Can AI become Evil
Is there a future where AI turns against us?
I grew up watching a movie called “Terminator 2” which forecast a world where AI (or artificial intelligence) has taken over the world and to win the war against humanity, the AI sends a terminator from the future back to the past to kill the leader of the human resistance when he was young.
When I discuss the plot of the movie, a question might arise. How likely is it that AI turns out to be evil in the future, given the fast rate of advancement in the field of AI?
Nowadays, we are bombarded with news saying that AI has beaten some humans at a certain task.
One of those major news was published in 2017 when an AI called AlphaGo bested South Korean Go master Lee Sedol in part by learning from a lot of example moves by humans.
Later on, a newer version was developed called AlphaGo Zero which did not even need examples of human moves and became better than humans by playing against itself.
So, when we hear about news like this, it is not far-fetched to conclude that machines will beat us at everything in the future.
Can it be true?
My Opinion
First of all, let me clarify why my opinion even matters on this topic.
I have finished my Ph.D. in artificial intelligence, and my thesis was based on how to automatically design the architecture of neural networks.
In plain English, I was trying to design algorithms that can do the work that used to be done manually but at a faster rate.
I have been asked by many people on quora about how AI can become evil and what options we have if it becomes evil.
My answer to those questions is that AI cannot be evil and here are my arguments:
1. Evil is an abstract concept that humans made up.
For example, let’s look at a knife. A knife can be used to cut vegetables for eating and kill people. But based on the number of people that are killed by a knife, do we assume that knife is evil?
No, because we know that it is just a tool.
The same goes for AI.
AI is just a statistical toolbox used by humans to model certain tasks. It is neither evil nor it is good. It is the responsibility of human that uses it that makes the application evil.
You can use AI to solve world problems or create one.
2. We cannot build an AI that is good at everything
An AI that is better than humans at everything will be called “Artificial General Intelligence” or “AGI” and this is what is portrayed in most movies.
But there is a mathematical limitation called the “No Free Lunch Theorem” which states that no one algorithm can be the best at all the tasks present.
Yes, you will read many news articles about how an AI has beaten humans on certain tasks like playing chess.
But the next question that you should be asking is can that same AI do anything other than that task?
Conclusion
One thing that I realized after all these years is that we are afraid of everything unknown. I know that I was afraid when I started my Ph.D. journey as I was constantly bombarded with questions like what if I am not good enough, what if I don’t get a job etc.
But after going through the motion of completing the Ph.D., the fear has gone away.
The same psychology applies when it comes down to the question“Can AI become evil”. We are afraid of this question because many people do not understand AI and look at movies and TV shows as the future predictions of society.
My suggestion will be to read more articles regarding AI from academics or people who have technical knowledge.
