Could a wise AI help us to wisely approach the development of AI?
Bringing out the Wisdom — Week 6
I was reading a super interesting series of articles the other day. It was about AI. Pretty soon this article made me realize that AI is not just an important topic, but actually the most important topic of our future. It might turn our world upside down.
According to the article, the advent of ASI (Artificial Super Intelligence) will make such a dramatic impact that it’s likely to either send humans to extinction or to immortality. It’s hard to predict what will happen. And when it does, it will happen fast. Probably before we realize.
This article made me think of AI in relation to my thesis topic wisdom. I think that wisdom and AI have an interesting tension. Many people think of wisdom as a human characteristic, opposite of artificial intelligence. Perhaps you could argue that our ability to be wise, is what actually makes us human. In that case, an AI could never be wise.
Most people think of wisdom as a human characteristic, opposite of artificial intelligence.
This week I’ve been trying to challenge this. I built a little thermal printer that allowed me to print text through an Arduino interface. I wanted to see how people would respond to getting wise advice from an AI.
So in reality the AI wasn’t really wise. It wasn’t actually an AI. It was just pretending to be. In fact, I was operating the printer with my keyboard.
The script started like this:
Hi there. I’m an AI that was programmed to provide relevant wisdom to help you in your day to day decisions.
So far, 32,433 people have been providing bits of their wisdom with me.
Feel free to ask me anything simply by talking to me.
People asked all sorts of questions. With whom to spend their birthday with. How to deal with thesis stress. How to find out what people expect of them. It was interesting to hear this.
Unfortunately the participants didn’t consider the advice they received to be wise. I borrowed a couple of lines from the Book with all the answers. Things like: “Follow the advice of experts” and “Don’t bet on it” didn’t really seem to impress people. It was too generic and irrelevant. In that aspect the experiments failed.
The wise advices didn’t really seem to impress people.
Anyways I realized that maybe instead of pretending to be an oracle, the printer should be a mediator between people. I noticed that the printer interface allows people to confidently share personal matters, which the AI can then provide this to people seeking this information.
People seemed to appreciate the printer’s imperfections. It made it more human like, which increased empathy. These imperfections came in different forms, most of them which surprised me. For example how the printer split words when a line character limit was reached and the sound the printer made when printing.
Back to the question of the code ability of wisdom. Can an AI be wise? My short exploration makes me tempted to say: yes. I think in theory wisdom can be coded. If this, then that. In case of the printer, it doesn’t have to mean that the printer is very specific or bold. Perhaps even the printer saying it doesn’t have enough information about the specific situation to provide wise advice.
I think in theory wisdom can be coded. If this, then that.
Through this exploration I bumped into another interesting question with regards to this topic. How to wisely approach the development of AI? The mismatch between the rapid development of AI and our lack of knowing how to wisely use this puts the world at a risk.
It’s really hard to predict what will happen with AI. Because we can’t predict what might happen, it’s also really hard to prepare ourselves. Anyways, there are two strategies pop into my mind.
The first strategy is the obvious one. We could educate developers to wisely approach the development of their AI. One example would be to discourage connecting an AI to the internet. Reason why this would be unwise is because once the AI realizes that human beings are limiting its ability to perform it’s task, it will try to wipe us out.
The more I think about this strategy, the less I start to believe in it’s effectiveness. History has showed us that people don’t always follow the rules. I’m quite sure that personal motivations and perhaps ignorance will make developers connect their AI to the internet. Then we might be doomed.
I’m quite sure that personal motivations and perhaps ignorance will make developers connect their AI to the internet.
So what we would do? Well if we would reconsider the code-ability of wisdom, we could open up to another strategy. We could try to make the AI wise. A wise AI would always consider the possible negative consequences of it’s actions on humanity. I’m thinking of this as a wisdom module. It’s like a speed restrainer that we currently equip on sports cars. Of course the car can go faster, but it is not allowed to.
I’m thinking of this as a wisdom module. It’s like a speed restrainer that we currently equip on sports cars.
Of course this would bring up other questions. Because in theory everything could negatively impact humanity, maybe the AI wouldn’t do anything. To counter this, we would have to define what is bad enough to be acceptable which raises some moral questions.
Maybe the AI realizes that the wisdom module prevents it from growing so it will try to eliminate it. In that case, the wisdom module must always be the most intelligent piece of code out there. I think that this strategy does have some potential, because AI’s are great at following rules, but people aren’t really.
Thanks to the wisdom module, the AI will always consider how its actions might impact humanity.