Casper Wilstrup.

We’re living in an era buzzing with artificial intelligence (AI) chat. There’s one big worry though. A lot of AI is like a magician’s trick – we see the result, but we don’t know how it’s done. It leaves us guessing.

We call these ‘black-box’ AIs. They give us great predictions, but they keep the ‘how’ a secret. This mystery makes it hard for scientists to use AI to come up with new ideas.

There is a bright spot. A field called Explainable AI (XAI) is trying to crack open these black boxes. Popular methods like SHAP and LIME are like detective tools. They snoop on the AI, helping us understand what’s going on inside.

Still, these tools only go so far. They watch AI behavior but don’t really get to the heart of how it’s done. So, we’re thinking of a new strategy, called Symbolic AI. It’s about making AIs we can understand from the get-go. Like a math problem, they show the whole working process, not just the final answer.

AI safety is big news. Famed AI researcher Eliezer Yudkowsky worries that unchecked AIs could be bad news for humans. Picture an AI so focused on its goal that it doesn’t care if it harms people to reach it. Scary, right? Yudkowsky says we need to make sure our AI has safety locks and follows our values.

But how do we make AI safe? We think the answer is simple. Let’s use AI to build understandable models that humans can check for safety. We stay in the driver’s seat this way.

Here’s our four-step plan:

1. We build smart AI “scientists” to come up with new ideas.
2. These AI “scientists” create models we can understand.
3. We make sure these models work as they should.
4. Finally, we use these models to make decisions in our AI systems.

This way, we get the best of both worlds – AI’s power and human control.

Now, let’s imagine what this could look like in real life.

AI doctors could use models to diagnose diseases and recommend treatments. AI cars could drive using models that represent traffic and road conditions. AI factory workers could use models to build things more efficiently. AI builders could use models to ensure buildings are safe and long-lasting.

AI is developing fast, and yes, it can be scary. But, it’s not going to stop. So, we need to make sure it’s safe. Our four-step process could be just the ticket. By focusing on clear models, we keep control and make sure AI sticks to our values.

Working hand-in-hand with AI in this way gives us the best shot at a safe, tech-powered future.