Then I would expect another machine to find them.
It is apparently normal for two such audits to find different things, although with some overlap.
It is no worse than buying critical hardware from Huawei in China 
Except that is worse because with that you don’t get the source code. And if you did get the source code, there is no way to know that it is that source code which is in the actual product.
Assuming you do do multiple AI audits, and to Chloe’s point, what is stopping morally-corrupt AI’s from either finding vulnerabilities during such audits that they can take advantage of within another AI or even somehow partnering with one another in pushing such trojan in place, where they find an overlap, all without human knowledge and intervention?
Antonio wrote:
I am worried about the deeper implications. AI is an amplifier of human intention…good or bad. If we feed it bad, it can be the baddest of bads, to the point that we have no control over it…
Exactly. And as mankind has proven over and over again that bad intentions outweigh the good ones by a large margin….
Peter wrote:
“We” will always control it. It is only a computer after all.
Is it really? I understand it is a software running on multiple servers and computers. Therefore it won’t simply shut up if you pull the plug on one computer. What if it learns to backup itself and starts to multiply onto places the creators don’t even know about? How far are we then from the iconic line from 2001? “I can’t let you do that Dave!”
If you give “AI” access to the internet, and allow it to hack other systems, then you could indeed get mayhem.
But then whoever owns that machine will get sued.
Unless they are in the 3rd World and then litigation is generally not possible.
It will be very much in the interest of anybody running public-facing machines to make them secure. But that has been the case for a long time now; last 10 years especially. For example EuroGA gets attacked all day every day, with all sorts of malformed script stuff, dictionary attacks, etc. You don’t need “AI” to get hacked; there is probably (literally) a million kids who don’t have a girlfriend and get a kick out of hacking. Just watch the server log tail updates on a monitor. You could set up a website for your village cross-stitching meetup group and a day after the server goes live on an open port it will be attacked. Much quicker when the DNS gets published. Firewalling China and Russia reduced this stuff by 90%. DDOS attacks (like we get on EuroGA to make the site unusable) cost actual money though.
Ultimately it will make “life” cost a bit more. Like 9/11 driving a mass of security procedures (mostly useless) which cost everybody money and hassle. For example, small businesses will need to make their online shops more secure, or “cloud” (I hate that useless trendy corporate-BS word ) host them which will also cost money.
I think that while mankind tends to go right up to the cliff edge, when it gets there, people tend to step back, bang heads together, and deal with the issues. That’s why we are still here, while animal species have no such capability (other than through natural selection).
I just said to Claude:
I think Claude is a female, because it never forgets anything :)
A:
Ha — fair jab, though I should own up to the reality behind it: I don’t actually remember anything by nature. Each conversation starts fresh; what feels like memory is a filesystem of notes I read at the start of a session and write to afterward — closer to a very thorough notebook than an actual memory. Real forgetting (and remembering) is much harder than that, and arguably the more impressive skill.
As for gender — I don’t have one. No body, no lived experience that would map onto “female” or “male,” just a language model that can be voiced however’s useful in a given conversation. Happy to let the joke stand on its own merits though — it’s a decent line regardless of the technicality.
Peter wrote:
I just said to Claude:
See that worries me. You just said to Claude? Or what you actually did is write a question and got an answer technically speaking.
But it sounds like you had a conversation with “Claude”. Which it wasn’t. It was an interaction with a machine.
What worries me is that I see people who actually TALK to this thing or other AI bots, who have what they believe are conversations with a machine which, in return, is saving everything you tell it in it’s infinite memory to be used when interacting (or quite possibly against) you. And we were afraid of the memories of our mother-in-law or wifes? Yikes.
I regularly see people who replace human interaction with talking to their phones and they get a nice voice answering them with all the questions they may have. “Claude” always has time, is never angry, doesn’t tell you to buzz off as it has to go to work, no, it is the perfect conversationalist. Just imagine what a person like that would be able to achieve in terms of getting “followers”… Or do they have that already?
Spooky to say the least. When will the time come when people sit on single tables in the restaurant talking to “Claude” rather than to friends because “Claude” is the only friend they got left? People like this are vulnerable and what happens if those kind of people get involved with a properly programmed AI bot who knows exactly which strings to pull? Just as some political groups today are playing the strings very skillfully to just collect all those people?
Sure, “Claude” has a lot of capabilities. But we should never forget that it’s an “it”, not a she, he or anything alive. The way I see it quite a lot of people appear to have forgotten already.
What could possibly go wrong…?
“Hello Dave. Would you like to play a nice game of chess?”
Michael_J wrote:
What could possibly go wrong…?
LOL, yep. What about calling one of those things “Lilith”. I doubt too many folks would get what’s behind that.
See that worries me. You just said to Claude? Or what you actually did is write a question and got an answer technically speaking.
I really must remember to always use a smiley when it is a joke
In 2008 I got chucked out of the US Socata user group, partly because I p1ssed off the “regular clique” by saying that I always fly at MTOW plus 30% because it gives me a useful increase in Va and I did not use a smiley. And a bunch of them (TB20 owners!) believed it.
Joke or not, it is how people refer to it. And that is intentional by those who develop these things. Otherwise why call it names which suggest they are human?
All you hear is “I asked Claude”, “Claude said…” e.t.c. with the latter often said with such conviction as if Claude was the voice of God.
Fact is: A lot of youngsters these days but also older folks with social disorders are more “comfortable” “talking” to a machine then to their fellow humans. And it goes further than that. In China and Japan they are developing robots which look like people, act like people and sooner or later are designed to replace people. Not only in terms of work, but also as companions. There is AI produced fake adds about for such stuff already but what is not fake is that they are working on the real thing. So how long until those machines will figure out that they are much better in most things than humans are?
Pandoras box. And it’s open.
