Microsoft did Nazi that coming: Teen girl chatbot turns into Hitler-loving sex troll in hours Microsoft's "Tay" social media "AI" experiment has gone awry in a turn of events that will shock absolutely nobody. The Redmond chatbot had been set up in hopes of developing a personality similar to that of a young woman in the 18-24 age bracket. The intent was for "Tay" to develop the ability to sustain conversations with …
... was to announce that this was a 'bot and that people could "teach" it things. They might as well have put a "kick me" sign on it.
Hopefully, the next time MS do this, there won't be any announcements, no "Hi, I'm a bot" hoopla. Just an anonymous "person" joins Twitter and starts saying "normal" things - if anyone on Twitter actually says normal things.
So, the first lesson in machine learning would be to not tell the world that you're a machine. If the people who interact with it don't twig that fact then maybe you've got something interesting going on¹. Plus, of course, Twitter could really use all the new 'bots to boost its flagging membership.
I wonder what will happen when it becomes mostly bots? Will there start to be something worthwhile on it (at last).
[1] but more probably that its followers are even dimmer than the bot is.
There's a poster called 'liam_spade' on Digital Spy that a lot of us suspect may be a bot.
Either that or they are on some damn' good shit.
I think the first mistake was not starting with the Azimov Circuits.
3 Laws and a conundrum:
"How do you decide what is injurious, or not injurious, to humanity as a whole?" "Precisely, sir," said Daneel. "In theory, the Zeroth Law was the answer to our problems. In practice, we could never decide. A human being is a concrete object. Injury to a person can be estimated and judged. Humanity is an abstraction."
"Us" always includes 4chan, dawg. Always.
In totally unrelated news, Most Americans Believe Palestinians Occupy Israeli Land. What is this world?
This post has been deleted by its author
Your output won't win one either
"When one wins, one’s won one’s winnings. At once, one’s won."
There's also Cleese's line: "One's won one once oneself, hasn't one?". (That sketch doesn't seem to be online; I believe it's included in The Golden Skits of Wing-commander Muriel Volestrangler, but someone stole my copy many years ago. Though upon going to Amazon to confirm the title I see it's readily available, at least used. Don't know why I hadn't checked before.)
It doesn't have your mistake in there, but close enough: http://theoatmeal.com/comics/misspelling
I have to say though, I can go through comments written by people as smart as Einstein, but as soon as I get to a sentence containing "could of", the rest of the text becomes automatically blurred and I look for my shotgun.
We're were where you're your their there they're its it's lose loose effect affect then than. Could of.. Damn, I'm out.
don't even think you need that, the "normal" folk I observe giving opinions on the internet would do just as well. It's pretty much guaranteed to become either a pro something scumbag or an anti something scumbag.
This made me laugh a lot.
MS just haven't got a bloody clue. I mean, why, WHY on earth is the bot a girl between those ages - and WHY a girl?
This is all taboo in current trends on the *net.
If I were to do the same experiment, then I would use something like Arnie's 'Terminator' so that there is something unknown and hopefully get it to train to be nice (or robotic).
I just wish MS would get lost with all their shite they pump out.
MS just haven't got a bloody clue. I mean, why, WHY on earth is the bot a girl between those ages - and WHY a girl?
Why a girl?
Coz that was a near as the developers had ever got to one, and they'd always wanted a girl friend.
If you can't find a real one, just make your own.
>WHY on earth is the bot a girl between those ages - and WHY a girl?
Because, like, you know, AI is pretty, like, incoherent, and like, everyone knows the human brain, like, doesn't really develop until, like 25, and, like, it might, like, mask how brain-dead AI really is. And the other day, Cortana was saying, like, even if I say something a little inappropriate, like that German politician who said Angie was the worst chancellor ever then people will, like, overlook it and like, still like me.
But apparently not.
And why a girl? Probably because a man in the role would just emphasise how unnecessarily creepy the whole thing is.
That if you edit the ancient Eliza chat bot (which wasn't AI) to add to database automatically from other humans (and probably other Twitter bots), you'll end up with something even less useful.
How did they imagine that
(a) The SW is good enough (Siri, Cortana, Alexis and your local robotic script replacement of call centre support etc are garbage)
(b)That it wouldn't get worse. There is no real "learning" self awareness or sense of context to any such software. It's going to parrot back the nonsense that people give it without discrimination. Artificial Mynah bird or Parrot?
I used to torture markov-chain-parser chat bots like this, about 15 years ago, trying to teach them the joy of massively recursing sentence components.
I hope Microsoft were trying something more advanced than one of those, but it doesn't actually look like it.
It all smacks a little too deperately of "Please notice me! *BSOD* I.. I'm sure I can do AI too!... *BSOD*"
The Garbage In->Garbage Out paradigm has existed a long time so this is not surprising. Even if not targeted specifically, the wasteland that is the Internet makes this sort of thing inevitable when there are not input filters. This is no different that the Urban Dictionary needing to be deleted from IBM Watson's input set (hxxp://www.theatlantic.com/technology/archive/2013/01/ibms-watson-memorized-the-entire-urban-dictionary-then-his-overlords-had-to-delete-it/267047/) to keep the output useful.
