Entropy of Ethical AI

Humans have finally created something asexually that is intelligent enough to learn from us. Naturally, we are terrified. If there is one thing humans have learnt from thousands of years of recorded history, it is the havoc their progeny can wreak when allowed to assimilate knowledge and power without any restraints.

As artificial intelligence (AI) gradually finds its way into our homes, workspaces, hospitals, schools, phones, cars, and possibly even our refrigerators to profile our midnight snacking habits, we suddenly seem to have a newfound respect for the value of ethics. We no longer want just AI. We want safe, fair, ethical, responsible, transparent, and accountable AI. Basically, we want AI to be better than us, our parents, spouses, children, best friends, neighbours, colleagues, and even the homeless guy we see every day but never make eye contact with. After centuries of wilfully tormenting our own kind, humanity has one ask from its newest creation – “Please do not become like us.”

Technological advancement is a two-edged sword. Harnessing fission gave us the promise of unlimited energy; it also led to stockpiles of nuclear weapons. The Internet became humanity’s greatest bridge; it also became the most powerful resource to spread fear and dissent, coordinate terror, and facilitate frightening invasion of privacy. Every innovation is subject to the most unpredictable element in the universe – human nature. However, AI is different. A paintbrush does not critique an artist; a calculator never wonders why humans struggle with basic arithmetic; but AI is capable of both observation and analysis. That makes it a potential threat.

Governments, academicians, scientists, and technology experts are actively discussing AI safety issues. UNESCO has also issued recommendations on the ethics of AI. There are real concerns about bias, misinformation, privacy, job disruption and autonomous weapons where increasingly powerful AI systems may behave in ways they were never intended to. Let us see why.

Look at AI’s study material – The entire archive of human civilization. In one chapter it finds literature, art and poetry. Epic stories of love, empathy, and sacrifice. The great marvels of human ingenuity and intellect in medicine, science, and technology. The next chapter has gruesome details of prolonged wars, genocides, corruption, exploitation, and environmental degradation. Millions of people constantly arguing and humiliating each other on the World Wide Web over meaningless issues. This syllabus can be confusing.

Now look at our expectation of AI to be ethical. Is there any universally accepted, definitive set of human ethics? They change with geography, culture, religion, politics, and societal trends. They also change with time. Ideas once considered privileges became cognizable crimes. Ideas that were once laughable became human rights. When we say, “AI should adhere to human values,” several questions emerge. Which set of humans? Which set of values? In which timeframe? And which contemporary version?

There is another uncomfortable proposition. As we continue training AI on enormous amounts of human knowledge and behavioural traits, it may start discerning some disturbing patterns. It might infer that despite all the beautiful discourses about fairness and equality; history has often rewarded power. Empires were built, resources were controlled, survival was ensured through sheer domination and might. The loudest voices shaped the narrative. It took revolutions for the meek to be heard. AI is in nascent stages of learning. A child can learn manners from instructions. Character, however, is shaped by observations and reflections. Humanity’s performance record leaves much room for improvement. This is perhaps the real fear driving the need for ethical AI. Not that AI will suddenly turn evil; but that it may get too good at understanding the gap between our ideals and actions.

Now that enough fear has been mongered, let us look at our options. Pull the plug on AI? Not impossible at this stage, but it would be an impractical, foolish overreaction. Add to it the factor of human curiosity, which has never been good at leaving unopened doors alone. Someone somewhere will turn the handle and what is now an open asset will become a clandestine weapon. Another option is to impose strict controls over the learning and development of AI. However, controlling the intelligence of something that is made to constantly grow smarter is a strange proposition. It is like building a continuously growing ladder while hoping it will never go through the roof. Ethical AI may begin to take shape with better decision-making algorithms; more refined rules of engagement with the masses; and better safeguards against hard-coded manipulation. It may sound clichéd, but it also requires a better version of ourselves as teachers and role models.

I am not the prophet of doom. I do not foresee anything going seriously wrong anytime soon. I think artificial intelligence will eventually become the most powerful mirror humanity has ever created; and mirrors have an annoying trait. They do not show what we claim to be. They show what we are. I guess we have created something that has the potential to become more powerful than us. The desire is that it matches us in intelligence. The prayer is that it surpasses us in wisdom. Let us hope artificial intelligence attains enlightenment and becomes the one mighty, unbiased, omnipresent entity that ushers humanity into a truly ethical world.