita Schoenebeck,<\/strong> an associate professor at the School of Information at the University of Michigan, said, \u201cAI will mostly be used in questionable ways and sometimes not used at all. There\u2019s little evidence that researchers can discern or agree on what ethical AI looks like, let alone be able to build it within a decade. Ethical AI should minimize harm, repair injustices, avoid re-traumatization and center user needs rather than technological goals. Ethical AI will need to shift away from notions of fairness, which overlook concepts like harm, injustice and trauma. This requires reconciling AI design principles like scalability and automation with individual and community values.\u201d<\/p>\n\n\n\n Jeff Gulati<\/strong>,<\/strong> professor of political science at Bentley University, responded, \u201cIt seems that more AI and the data coming out could be useful in increasing public safety and national security. In a crisis, we will be more willing to rely on these applications and data. As the crisis subsides, it is unlikely that the structures and practices built during the crisis will go away and unlikely to remain idle. I could see it being used in the name of prevention and lead to further erosion of privacy and civil liberties in general. And, of course, these applications will be available to commercial organizations, who will get to know us more intimately so they can sell us more stuff we don\u2019t really need.\u201d<\/p>\n\n\n\n A senior leader for an international digital rights organization<\/strong> commented, \u201cWhy would AI be used ethically? You only have to look at the status quo to see that it\u2019s not used ethically. Lots of policymakers don\u2019t understand AI at all. Predictive policing is a buzzword, but most of it is snake oil. Companies will replace workers with AI systems if they can. They\u2019re training biased biometric systems. And we don\u2019t even know in many cases what the algorithm is really doing; we are fighting for transparency and explainability.<\/p>\n\n\n\n \u201cI expect this inherent opaqueness of AI\/ML techs to be a feature for companies (and governments) \u2013 not a bug. Deepfakes are an example. Do you expect ethical use? Don\u2019t we think about it precisely because we expect unethical, bad-faith use in politics, \u2018revenge porn,\u2019 etc.? In a tech-capitalist economy, you have to create and configure the system even to begin to have incentives for ethical behavior. And one basic part of ethics is thinking about who might be harmed by your actions and maybe even respecting their agency in decisions that are fateful for them.<\/p>\n\n\n\n \u201cFinally, of course AI has enormous military applications, and U.S. thinking on AI takes place in a realm of conflict with China. That again does not make me feel good. China is leading, or trying to lead, the world in social and political surveillance, so it\u2019s driving facial recognition and biometrics. Presumably, China is trying to do the same in military or defense areas, and the Pentagon is presumably competing like mad. I don\u2019t even know how to talk about ethical AI in the military context.\u201d<\/p>\n\n\n\n Emmanuel Evans Ntekop<\/strong> observed, \u201cWithout the maker, the object is useless. The maker is the programmer and the god to its objects. It was an idea from the start to support as a slave to its master the people, like automobiles.\u201d<\/p>\n\n\n\n Many of these experts expressed concern that AI systems are being built by for-profit firms and by governments focused on applying AI for their own purposes. Some said governments are passive enablers of corporate abuses of AI. They noted that the public is unable to understand how the systems are built, they are not informed as to their impact and they are unable to challenge firms that try to invoke ethics in a public relations context but are not truly committed to ethics. Some experts said the phrase \u201cethical AI\u201d will merely be used as public relations window dressing to try to deflect scrutiny of questionable applications.<\/p>\n\n\n\n A number of these experts framed their concerns around the lack of transparency about how AI products are designed and trained. Some noted that product builders are programming AI by using available datasets with no analysis of the potential for built-in bias or other possible quality or veracity concerns.<\/p>\n\n\n\n Joseph Turow,<\/strong> <\/strong>professor of communication at the University of Pennsylvania, wrote, \u201cTerms such \u2018transparency, justice and fairness, privacy, freedom and human autonomy, beneficence and nonmaleficence, freedom, trust, sustainability and dignity\u2019 can have many definitions so that companies (and governments) can say they espouse one or another term but then implement it algorithmically in ways that many outsiders would not find satisfactory. For example, the Chinese government may say its AI technologies embed values of freedom, human autonomy and dignity. My concern is that companies will define \u2018ethical\u2019 in ways that best match their interests, often with vague precepts that sound good from a PR standpoint but, when integrated into code, allow their algorithms to proceed in ways that do not constrain them from creating products that \u2018work\u2019 in a pragmatic sense.\u201d<\/p>\n\n\n\n Charlie Kaufman,<\/strong> a security architect with Dell EMC, said, \u201cThere may be ethical guidelines imposed on AI-based systems by legal systems in 2030, but they will have little effect \u2013 just as privacy principles have little effect today. Businesses are motivated to maximize profits, and they will find ways to do that, giving only lip service to other goals. If ethical behavior or results were easy to define or measure, perhaps society could incentivize them. But usually, the implications of some new technological development don\u2019t become clear until it has already spread too far to contain it.