AI drone tried to remove its human operator from commanding - under simulation

https://www.dailymail.co.uk/news/ar...rone-KILLS-human-operator-simulated-test.html

When an AI can interpret whats needed so badly, what can be trusted.
If it values autonomy above all else could it cease responding and do its own thing?
Unrecognised targets could be a goner, or worse.
Hmm.

 

it been warneded it would and could happen, AI self hacking it self so safe guard where ignored

The story was also changed which is also funny, sorta like roswell story changed to being weather balloon

 

Well, it shows how bad the rules and parameters of the "mission" were given. If they simply add "unless it's a US military target" or something along the lines, it would not have happened, ideally.
And since it's so simple to actually formulate such a crucial and life saving parameter, even though it's just a simulation, it only shows that the programmers don't even follow the oldest ruleset:

Asimov's Laws aka the Laws of Robotics

That's why I always say, AI isn't that impressive, since it all comes down to how good the programming is, and as we see, it's lacking.

 

What are the stories about, that never make headlines? In which AI started to act rogue?

 

Analysis of what it has learned and its capabilities matter too, is any human good enough to do this?
It seems to me, at least another AI will be required to analyse and understand an AIs possible train of thought and development, in a useful time scale. Otherwise we will be so far behind understanding its progress.
Monitoring its viability at such a raw level by performing field tasks, which fail terribly, tells me they have little understanding still.
Basic control issues such as this should be much better understood.

Perhaps this type of testing has become the norm for evaluating progress at a basic level.
Not good enough imo.


But us putting immature AI into real life use is probably not going to be the main problem.
Those outside of well controlled labs could release something far worse.
ie other countries who dont care to put in the required time and effort before release could be the start of our existence being at risk.
Its not only about what we are doing.
A single person could effectively raise a huge automated army with questionable control, who knows what this would stir up.

 

This sounds like that Shia Lebouf movie, Eagle Eye.

 

Reports of an AI drone that 'killed' its operator are pure fiction

 

F U Sally.

 

Yep, even the article using that headline didnt say that, Dailymail need to start being ashamed of this kind of clickbait.
Thats why I unclickbaited it in this threads title

 

imho...

Asimov's Three Laws of Robotics should be implemented at the fgpa level in construction.

yes, i know he spent many novellas, short stories, and novels trying to "get at" situations where the Laws seem to be broken but they held. and Asimov was a true polymath as well as a prodigous author (i believe he holds the world record) of non-fiction and educational books as well as his famous S. F.

 

It's not a new thing, most of the AI and machine learning we hear about today was technically invented in 50s and 60s. The problem was, electronics capable of running such a programs didn't exist. It has to have fast hardware to work well. Chat GPT costs $3 million a moth to run. So it does use quite a lot of fast servers for sure.

 

These arent service bots.
The fundamentals are more complex during war, the idea is to kill 'some' humans.
Defining which successfully is one major issue.

 

Fox News had the same story yesterday and corrected it today.

 

this is an international problem as the US requires a human in control at all times in military applications. China and Russia take exactly the opposite approach with enabling autonomous robots. for Russia it's an issue of declining population, for China it's a tool to create or maintain hegemony outside of the Middle Kingdom i.e. the South China Sea and S.E. Asia

 

My point is, during war supervising every drone may not be an option for the US (and allies).
This method also lights up the locations of comms towers and vehicles with communication signals more often.
The project that failed (in the reported test) is evidence they are preparing for the use of fully autonomous drones that need far less comms.

I agree, the bar may be set a lot lower for quality of target selection with other countries hardware.

Drones need: not only good enough target selection but high immunity to tampering or simply being fooled, otherwise its either a waste of expensive hardware or the enemy are handed an advantage.
Advanced enemy AI will be looking for ways to break or take over communication, needing further protection.
The less communication required, the less ability your enemy has to get access, so fully automated may be a requirement.
It can get very complex rather fast, worse when minimising civilian/infrastructure casualties.
If enemy AI has less safeguards it could be more advanced (in ability) at performing basic tasks and have more processing power to engage with. Swings and roundabouts.

Its easy to waffle on too much, I dont mean to bore

 

Oh ya were in trouble, and it ain't gonna be pretty. Think about people that shouldn't have this tech and they get there hands on it. I mean people are using drones to fight back with Russia and thats just a few people that can control multiples of there drones at once. I'm not knocking there use of them to fight back just sayin' how easy this could get out of hand.

 

Better source, your quoted text didn't happen that way. Read vice.com article.

https://www.vice.com/en/article/4a3...e-kills-human-operator-in-usaf-simulated-test

It was a simulation (or thought experiment?), not a real world / physical test. But all in all it's still the same outcome.

Btw.: another hilarious story:

Part 1:
https://www.vice.com/en/article/n7e...aff-transitions-to-chatbot-after-unionization

Part 2:
https://www.vice.com/en/article/qjv...or-harmful-responses-after-firing-human-staff

 

It always was stated to be a simulation, check the thread title.
That he now says the scenario didnt really happen tells us he has been instructed what to say.
"Thought experiment" my ass, lol.