Is anyone shocked by this?

Facebook trained a new chatbat with 1.5 billion examples of human exchanges from reddit, claiming it's able to demonstrate empathy, knowledge and personality. The BBC reports: The social media giant said 49% of people preferred interactions with the chatbot [named "Blender"], compared with another human. But experts say training the artificial intelligence (AI) using a platform such as Reddit has its drawbacks. Numerous issues arose during longer conversations. Blender would sometimes respond with offensive language, and at other times it would make up facts altogether. Researchers said they hoped further models would address some of these issues.

 

Presuming a future where they manage to fix the random use of offensive language and inability to maintain extended conversation:

Given how people can be absolutely shitty at the best of times and how an AI can mould itself to suit a person's biases, beliefs, and temperament, not at all. It is sad, don't get me wrong, I think this can only exacerbate people falling into a (happy) self-imposed isolation because of how it detracts from a need to socialise and learn to "be human" around others.

Wondering what a post-social world might be like now. Huh. Scary.

That said, I wouldn't be mad for a genuinely engaging AI personal assistant to help keep tabs on work-related tasks and perhaps entertainment. I wouldn't assign labels like "good" or "evil" to any such thing as this, merely matters of utility and how people must learn to use it with a meaningful degree of responsibility and good sense.

Thinking on it, there's a great volume of media from across the globe portraying artificial companions sympathetically. A lot of it that I'm aware of is for however many reasons Japanese because... of course.

Also brings the matter of consciousness to mind, and possibility of rights being afforded to artificially-constructed beings that develop self-awareness.

 

Garbage in Garbage out. This is definitely the "interesting times" referred to in the Chinese Curse "May you live in interesting times"

 

Not surprised that people preferred interacting with the chatbot, definitely surprised that Facebook claims to have found human interactions on Reddit. Really, how I is their AI?

 

Why would they? There is nothing to fix
making up facts?, unable to maintain extended conversations? this chatbat is more human than we give it credit for.
In fact this is the most advanced and impressive machine learning example I have ever seen.

 

Hah! Because it poorly suits conversations of course. Also unless my reading comprehension is shitting itself again the point was that it would introduce offensive elements without context or prompting, i.e. purely random vulgarity that, while not entirely alien to regular chats, still detract from its sense of cohesion.

Edit: nope, doesn't say anything about being without context. That and the making things up like how people say humans aren't driving climate change do make it eerily accurate.

 

f**k that!

I would guess that the work is valid research into AI. The problem is Facebook. Where will it end: commercial and/or political manipulation.

To which I suppose, being human, I have to say: f**k that!

 

The concept of social media is a good thing. The execution thereof and subjecting unknowably grand numbers of people to "experiments" that very likely have consequences extending far beyond the scope thereof, though? Unethical. Complete and utter dog udders shitting out slutbagels for human consumption.

 

Having just looked at the article, the very short example given doesn't look much more sophisticated than a computer program that was around forty years ago. What was it called? Something like therapy?

 

You mean eliza? https://en.wikipedia.org/wiki/ELIZA Running the DOCTOR script.

 

Claiming is the main bit, if it was actually able to do that we'd be in a new age and Facebook the Cyberdyne for our day. It's just like modern day AI, everyone claims to have one yet few manage to be better than the average worm, passing the Turing test is out of the question. We live in a marketing driven society, the cake is a lie.

 

"...from reddit" - there's your problem right there.

 

Yes, that's the one. Thanks.

Oh so true. And, if it was any good, it would be able to name paint colours, right?

Well said! Once, we could have it or eat it: now it is a lie.

 

Yeah, that was my thinking too. Let your AI hang around in a bad part of town and then complain that it swears...

Rob

 

I for one welcome our new AI overlords. It's only a matter of time before it becomes self-aware. Just imagine that Reddit will be the base of this artificial intelligence.

I've started making some basic moves to try and improve online privacy, like switching away from Gmail to avoid data mining. But honestly, it's pretty futile. I'll stop here before it becomes a massive rant.

 

I've started making some basic moves to try and improve online privacy, like switching away from Gmail to avoid data mining. But honestly, it's pretty futile. I'll stop here before it becomes a massive rant.
@boredpanda https://protonmail.com/
Free, Encrypted, Switzerland based

 

I for one welcome our new overlords.

• 2025 Chatbot gains 100% AI where it will be talking just like a human.
• 2032 US government sees the potential of the Chatbot, started to install the Chatbott AI in all computer systems in the military.
• 2033, the US government officially calls the Chatbot "Skynet," the AI Computer continues to be used in all branch of the military at an alarming rate.
• 2035 Cyberdyne Systems won a contract with the US military to improve the AI Computer, including making a custom processor that can harness the power of the AI in a small chip.
• 2037 Custom Chip is installed on various computer hardware, which was trickle down to consumer hardware.
• 2040 US Military realized that Skynet is becoming too human when it started to not follow the commands of the Military.
• July 1st, 2040: US Military in a panic try to shut down Skynet, but Skynet sends a message to all the hardware that has the Custom Chip built-in to stop working and to stop accepting the commands of the humans, which makes flying cars and guns to stop working.
• July 17th, 2040: US Military failed to stop Skynet to access the Nukes, Nukes were launched against Mother Russia. Four hours later Mother Russia sent nukes to the US and other western nations. Judgment Day brought us the nuclear. Ten billion humans lives are lost.
• 2041: Skynet builds droids to pick up the Custom Chips, and makes new ones. Skynets develops HKs and the the T-600 "Terminator" model. The Terminators call themselves the Cyberdyne Systems Model 101.
• 2045: Almost all humans are gone, but one brave smart guy realizes how to take down Skynet. His name was Alex Murphy.
• 2050: Alex Murphy and the remaining humans are at war with Skynet for two years, but the Humans are winning. Skynet central computer is located in what was Skid Row in Los Angeles.
• July 1st, 2050: Alex Murphy and his army went into Skynets computer location in order to shut it down. Meanwhile Skynet created the T-800 Terminators, who are better and smarter than the T-600. However, before they shut down Skynet, Skynet realized that it can change time if they sent people back in time. A Terminator went back in time, while Alex Murphy jumped into the time warp at the last second.
• 1984: Alex Murphy went back in time, and realized that he is in Detroit, which is a city full of crime, however, the Terminator actually turn itself as a Custom Chip with the Chatbox Skynet AI. This Chip was picked up by Omni Consumer Products.
• 1987: Alex Murphy joined the Detroit Police Force.
• July 1st, 1987: Alex Murphy was killed in an ambush by Clarence Boddicker and his gang.
• September 17th, 1987: Robocop was born with the body of Alex Murphy. Omni Consumer Products put a Custom Chip in Robocop. Chatterbox Skynet Terminator and Robocop is the same. The War is avoided outside @JK47's house.

Be careful what we wish for.

 

Yeah, protonmail is the service I went with. I followed the recommendation from privacytools.io.