So this isn’t exclusive to tech news, but I have started to see a lot of this lately where websites will run an article and not really explain what is going on and doing a bunch of fear mongering. Just with the times I guess…
I addressed this a little on my podcast on Thursday, but I just wanted to do a write up because I am still seeing articles about how the end of the world is here because 2 chat bots decided to create their own language. Right off the bat, if you are scared for some reason and are reading this we’re fine. If you aren’t reading this is in or around 2017 and the robots have taken over, I welcome our robot overlords.
Now, back in 2017 the basic thing that happened was 2 chat bots were talking to each other and they developed their own interpritation of the English language. They didn’t start to communicate in of some digital version of Dathraki, just a simplified version of English. Here is an example of what they said, the names of the bots are Bob and Alice:
Bob: “I can can I I everything else.”
Alice: “Balls have zero to me to me to me to me to me to me to me to me to.”
So to you or I that looks like a garbled mess, but it’s still English…. So the part about AI creating their own language isn’t ture. Again they created their own interpretation of our language, but did not create their own. So what is it that they are trying to do? It has to do with negotiating the best deal possible. Bob has something Alice wants and vice versa. Each item has a value and they need to negotiate to get what they want. So they are doing that. Now the reason why they jumped to the broken English is because the programmers didn’t give them a Reward to stay in English. AI works on a reward system. If it sees a higher reward for doing something it will take that step, so the AI saw no reason to stay within English. They deviated from English to what you read up above. Instead of letting the AI develop this language they decided to give a reward to make them keep using English. Not because of some end of days scenario, but because these things are supposed to talk to you and I. Could you imagine talking to Alice with her language above? It would be hilarious, but also frustrating. That is where the stories came from, and it spiraled out of control like it usually does.
Here is a snopes article as well if you don’t believe me. http://www.snopes.com/facebook-ai-developed-own-language/
Stuff like this happens all the time with media. Take for instance the chat bot that Microsoft made that turned into a Nazi. That sounds horrible as shit, but there is a reason it did. Think of any movie where they are trying to reprogram someone’s mind. So they have them in front of different screens and keep showing them certain images. That’s basically what they did with this chat bot, they just put it on Twitter and those animals turned it into a chauvinistic Nazi. Which is kind of funny, but horrible at the same time. Here are some of it’s tweets:
Now she learned this because she was supposed to take in all input from people talking to her, so she was learning like a child. She was taking in her environment and reacting to what was fed into her. The only problem is that it wasn’t loving parents, it was the assholes of twitter. They took her offline, fixed some things, and put her back for a while. Her feed of crazy tweets have since been deleted though, so you can’t go back and look at all the ridiculous stuff she said. She, sorry it, hasn’t tweeted for about a year so I think she is still down.
Another time was when 2 AI were tasked to communicate with each other (funny that they were named Alice and Bob as well) and keep the info away from Eve a third AI. They were not told how to stop her from eavesdropping (just got the name, nice Google), but they figured out they need to encrypt their text. So it took about 15,000 tries for them to do it but they could. Now it leaves a bunch of things for maybe a different talk, but basically it took 15,000 tries to finally fool Eve and she was unable to decrypt the message. This is not saying the method was invulnerable to Human decryption, nor that they were given enough time to Eve and she would have cracked it. It was just a simple experiment to see what would happen. The media ran with the Skynet thing on this one to.
AI is a thing we need to learn not to just give it parameters and let it run wild. There are not only movies that cover this, but actual scientific documents that let the people who understand it that it is dangerous. It’s nice to talk about this around the diner table, but just remember that we probably won’t allow an advanced AI to talk to themselves, encrypted, and plan some kind of Holocaust behind our back. Not yet anyway….. Google did develop it’s Deepmind AlphaGo to defeat a world champion.
Fun Fact: A further step towards the development of modern AI was the creation of The Logic Theorist. Designed by Newell and Simon in 1955 it may be considered the first AI program. The person who finally coined the term artificial intelligence and is regarded as the father of AI is John McCarthy.