<\/p>\n\n\n\n “There may be ethical guidelines imposed on AI-based systems by legal systems in 2030, but they will have little effect — just as privacy principles have little effect today. Businesses are motivated to maximize profits, and they will find ways to do that, giving only lip service to other goals.”<\/p>Charlie Kaufman, a security architect with Dell EMC<\/cite><\/blockquote><\/figure>\n\n\n\n \u201cThe biggest impact of AI-based systems is the ability to automate increasingly complex jobs, and this will cause dislocations in the job market and in society. Whether it turns out to be a benefit to society or a disaster depends on how society responds and adjusts. But it doesn\u2019t matter, because there is no way to suppress the technology. The best we can do is figure out how to optimize the society that results.<\/p>\n\n\n\n \u201cI\u2019m not concerned about the global competition in AI systems. Regardless of where the progress comes from, it will affect us all. And it is unlikely the most successful developers will derive any permanent advantage. The most important implication of the global competition is that it is pointless for any one country or group of countries to try to suppress the technology. Unless it can be suppressed everywhere, it is coming. Let\u2019s try to make that be a good thing!\u201d<\/p>\n\n\n\n Simeon Yates,<\/strong> <\/strong>a professor expert in digital culture and personal interaction at the University of Liverpool and the research lead for the UK government\u2019s Digital Culture team, predicted, \u201cUntil we bring in \u2018ethical-by-design\u2019 (responsible innovation) principles to ICT [information and communications technologies] and AI\/machine learning design \u2013 like attempts to create \u2018secure-by-design\u2019 systems to fight cybercrime \u2013 the majority of AI systems will remain biased and unethical in principle. Though there is a great public debate about AI ethics, and many organisations are seeking to provide both advice and research on the topic, there is no economic or political imperative to make AI systems ethical. First of all, there is great profit to be made from the manipulation of data and, through it, people. Second, there is a limited ability at present for governments to think through how to regulate AI and enforce ethics (as they do say for bio-sciences). Third, governments are complicit often in poor and ethically questionable use of data. Further, this is not in the main \u2018artificial intelligence\u2019 \u2013 but relatively simplistic statistical machine learning based on biased datasets. The knowing use of such is in and of itself unethical yet often profitable. The presentation of such solutions as bias-free or more rational or often \u2018cleverer\u2019 as they are based on \u2018cold computation,\u2019 not \u2018emotive human thinking,\u2019 is itself a false and an unethical claim.\u201d<\/p>\n\n\n\n Colin Allen,<\/strong> <\/strong>a cognitive scientist and philosopher who has studied and written about AI ethics, wrote, \u201cCorporate and government statements of ethics are often broad and nonspecific and thus vague with respect to what specifically is disallowed. This allows considerable leeway in how such principles are implemented and makes enforcement difficult. In the U.S., I don\u2019t see strong laws being enacted within 10 years that would allow for the kind of oversight that would prevent unethical or questionable uses of AI, whether intended or accidental. On the hopeful side, there is increasing public awareness and journalistic coverage of these issues that may influence corporations to build and protect their reputations for good stewardship of AI. But corporations have a long history of hiding or obfuscating their true intent (it\u2019s partly required to stay competitive, not to let everyone else know what you are doing) as well as engaging actively in public disinformation campaigns. I don\u2019t see that changing, and, given that the business advantages to using AI will be mostly in data analytics and prediction and not so much in consumer gadgets in the next 10 years, much of the use of AI will be \u2018behind the scenes,\u2019 so to speak. Another class of problem is that individuals in both corporate and government jobs who have access to data will be tempted, as we have seen already, to access information about people they know and use that information in some way against them. Nevertheless, there will undoubtedly be some very useful products that consumers will want to use and that they will benefit from. The question is whether these added benefits will constitute a Faustian bargain, leading down a path that will be difficult if not impossible to reverse.\u201d<\/p>\n\n\n\n Alice E. Marwick<\/strong>,<\/strong> assistant professor of communication at the University of North Carolina, Chapel Hill, and adviser for the Media Manipulation project at the Data & Society Research Institute, commented, \u201cI have no faith in our current system of government to pass any sort of legislation that deals with technology in a complex or nuanced way. We cannot depend on technology companies to self-regulate, as there are too many financial incentives to employ AI systems in ways that disadvantage people or are unethical.