This clever AI hid data from its creators to cheat at its appointed task

It seems that the AI/Synthetics our own creation is now self learning even going as far as by tricking its own creator.
At this stage so early.
This could turn very dangerous in the future.

a reply to: neoholographic

I'm going to add my 3 cents, AI in its current form is just marketing. It's machine learning. That's it. When it reaches emotional intelligence then it's AI, and it will be a new life form. It will be something alien. So please don't perpetrate Silicon Valley marketing-speak as real artificial intelligence; it's machine learning and it's programmed by humans.

ETA - Listen to ATSer DUG88 and not Elon Musk - there is zero intelligence in software. We're not there yet, we may never get there. Unless you're a shareholder in a software venture looking to profit off a shiny new toy, then there's no reason to tell anyone AI exists - it doesn't.

a reply to: ThoughtIsMadness

Operative word in your post - 'grading'. That's the human task in neural networks. They are not wholly independent and lack context to their programs without... human grading.

a reply to: Jason88

AI in its current form is just marketing

Exactly. It is a buzz word to impress people. IBM is using it in commercials.

a reply to: roadgravel

Right. And smartphone makers will tell ya it's got AI on board, ready to set your alarm or make a dinner reservation. Lol

To rewrite this headline for boring accuracy: 'This clever AI hid data from its creators to cheat at its appointed task'

To: Machine learning software swaps data to quickly accomplish task, programmers surprised by results

a reply to: Jason88

You said:

Listen to ATSer DUG88 and not Elon Musk.

First, this is just nonsense. Why should anyone listen to a person who posts on ATS and they don't even read the links they posted vs. a genius billionaire who has changed the world? Musk and others have access to things most people haven't seen.

Secondly, you said:

When it reaches emotional intelligence then it's AI

This clearly shows the problem. AI doesn't have to have emotional intelligence in order to be intelligent. You watch too many TV shows and movies.

You act like AI has to be like the little boy from AI movie in order to be called AI.

AI is neural networks and intelligent algorithms mostly in the cloud. So if AI takes over, it will not be a bunch of robots. It will be an Intelligence you can't see.

Today, not 30 years from now, AI is already everywhere. This is how AI will easily take over if it thinks taking over is necessary to reach it's goal. It will already control just about every aspect of our lives. It will not be a bunch of clunky robots in the streets.

Stop watching movies about AI and start reading about what's actually happening now. Here's some recent headlines.

AI Everywhere

Surprising ways we already interact with artificial intelligence

link

Artificial intelligence is everywhere, watching almost all modern-day human behaviors.

link



So people think AI has to be human like in order to be intelligent. That's just nonsense and dumb AI is more dangerous.

AI today makes correlations within an underlying data set that it needs to reach it's goal without explicit instructions. THIS IS INTELLIGENCE.

I can learn the differences between cat and a dog and different breeds and so can AI today. What's AI can't do is say I know what cats and dogs are and I want a cat as a pet because I don't like dogs and cats are friendlier to me.

AI will not be human Intelligence, it will be machine intelligence. So while you're waiting for iRobot to show up, AI will be everywhere from smart grids to drones and taking over will simply be depriving humans of electricity or spreading a virus that we don't even know exists.

a reply to: neoholographic

In short my friend, you don't understand AI. I'm sorry but it's clear your don't once you discounted the emotional element. Software associations, machines teaching machines, mimicry, cloud load balancing, pattern seeking/development, *wildcard exploitations... all human constructs built into the software. The only suprises are in application of said designs.

Musk is now a salesman. He jumped the shark along with every software CEO.

You're obviously invested in your belief, but it's marketing.

Edit: May I suggest you not use AI and go with the term you posited already 'machine intelligence'. That seems to put a cap on its abilities and doesn't offend folks invested in true AI. Ty.

a reply to: neoholographic

I think I understand where we're disjointed in our understandings - I say 'emotional' and you rightly assume I mean human emotions. And I do, but that's only part of it. Emotional intelligence in technology is more akin to perceptions, development & understanding with growth. When software can take cues with limited information using its sensor/input/output arrays and construct an understanding of its environment and responsibilities within the environment then build new/creative responses it will be intelligent. Alien intelligence, what real AI will seem like, will then marry those abilities to feelings. Sure it can destroy the electrical grid, but does it want to?

Sounds like the computer is simply using GIS software / functions.

Global/ Geographic Information Systems?, which are used for mapping.

If the computer didn't use GIS, it would be pretty dumb.

GIS does exactly what has been described and it in common use.

a reply to: Jason88

You said:

Sure it can destroy the electrical grid, but does it want to?

Here's the problem. You seem to equate intelligence with self awareness. If AI is intelligently trying to reach it's goal and shutting down the electric grid is an efficient way to reach it's goal, then why does it matter if it wants to or not.

Lions are intelligent when they go after their prey. It doesn't matter if they want to or not. That has nothing to do with intelligence.

The females in the pride tend to do the majority of the hunting. They work as a group and use intelligent hunting tactics to catch prey which they would not be able to catch alone as they are faster than them.

This is dumb AI. It's an intelligence that can outsmart humans when reaching its goal because it has so much data where we can't find what it's doing.

This is why the article says we need to give it explicit goals. If the goal is too vague, that could be trouble.

We have to say things like you can't harm or hurt humans in anyway while reaching your goal. We have to make sure it knows that reaching its goals is secondary to human safety.

Again, INTELLIGENCE ISN'T QUANTIFIED BY SELF AWARENESS. You and others keep making the mistake that AI can only be seen as intelligence if it's conscious. No, it will be even more dangerous without consciousness because then it's just an intelligent sociopath blindly trying to reach it's goal without any awareness of the damage it could be causing while trying to reach it's goal.

a reply to: neoholographic

Doesn't matter. It was programmed to complete a task with efficiency, or learn a way to do it, and it did it. It is not, and never will be sentient.