In recent years machine learning has been making inroads into what used to be human domains. However, Microsoft’s recent experience with its chatbot known as Tay revealed something every more interesting about what it means to be human and/or even attempting to mimic human.
While natural language processing has been the biggest challenge for computer scientists and artificial intelligence researchers alike, in recent memory it is increasingly become evident that machine learning can overcome earlier limitations. While this may be largely because of improvements in hardware and software designs and tools, it has nonetheless open up many possibilities. It is from these advances that systems like Watson, Siri, Google Now, Cortana etc have made into the market. Machine learning, combined with machine visionary has given rise to autonomous vehicles with varying capabilities.
However, while these systems are taught to learn a limited set of rules e.g. the structure of sentences in order to extra meaning or traffic rules in combination with detection and identification of obstacles, it has become increasingly obvious that NPL works within the broader context of human society which means that many of what passes for rules are neither hard and fast nor rigidly imposed. Humans are equally adept at processing norms and mores as they are in processing grammar and traffic rules. More importantly humans also possess the ability to create their own meaning and context because of what the possible fall out of not adhering to social norms would be.
Tay was able to learn but could not project it’s own objective in the conversation in which in participated; that simple fact made it easy for people to have it say things that are not polite. That is not the end of the story though: this entire episode only serves to highlight the fact that understanding a sentence does not mean appreciating the impact of the sentence on an audience. While many, including the designers and developers of Tay, will definitely be looking at what went wrong, the more interesting part is that perhaps we need other disciplines to enter into the effort of adding more nuanced abilities to AIs that will have to operate along side human beings.
At the moment, the main point of departure for most of the expanded effort to integrate AI into human society is laws and ethics. However the proper point of departure should be building in more abilities for AIs to appreciated the nuances of human society. The truth of the matter is that we as a species are not great at being governed by laws and/or being ethical. It is a matter of appreciating human nature as it relates to laws and ethics generally. As many have noted the greatest risks that AI (in their current form) pose is from human beings themselves.
The more worrying aspects of not getting the law and ethics angle of AI right in the first place is that it will tap into the disturbing human capacity to feel less responsibility for actions and decisions made on your behalf. Very few people have the conscience to question, at the very least, laws that are clearly unethical.
Microsoft Tay was an experiment first and foremost though unleashing on the greater internet may have been premature in the process. That misstep is a testament to the fact that there is gradually more confidence in the abilities of machine learning algorithms. The embarrassment created by Tay seems to suggest that people increasingly related to AIs as if they are people, in a general sense, though it is more likely a question of wondering what Microsoft was trying to do.