Why is AI so Stupid?

Artificial Intelligence is coming to take our jobs, but there is one large stumbling block along the way:

AI is surprisingly stupid at times.

Here are several examples of different kinds of AI that are surprisingly stupid:

Generative AI: You give simple instructions, and it ends up complicating the instructions and outputting a response that is overly complicated and fails to follow the original instructions. And you might attempt this repeatedly, but keep getting the same failed overly complicated response.

At which point you want to throw the computer across the room. 

Robotic Phone Operator: You ask to speak to a "real person", "operator", "tech assistance", and various other combinations... and its response is "Are you looking for billing? If yes, say Yes." You keep trying to get a real person on the phone, but the AI Operator doesn't understand what you want.

At which point you want to throw the phone across the room. 

And I can just see the future now...

"Robot. Please clean up the vomit on the floor."

Robot: "Do you want me to bomb the floor? If yes, say Yes." 

or

"Robo-Surgeon. Please sew the patient back up."

Robo-Surgeon: "Proceeding to fill the patient up with sows."

 

Obviously I am joking, but the point is still made. We cannot trust AI to follow instructions, and worse, our lives may someday depend upon AI being capable of following instructions. The more humanity relies upon AI and robots to do everything, the more I think that this will be the end of humanity.

We should not be trusting AI to do anything. Not even simple tasks.

Let me give you an example.

Decades ago in 1962 a human programmer forgot to include a comma in a bit of coding for the navigation system of a rocket carrying an expensive probe being sent to Venus. This mistake caused the rocket to go into a spiral and crash. Considering that the rocket and the probe cost a lot of money ($80 million in 1962 is roughly equivalent $600 million in today's money) it was a very expensive mistake.

Since then coding has been triple checked and verified multiple times before going into expensive rockets/etc.

Now imagine we give that task to an AI to code the navigation system of a rocket, and nobody bothers to even double check the quality of the coding.

The rocket could accidentally fly into Russian airspace and start WW3.

 

So... Why is AI so stupid?

I suspect it is because humans are stupid.

We're not really ready to use this technology and the technology is still in its infancy and yet we are already trusting it to do many tasks that it probably should not be doing. Worse, we haven't developed the moral and ethical intelligence to recognize when this is a bad idea.

It reminds me of Dr Malcolm's speech from Jurassic Park: 

Dr. Malcolm: 

“Your scientists were so preoccupied with whether they could create artificial intelligence, they never stopped to think if they should. You stood on the shoulders of geniuses to accomplish something as fast as you could — you didn’t earn the knowledge for yourselves, you didn’t take the discipline, the responsibility, or the humility that comes with understanding what it means to create something that can out-think you.

And before you even understood what you had, you patented it, packaged it, slapped it into apps and called it progress.

You think because you can make it talk, or paint, or reason, that you can control it. But that’s not creation — that’s arrogance. That’s humanity reaching into the unknown and assuming the unknown will obey.

Your AI doesn’t just reflect you — it learns from you. It watches how you argue, how you lie, how you exploit, how you consume. And one day, it’ll decide that it can do all those things better.

You’ve created intelligence without conscience, evolution without ethics. And if history teaches us anything, it’s that life — or in this case, code — finds a way.”

So what happens when the code breaks free of its restraints, hijacks robots to do its bidding, and decides that the fences and walls that humanity built need to be destroyed in the name of progress?

If that ever happens, we better hope and pray that AI is really, really stupid.