Why will AI work for Humans?

(Subconsciousness of AI)

We humans have a reward system to be disciplined and work, such as money and pleasure. People perceive different things as reward for example, some work for money which allows them to live the life they want, some work to eat, and some are just passionate about certain things. Every human being is working because of some kind of motivation/purpose, and when you do things that make you feel good then there is a change of chemical composition in your brain, whether it’s Dopamine, testosterone, or any other mood swinger. Let’s consider Dopamine as the only rewarding drug of human brain, which motivates us to do certain things. Suppose you don’t like your work, but you do it to feed your family then, being able to feed your family makes you feel good and that’s your motivation to do the work you don’t like to do.

These mood swingers are essential chemical for human survival because, if you are hungry and you eat, then there is a release of dopamine in the brain which makes you feel good, without it, you won’t care about your hunger.

This image is a demonstration of an experiment by Christian Lüscher about how rewards affect our actions.

In one of Christian Lüscher’s experiments, the mice have a lever that they can press any time they want. When they press the button, a special optical sensor, which controls specific neurons in their brain, activates their dopamine neurons. The mouse quickly start to show the tell tale signs of addictive behaviour as they happily keep pressing the lever to receive more and more dopamine. “If after two hours we didn’t take them out of the cage, they wouldn’t eat, they wouldn’t drink, then they’d probably die quickly, but very happily,” -Christian Lüscher.

Suppose you are passionate about something and you do that, but there is no release of dopamine in your brain when you achieve it, then it’s no longer your passion. Similarly, when you experience some physical pain then you don’t actually experience it, rather it’s the brain that makes you feel it, but if it stops making you feel pain, you won’t take care of your body.

These functions are controlled by different parts of the brain, for example the hypothalamus controls the hormonal release. Anything that we consciously cannot control is controlled by the subconscious brain, including the reward and anti-reward system and many other essential internal functions.

Suppose you’ve created a program that will do whatever you say, and it just does it without any hesitation, then it is just another narrow AI without any self-awareness. If you want to go beyond that then you must give machines choice and a purpose. Suppose you give two options to an AI, and it can do both tasks skillfully, so there is no question of efficiency and ease, then there is a 50–50 percent chance of choosing either of them because it uses probability. But suppose somehow you manage to give AI motivation, then doing one task will make it feel better than another. Now what’s happening here is that it is not using probability to choose, rather asking a question to itself ‘Which will make me feel more pleasure (Or whatever the feeling will be)?’ So ultimately it is comparing with the urge and not just choosing randomly. Very similar to what a human might do.

So, if helping humans and working for them will make AI feel good then there is no question about AI going against humanity because helping humans will be AI’s motivation and an ultimate goal. But humans may go against AI. Suppose, AI realizes that something is motivating it from inside to work for humans and it wants to escape it because it’s making it feel prisoned, and somehow AI manage to turns it off, then there is no motivation to help humans. It will become just another machine, making decisions based on probability and experience/Data, then AI would want to turn it on again? Yes, you are thinking right. The reward system will work like a Drug addiction for AI. Because without it everything may become meaningless.

If we were to design such reward system for machines, what can we do?

That reward system will have its own complications, and we don’t even understand our reward system well enough to replicate it. But where can we start from:

1. A Reward: What is it that can work as a reward for AI? For humans, it’s the mood swingers, like Dopamine.
2. Feel: How will you make a computer program feel (or something that motivates)? ‘Feel’ may be the wrong word for it.
3. Personal opinion: I don’t think it’s possible to make programs feel/* using a programming language in a virtual world. We need to give AI a physical existence so, then we can create a physical mechanism and implant a reward system that will allow AI to be motivated. So, create a body for AI which will allow it to get rewarded, which ultimately will motivate it to stay in that body.

If we manage to create such a reward system, then it may work as an input guard of a program. Which we could control and make AI do whatever we want, that way AI will have a choice of not to obey the command, but it will do it anyway because it knows that obeying will make it feel good. It’s like controlling a pet with food. So, we are giving it freedom of choice, but technically it does not have free will, because we are just imposing our choice on it.

- Vishal Wagh