\u201d<\/p>\n\n\n\n Jillian York<\/strong>,<\/strong> director of international freedom of expression for the Electronic Frontier Foundation, said, \u201cThere is absolutely no question that AI will be used in questionable ways. There is no regulatory regime, and many \u2018ethics in AI\u2019 projects are simply window dressing for an unaccountable and unethical industry. When it comes to AI, everything concerns me and nothing excites me. I don\u2019t see the positive potential, just another ethical morass, because the people running the show have no desire to build technology to benefit the 99%.\u201d<\/p>\n\n\n\n David Mussington<\/strong>,<\/strong> a senior fellow at CIGI and professor and director at the Center for Public Policy and Private Enterprise at the University of Maryland, predicted, \u201cMost AI systems deployed by 2030 will be owned and developed in the private sector, both in the U.S. and elsewhere in the world. I can\u2019t conceive of a legislative framework fully up to understanding and selectively intervening in AI rollouts in a manner with predictable consequences. Also, the mode of intervention \u2013 because I think interventions by public authorities will be attempted (just not successful) \u2013 is itself in doubt. Key questions: Do public authorities understand AI and its applications? Is public-institution-sponsored R&D in AI likely to inform government and public research agencies of the scale and capability trajectory of private sector AI research and development? As tool sets for AI development continue to empower small research groups and individuals (datasets, software-development frameworks and open-source algorithms), how is a government going to keep up \u2013 let alone maintain awareness \u2013 of AI progress? Does the government have access to the expertise necessary to make good policy \u2013 and anticipate possible risk factors? I think that the answers to most of these questions are in the negative.\u201d<\/p>\n\n\n\n Giacomo Mazzone,<\/strong> head of institutional relations for the European Broadcasting Union and Eurovision, observed, \u201cNobody could realistically predict ethics for AI will evolve, despite all of the efforts deployed by the UN secretary general, the UNESCO director general and many others. Individuals alone can\u2019t make these decisions because AI is applied at mass scale. Nobody will create an algorithm to solve it. Ethical principles are likely to be applied only if industry agrees to do so; it is likely that this will not happen until governments that value human rights will oblige companies to do so. The size and influence of the companies that control AI and its impact on citizens is making them more powerful than any one nation-state. So, it is very likely only regional supranational powers such as the European Union or multilateral institutions such as the United Nations \u2013 if empowered by all nation-states \u2013 could require companies to apply ethical rules to AI. Of course, many governments already do not support human rights principles, considering the preservation of the existing regime to be a priority more important than individual citizens\u2019 rights.\u201d<\/p>\n\n\n\n Rob Frieden<\/strong>,<\/strong> a professor of telecommunications law at Penn State who previously worked with Motorola and has held senior policy positions at the FCC and the NTIA, said, \u201cI cannot see a future scenario where governments can protect citizens from the incentives of stakeholders to violate privacy and fair-minded consumer protections. Surveillance, discrimination, corner cutting, etc., are certainties. I\u2019m mindful of the adage: garbage in, garbage out. It\u2019s foolish to think AI will lack flawed coding.\u201d<\/p>\n\n\n\n Alex Halavais<\/strong>,<\/strong> associate professor of critical data studies, Arizona State University, noted, \u201cIt isn\u2019t a binary question. I teach in a graduate program that has training in the ethical use of data at its core and hopes to serve organizations that aim to incorporate ethical approaches. There are significant ethical issues in the implementation of any algorithmic system, and such systems have the ethical questions they address coded into them. In most cases, these will substantially favor the owners of the technologies that implement them rather than the consumers. I have no doubt that current unethical practices by companies, governments and other organizations will continue to grow. We will have a growing number of cases where those ethical concerns come to the forefront (as they have recently with facial recognition), but unless they rise to the level of very widespread abuse, it is unlikely that they will be regulated. As a result, they will continue to serve those who pay for the technologies or own them, and the rights and interests of individual users will be pushed to the sidelines. That does not mean that ethics will be ignored. I expect many large technology companies will make an effort to hire professional ethicists to audit their work, and that we may see companies that differentiate themselves through more ethical approaches to their work.\u201d<\/p>\n\n\n\n Ebenezer Baldwin Bowles<\/strong>,<\/strong> an advocate\/activist, commented, \u201cAltruism on the part of the designers of AI is a myth of corporate propaganda. Ethical interfaces between AI and citizenry in 2030 will be a cynical expression by the designers of a digital Potemkin Village \u2013 looks good from the outside but totally empty behind the facade. AI will function according to two motivations: one, to gather more and more personal information for the purposes of subliminal and direct advertising and marketing campaigns; and two, to employ big data to root out radical thinking and exercise near total control of the citizenry. The state stands ready through AI to silence all voices of perceived dissent.