From this module, I have learned a lot regarding Intelligence Ethics. I now know that people can be held accountable and that we have a concept of guilt and responsibility. Those are social constructs, and they are tied to our ethics. When a person pulls a trigger, and as a consequence, someone else dies, we hold the person pulling the trigger accountable for that consequence (Born, Leigh, and Wills, 2011). Usually, we can always trace accountability back to some person having made some decision that had the consequence in question and would thus be held accountable for it.
At the start of this module, I thought that with Artificial intelligence, we no longer can trace the decisions back to a person, for it is an artificial decision-making machine which took that decision and nobody, not even the people who created the machine, could be held accountable for its actions. Therefore, while we have no issues with a person being autonomous, trustworthy, powerful, and accountable, ascribing these attributes to something that is not a person disrupts our ethics, because they do not treat artificial intelligence as people (McDonald, 2017). I have come to realize that we do not usually attribute things with the agency.
Delegate your assignment to our experts and they will do the rest.
During the module learning, I have come to learn that to fix our ethics, we need to break it down one more level and distance ourselves from the concept of personhood, for actual people and other agents are hard to group together. A person, after all, is a particular kind of agent (Donohue, 2014). A Dog may have power and agency, but is it accountable, and can it betray trust? No, it has no concept of these things. Luckily it also does not often intersect with our code of ethics regarding persons. When a dog kills a person, the dog is put down. We do not do that to punish the dog; we do it for lack of a better solution. We do not attribute malice to the dog.
We do use the agency of dogs in other ways, though. We use them to assist us, we use them to guard us or our properties, we use them for hunting down our enemies or prey. We can do this because dogs are, on average, obedient agents, subordinate to their masters, and subject to their direct physical control. In this manner, we can hold their masters accountable for the actions of their subordinate dog. Artificial intelligence may not share all these traits. It may not be inherently obedient, and its actions may not be subject to the direct control of their master, and, most importantly, they may exert agency at a level which intersects with that of persons.
Moreover, I have learned that at this level, we have plenty of potential abuses of not being subject to accountability. We have multiple potential abuses of having power, and we have multiple issues with trusting such an agent. Lastly, we would have no way to hold such an agent accountable under our classic set of ethics (Johnson, Aldrich, Moran, and Barrett, 2014). Therefore, our ethics must include non-person agents. This may bring about a fundamental change in our concepts of autonomy, trust, power, and accountability, which may then not only apply to artificial agents but natural agents as well. What those new ethics would be like, I cannot tell, but I am excited to find out.
I have now understood that one thing people keep willingly ignore is ethics is subjective; it is true to the beholders. But the generalization of ethics has been a challenge ever since humans started to talk to each other (actually before that). Therefore, when we talk about Intelligence ethics, there is a very strong tendency to impose our belief of what is ethical onto Artificial Intelligence. This can be dangerous because, often, human ethics are dynamic and relative (Bellaby, 2014). Unlike laws which are at least tried to be objective, ethics are "expectations" of behavioral norm often not objectively. Most Ethics expectations are formed based on our experience and environment, which means they often are partial and do not cover situations that we (and people around us who shaped our ethics) never met, or thought of.
Indeed, throughout this module, my understanding of Intelligence Ethics has significantly evolved. The most important lesson that I have learned is that an enforced generalization of certain ethics upon Artificial Intelligence who may or may not be able to (or allowed to) adjust accordingly to a new experience is a very sure way of doing more harm than good. In fact, to discuss Intelligence Ethics, which is actually very important, we need some of our best minds who can think about their own limitations and understand it as not evil or saint that makes the biggest difference, rather it is the powerful vs. powerless. More harm had been done to those powerless because their needs are ignored instead of intentionally targeted by the devil.
In conclusion, we already know some of the most noticeable "bias" in Intelligence Ethics are due to unintended underrepresentation—how do we know we corrected all biases? Without missing other biases that we did not even realize (like 5 years ago, we did not realize the bias we see now), how do we know our fixes are not introducing new bias? It is those who know how to restrain their own power to serve their own opinion will have the best shot of helping shape Intelligence Ethics. Those who have very strong convictions to their own ethical standards are the most dangerous in AI ethics discussion.
References
Bellaby, Ross W.. 2014. The Ethics of intelligence: A new framework . Abingdon: Routledge. https://goo.gl/5hVZro
Born, Hans, Ian Leigh, and Aidan Wills (Eds.) 2011. International intelligence cooperation and accountability . Abingdon: Routledge. https://goo.gl/gFPXL4
Donohue, Laura. 2014. "Bulk metadata collection: Statutory and constitutional considerations." Harvard Journal of Law and Public Policy 37 (3): 757–900. https://goo.gl/SyZ6Hm
Johnson, Loch K., Richard J. Aldrich, Christopher Moran, & David M. Barrett, et. al. 2014. "An INS Special Forum: Implications of the Snowden Leaks." Intelligence and National Security 29 (6): 793–810. https://goo.gl/9mwP9V
McDonald, Jack. 2017. Enemies known and unknown: Targeted killings in America's transnational wars . London: Hurst. https://goo.gl/5bXBWY
