Let's talk about Artificial Intelligence

[old man emu]

Are we afraid that someone has let AI out of the bottle and it will turn into something like the water-bearing brooms of the Sorcerer's Apprentice? Are we afraid that AI will self-replicate and eventually create bodies for itself and we will be under attack from Dalek-like machines?

[rgmwa]

I think that would be the least threatening possibility. It's what it might do when it's turned loose deliberately or accidetally on the interconnected systems and devices that we all use these days that would be the bigger threat. Banks and other financial institutions, public facilities like hospitals, power stations and distribution networks, the military etc., and especially if nobody knows that it's there or what it's doing.

Edited yesterday at 10:55 AM by rgmwa

[Marty_d]

There was some discussion on RN's Future Tense about it today.

One presenter made the point that pre-AI, hackers needed a fair amount of knowledge to do damage, and still do it at human speed.

Soon there will be the ability for anyone to buy off the shelf AI hacking tools, which will work thousands of times faster, with no knowledge whatsoever.

[old man emu]

13 hours ago, rgmwa said:

think that would be the least threatening possibility

What you describe might just well be the first incursion. What I was thinking of is an ultimate result if it was turned loose in some way. 

 

12 hours ago, Marty_d said:

off the shelf AI hacking tools, which will work thousands of times faster, with no knowledge whatsoever.

That's the point. AI tools work at the speed of light (?). The horse could have bolted before we knew it was thinking of bolting.

[red750]

13 hours ago, Marty_d said:

Soon there will be the ability for anyone to buy off the shelf AI hacking tools, which will work thousands of times faster, with no knowledge whatsoever.

This is being advertised on Facebook for US$59.00.

 

https://www.artspace.ai/lifetime?utm_campaign=120238296206500490&utm_content=120240627676020490&utm_id=120238296206500490&utm_medium=paid&utm_source=fb&utm_term=120238296206490490

[red750]

Here's another...

 

https://pyxa.ai/?utm_source=facebook&utm_medium=cpc&utm_content=8+Sep+Viral+marketing+-+2+Video+ads+%26+1+Image+ad&utm_campaign=ABO+New+Interests+19May&utm_term=broad+ImageAD+1&utm_id=120225417666210542&ad_id=120232826134050542&adset_id=120232826134020542&placement=Facebook_Desktop_Feed&site_source_name=fb

[rgmwa]

1 hour ago, old man emu said:

What you describe might just well be the first incursion. What I was thinking of is an ultimate result if it was turned loose in some way. 

The damage that could be caused by the first incursion would likely make your ultimate scenario unnecessary. 

Edited 16 hours ago by rgmwa

[old man emu]

15 minutes ago, rgmwa said:

The damage that could be caused by the first incursion would likely make your ultimate scenario unnecessary. 

Two different, but possible scenarios. 

 

However we don't need AI to put the financial or military structures at risk. We' got the orange A-hole creating the risk.

[facthunter]

red. how can you be sure that Link is safe? Nev

[red750]

I have had Avast antivirus software installed for over 10 years. It regularly blocks dangerous sites, but accepted these ones when I opened them. Avast also regularly runs system checks to look for dangerous files.

[Jerry_Atrick]

On 27/07/2026 at 4:45 PM, rgmwa said:

How does it decide how the probability maps should be altered to get a better result?

Sorry - I didn't see this...

 

There are two ways - human and machine learning. And Machine Learning is often automated feedback loop between humans and machines, but can be, for example, response to external stimuli.

 

Human learning, or tagging, is where a human takes the output and if it is inaccurate, explicitly tags the output to the degree it is inaccurate and corrects it. The data held to calculate the probability will be updated as a result. 

 

For machine learning, it is where the machine works it out for itself. For example, if a LLM answered your prompt, but answered the wrong question and you corrected it, it would update the statistical data where it think it went wrong. For non-human interactions, and to take an extreme example.. Say a self driving car predicts the person on the side of the road is not moving in front of it (say moving away from it) but ends up hitting it, it will use the data it received through the camera and radar to adjust the statistical data and learn whatever triggered it "thinking" or statistically deducing it would move away from the car, actually meant it would move toward the car and update its map accordingly... automatically. 

 

This is an oversimplification, as it would feed some central AI DB that would recompute the data maps and download it back or something similar. But the concept should be clear enough. 

 

 

On 31/07/2026 at 10:03 AM, old man emu said:

Are we afraid that someone has let AI out of the bottle and it will turn into something like the water-bearing brooms of the Sorcerer's Apprentice? Are we afraid that AI will self-replicate and eventually create bodies for itself and we will be under attack from Dalek-like machines?

It isn't inconceivable.  But it would require almost autonomous and dextrous robots to do so, and I don't think we are quite there yet. In the future though, it may well be the case.

 

On 31/07/2026 at 1:22 PM, Marty_d said:

One presenter made the point that pre-AI, hackers needed a fair amount of knowledge to do damage, and still do it at human speed.

Soon there will be the ability for anyone to buy off the shelf AI hacking tools, which will work thousands of times faster, with no knowledge whatsoever.

Without context, the presenter is not making an entirely accurate statement. Yes, AI will work a lot quicker and uncover more vulnerabilities. But firms and governments are already using  (and have been for some time) AI to scan networks and systems looking for and patching vulnerabilities. In addition, AI has been deployed to monitor real time behaviour to determine id actions being performed are likely to to be cyber breaches of some sort. So, it will settle down to a bit of like it is now, where vulnerabilities are found by those looking for them from both from a defensive and offensive/nefarious point of view and, like today.. some of the nefarious ones will get through, and some won't. The good thing is that AI will be a far better defence - especially monitoring real time activity that is suspicious. 

 

Also, the number of "hackers" (crackers - hacking is what I used to do - programming with "short cuts") that really know the nuts and bolt of what is happening is a relatively small percentage and most of the systems/networks hacking is already available in automated tools - usually free - that can be downloaded. Khali Linux is a cybersecurity "optimised" version of Linux for penetration testers. Basically it comes with most, if not all of the major available hacking software. Not many of these 16 year-old bright sparks know much more than how to download Khali Linux and point it to a server and run the software.. a real cracker would know how to cover their tracks and not get caught - like some of the ransomware gangs...  So, this is already available today. And note, at least in today's world and I suspect when the AI between nefarious and defensive actors catch up, the weakest link in cybersecurity is humans.. social engineering (phishing, vishing, smishing and the like) are by far away the most common cause of breaches rather than the stereotyped hooded cracker at a keyboard. Real time AI monitoring will be much quicker at finding this (and already is). What will happen is you Windows Defender will become AI enabled. 

 

I found it ironic that Hugging Face, an AI cybersecurity start up was breached so easily. I think that may have damaged its reputation somewhat.

 

The bigger threats on the horizon will be with quantum computing. This is a game changer.