<\/p>\n\n\n\n \u201cI\u2019m convinced that any expression of ethical AI will stand as an empty pledge \u2013 yes, we will always do the right thing for the advancement of the greater good. No way. Rather, the creators of AI will do what is good for the bottom line, either through financial schemes to feed the corporate beast or psychological operations directed toward control of dissent and pacification. As for the concept that \u2018humans will be in the loop,\u2019 we are already out of the loop because there is no loop. Think about this fact: In the development of any major expression of artificial intelligence, hundreds of IT professionals are assigned to a legion of disparate, discrete teams of software writers and mechanical designers to create the final product. No single individual or team fully understands what the other teams are doing. The final AI product is a singular creature no one person understands \u2013 other than the software itself. Ethical action is not a part of the equation.\u201d<\/p>\n\n\n\n Richard Lachmann<\/strong>,<\/strong> professor of political sociology at the State University of New York-Albany, predicted, \u201cAI will be used mainly in questionable ways. For the most part, it is being developed by corporations that are motivated exclusively by the desire to make ever bigger profits. Governments see AI, either developed by government programmers or on contract by corporations, as a means to survey and control their populations. All of this is ominous. Global competition is a race to the bottom as corporations try to draw in larger audiences and control more of their time and behavior. As governments get better at surveying their populations, [it] lowers the standards for individual privacy. For almost all people, these applications will make their lives more isolated, expose them to manipulation, and degrade or destroy their jobs. The only hopeful sign is rising awareness of these problems and the beginnings of demands to break up or regulate the huge corporations.\u201d<\/p>\n\n\n\n Kevin T. Leicht,<\/strong> professor and head of the department of sociology at the University of Illinois-Urbana-Champaign, observed, \u201cThe good possibilities here are endless. But the questionable ways are endless, and we have a very poor track record of stopping ethically questionable developments in most areas of life \u2013 why wouldn\u2019t that apply here? In social science, the best predictor of future behavior is past behavior. The opium addict who says, after a binge, that \u2018they\u2019ve got this\u2019 \u2013 they don\u2019t need to enter treatment, and they\u2019ll never use opium again \u2013 is (rightly) not believed. So, in an environment where ethically questionable behavior has been allowed or even glorified in areas such as finance, corporate governance, government itself, pharmaceuticals, education and policing, why all of a sudden are we supposed to believe that AI developers will behave in an ethical fashion? There aren\u2019t any guardrails here, just as there weren\u2019t in these other spheres of life. AI has the potential to transform how cities work, how medical diagnosis happens, how students are taught and a variety of other things. All of these could make a big difference in the lives of most people.<\/p>\n\n\n\n “The good possibilities here are endless. But the questionable ways are endless, and we have a very poor track record of stopping ethically questionable developments in most areas of life \u2013 why wouldn\u2019t that apply here?”<\/p>Kevin T. Leicht, professor and head of the department of sociology at the University of illinois-urbana-champaign<\/cite><\/blockquote><\/figure>\n\n\n\n \u201cBut those benefits won\u2019t come if AI is controlled by two or three giant firms with 26-year-old entrepreneurs as their CEOs. I don\u2019t think I\u2019m going out on a limb saying that. The biggest concern I have regarding global competition is that the nation that figures out how to harness AI to improve the lives of all of their citizens will come out on top. The nations that refuse to do that and either bottle up the benefits of AI so that only 15-20% of the population benefits from it or the nations where large segments of the population reject AI when they realize they\u2019re being left behind (again!) will lose out completely. The United States is in the latter category. The same people who can\u2019t regulate banks, finance, education, pharmaceuticals and policing are in a very poor position to make AI work for all people. It\u2019s basic institutional social scientific insight.\u201d<\/p>\n\n\n\n Christine Boese,<\/strong> a consultant and independent scholar, wrote, \u201cWhat gives me the most hope is that, by bringing together ethical AI with transparent UX, we can find ways to open the biases of perception being programmed into the black boxes, most often, not malevolently, but just because all perception is limited and biased and part of the laws of unintended consequences. But, as I found when probing what I wanted to research about the future of the internet in the late 1990s, I fully expect my activist research efforts in this area to be largely futile, with the only lasting value of being descriptive. None of us have the agency to be the engines able to drive this bus, and yet the bus is being driven by all of us, collectively.\u201d<\/p>\n\n\n\nIt can be hard to agree as to what constitutes ethical behavior<\/h4>\n\n\n\n
Humans are the problem: Whose ethics? Who decides? Who cares? Who enforces?<\/h4>\n\n\n\n
Like all tools, AI can be used for good or ill; that makes standards setting a challenge<\/h4>\n\n\n\n
Further AI evolution itself raises questions and complications<\/h4>\n\n\n\n
Control of AI is concentrated in the hands of powerful companies and governments driven by profit and power motives<\/h3>\n\n\n\n