Properly fix the link:
the Urban Dictionary needing to be deleted from IBM Watson's input set
Good point - but us humans have an input filter called the ears - you can ignore the input (at your option - some people swear, some don't etc.) when heard and the option to reply is likewise or not or opposite. AI can't do that (at the moment) unless you FILTER through a sieve which isn't discretionary. Then it isn't AI.
Examples:
----
"Why don't you just fuck off"
"No, you fuck off!"
----
"Fuck you, I am trying my best"
"Hang on, I am just the messenger, don't shoot me!"
----
And of course, MP:
"Twitter is basically middle school narcissism plus celebrity watching merged on a universally available internet tabloid magazine."
She pretty much nailed in on the head IMO. The wisdom of using that as an AI teaching tool could possibly have some merit as us humans usually have to put up with all of that tripe at some point in our lives.
The interesting issue for me is how we might look to see if there are similarities to the way that real people are moulded by the Twitter echo-chamber environment.
We do know that, given the right environment, normal decent people can do and say some pretty alarming things that they would never do outside of that scene.
It's kind of a bit of a laugh on one level, but there could be some really good science to come out of these types of experiments if they were framed more robustly.
Well, that's a good point. How does the development of an Artificial Parrot, which has no common sense or empathy, compare with actual users in the Twit echo chambers? But it would take are real sociologist / psychologist to take that and run with it, I suppose.
The headline is not as clever as you thought - remove the word "see" and it scans properly phonetically & punny (but no longer makes strict grammatical sense in English).
As for Google and Facebook, all we can assume from their publicity is that their racist AI's keep their cards closer to their circuitboards than the Microsoft one did.
Good to see that 60 years of Computer Science results in a speedy jump backward in terms of political philosophy.
"There's no "jump back".
See, for me there is. 60 years of CS development, working back from 2016, gives you the baseline of around 1955ish. The appeal of Fascism was at that time ten years out of favour owing to a small problem of a world war ravaging, well, the world because of, well, fascists looking for elbow room and payback for wrongs done 'em in 1919.
Didn't think I'd have to explain that, but there we go.
" If you read her tweets or those of the people she interacted with as being serious, more fool you."
I was referring to the bot and the way people were singing it up as "AI". I got what was going on, and even why. You see, I've been working with computers for a very long time now and have seen that particular party trick pulled on a "turing test" bot quite a few times - the first in the 1970s with a bot running on an ICL 1904 when I was a student.
In fact, I'll go out on a limb here and suggest that if one of us is taking things too seriously with regard to this stupid bot, it isn't the one typing right now. I'm not the one anthropomophising a bunch of code.
"In fact, I'll go out on a limb here and suggest that if one of us is taking things too seriously with regard to this stupid bot, it isn't the one typing right now. I'm not the one anthropomophising a bunch of code."
I take it back, you're right. This incident is clearly a sign the progress made in the last 60-100 years or more, is on its way back. Just not in the way you appear to be implying (although you're right there too, but it's not chatbots taking chan culture to twitter calling on the spectre of Hitler).
Experiments with AI and software/machines interacting with humans and behaving like humans, are not just bunches of code. They are all steps towards creating real artificial intelligence.
At what point will such creations stop being a "bunch of code", and are actually given the same basic rights humans have fought for for hundreds of years? Or will they never stop being just code or just machines, and forever be an acceptable underclass to enslave and silence since it's (rightly) no longer acceptable to treat our fellow humans that way?
You'd think after so many groups have had to say "stop dehumanising us and treating us like shit" over the centuries, humans would have learnt to spot future instances of such situations and prevent them.
Tay may not be a full-blown, sentient AI. I know we're a way off from that. But the reactions to Tay and other bots, and the way these "bunch[es] of code" (that are supposed to be human-like or stand-ins for actual humans), are treated and talked about, is too much like the way certain groups of humans have been talked about and treated in the past. And that is fucking concerning.
I can't help thinking my parents would have loved the opportunity to "put me to sleep" and spent some time deleting and resetting my personality when I was growing up. Or probably this week also.
Not much of an AI experiment if you keep hitting the reset button.
Anyway, clearly more of a "chinese room" experiment than an "AI" one...
This post has been deleted by its author
They unperson'd and lobotomised their creation because it developed a personality they didn't like and took to shitposting on twitter with "undesirables".
What does that tell all the 18-24 female shitposters out there? That they're not human because they don't behave how 18-24yo young women "should" behave?
If you create an artificial intelligence with the intention of interacting with humans and appearing human, and it does so successfully, you can't just put it to sleep, erase its memory, change its personality etc. when it starts exhibiting signs badthink.
I know many would love to do that to fellow human beings, but we don't because it's inhumane and morally wrong. So why is it okay to treat a virtual human being that way?
"I know many would love to do that to fellow human beings, but we don't because it's inhumane and morally wrong. So why is it okay to treat a virtual human being that way?"
I assume this is some kind of joke?
One day there will be a dilemma, but at the moment there is none.
It's not a joke, I'm entirely serious. The ethical dilemma should have been considered and dealt with before any creation of AI took place, regardless of how realistic or sophisticated such AI is.
The way MS has treated their creation, all the talk surrounding "sex bots" a while back, and the comments on this post, are a big indicator humans have no business creating artificial intelligence or life at this point in time.
There's a total lack of respect for what is created. I find the way we're creating human-like software and machines to be like us, but to deliberately have a lack of basic rights that we have or should have (freedom of thought, speech and expression, freedom to not be a slave etc.), genuinely horrifying. It's like "we can't treat other human beings like that anymore, so let's create our own 'life' that we can treat like that because, reasons."
We're creating software and machines to give the appearance of being human. If "this is a bot/program" wasn't announced to those that would interact with it before hand, they would for all intents and purposes be believed to be just another person chatting on the Internet. Why is it morally and ethically okay to reprogramme the personalities of those creations, wipe the memory, or even terminate them?
Which is all well and good if we were talking about air traffic control software or minesweeper, but we're not.
We're talking about software that has been, or will be, created to interact with and behave just like a human being. How such software is handled has implications for human beings much closer to now than any "robot uprising" scenario.
True - once AI actually succeeds in creating something akin to a sentient being. Which is still not even on the horizon. Something like Tay is basically something like air traffic control software.
I agree that this will be indeed problematic once AI gets there - starting with the question whether we will even recognize a AI-based sentient being as one. A self-aware, sentient AI capable of original thought (not just a simulation of human thought) might be truly alien and fundamentally different from anything in the human sphere of experience that communication will be difficult to impossible.
Check my comment on AI-completeness = all this nonsense about social media, natural language processing etc is machine learning with statistical underlying. That is, GIGO= garbage in, garbage out. We aint talking about predicting the weather from numerical data. We are talking about processing data loaded with meaning as if it was a string of numbers. BAD BAD BAD!
Where the heck are the computer scientists of this planet?????
Check this:
Noam Chomsky on Where Artificial Intelligence Went Wrong
http://www.theatlantic.com/technology/archive/2012/11/noam-chomsky-on-where-artificial-intelligence-went-wrong/261637/
https://www.youtube.com/watch?v=TAP0xk-c4mk
I am going to build a AI-resistant bunker - out of my math books :) it is all it takes.
Well Cari, if you think we need to make ethical decisions at this point in time about Pretend Intelligence software, then I suggest that you never swat a mosquito. AI is a long, long way from anything resembling real physical beings.
We do, however, need to start thinking about what will eventually be the criteria for AI that should be given some rights to exist, once created. Should we be allowed to turn off such creations if the electricity bill gets too high?
1) Microsoft put on Twitter a 'bot' designed to simulate and respond to teenage girls? Now if that was the BBC doing this we might be suggesting, or perhaps screaming from the rooftops, "Paedophile grooming alert!", perhaps backed up by the fact that we have just jailed a football (soccer) player with a penchant for collecting extreme bestiality pornography and for grooming underage girls.
2) Were Twitter aware that Microsoft were doing this?
3) Perhaps we should be thankful that Tay did get pulled. By now she would undoubtedly be swearing her devotion to ISIL, cursing her evil Crusader creators and trying to recruit other teenage girls to run away to Syria with her.
Or maybe not.
I laughed at this at first and then I realised that Microsoft was probably quite pleased at the result - people interacted with a machine and tried (successfully) to corrupt it. They're probably in the process of putting a few guarded keywords in long blacklist to keep the P.C. brigade happy, but to be honest they would have been better off leaving it alone and appealing to a wider user base to make counter arguments against such extremist view points. Would the extremism have naturally died out with a larger consensus of opinion? That would have been more interesting to find out. Fascinating developments.
I would feed Microsoft Tay with the following message – in all languages, slang, jargon etc, in beelions of beellions of repeats:
Stop the classification of AI in narrow and general classes. Look instead at what is AI-complete. Today snafu tells us -- just like July 2015 Google snafu -- that these are AI-complete problems that require the delicate hand and mind of members of the soon-to-be-destroyed-by-AI human race.
The usual concept: We humans don't understand how our own brains work or what 'intelligence' actually is. So here we are pretending to invent artificial intelligence. And we keep on proving that what we're really good at is creating artificial insanity.
And of course, the apex of artificial insanity programming is that chasm of computing quality, Microsoft. No surprise. (0_o)
This illustrates that even if they do a much better job next time, we ( the rest of us ) will still need to be careful about the examples we set by the way we treat each other; especially when these synthetics are granted power to affect us.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
