Don't Name the AI

It’s a general rule that you don’t name the animals that’ll later end up being dinner. This may not be a huge issue in suburbs but the homes keeping sheep or cows or chickens, you can see where we'd run into this. Why shouldn’t we name them? Because once it’s been named it’s a pet or adopted family member and no harm can come to it without many tears from the small children that have fallen in love with it. It’s the same with our children’s toys; once named the toy is forever referred to by its name. In our children’s mind, it’s a living thing that must be loved and protected!

It’s not just children who name the things that are important to them. Dean’s ‘Baby’ of a 57 Impala, every ship that’s ever been built are good examples of our affinity to give things names. Even us land-loving, non-fictional adults, who also understand the difference between living things and non-living things, give names to what’s important to them. I'm sure that even your Roomba has a cute nickname. And it’s not just the physical objects that we anthropomorphise!

During my PhD years I got attached to a piece of code. Ok, it was more than a single piece of code, it was a whole processing suite taking data from flat files to features to the structure and creation of an Artificial Neural Network to collating the results and presenting us with the answers we so needed. But I’d built, it was my baby, I named it ‘The Machine’!This was sadly not it's final name in my paper.

I’d written the code, built it up from nothing, and then I went and NAMED it! That was it, this code was a thing now, it existed, it was a life! This meant I had to take care of it. If someone suggested a change or that it wasn’t good enough, I’d go all Mama-bear. (Well maybe not that extreme, this was academia.)  But I’d created an emotional connection to a bunch of scripts. The scripts didn’t even know I existed! They had no concept of creation, ownership, communication, love, nothing! The scripts ran their instructions on my computer's CPUs, and they didn’t even know I cared.

Was this loving connection to code born out of growing up with science fiction? Where there were robots and androids that could think and feel for themselves, like Data from Star Trek or Sonny from I, Robot. Or was it because I’m human and we like to believe our creations are beings too? When I’d named it, I didn’t understand the emotional connection I’d just created. Or that I could have an emotional connection to something non-living and not physically existing.

This brings us around to Artificial Intelligence (AI). It’s non-living and when it’s not built into a robot it’s not really physical either. We can’t look it in the eye or shake it’s hand to introduce ourselves. There are different ways we can interact with AI-like applications; It could be like my PhD scripts where the interaction was me running a script to kick of the process; or it could be through a chatbot window or online interface where we ‘talk’ to it. In each way the thing that has been created out of lines of code gives us something back, just like the give and take of a conversation. Sometimes it’ll be a file and other times it could be words or numbers.  I was able to emotionally connect with something that outputs a bunch of files. So, you can imagine what happens when a human interacts with something that responds to us less like a computer program and more like another human being. Suddenly these machines are ALIVE!

But they’re not. Underneath their cool exteriors they are made up of some very fancy coding and a whole lot of data. And this is where we need be careful. We love to anthropomorphise and the closer to human something acts the more we are going to believe it is human, or at least human enough, and that it is capable of sentient thought. If we easily convince ourselves that a computer program is alive, we can then be easily convinced to trust it and what it gives us.

Our children trust their toys and teddies, especially those that have names and go on car rides with them. They talk to them, tell them secrets, give them cuddles, and trust implicitly when their Foxy says she LOVES broccoli and it’s the best thing EVER! This is a safe trust because it is us as parents or even the children themselves who are the ones doing the thinking. We are capable of amazing thought processes, taking in input from our emotions and memories. AI doesn't have this background ability but it's responses to us can make sense and seem like a person responding we can start to trust it like we would trust Foxy. But it’s just a façade.

These machines very clearly parrot back at us the information it has seen before, historical information. By trusting systems that are based on passed behaviours, or information that’s not in the best interest of people like some of human hisotry, we are propagating the same issues and biases into the now and the future. But we trust it because we’ve created an emotional connection; This emotional connection is because we humans like to anthropomorphise things. We named it, and therefore it’s ours to love and protect. But what we really need to protect are those that use our AI/Machine Learning Models or those who have it used on them.

This post has been a long time in coming. The popularity of Large Language Models recently has seen the world getting closer to building true AI, but we aren't even close yet. What the LLMs have done is they have made Machine Learning models avaliable to people who wouldn't normally have anything to do with them. Now that Machine Learning is a household term we need to start thinking about how these very cool systems are impacting our lives well before a true congnisent AI is built, if it ever is.

My next post is going to delve into the data behind machine learning algorithms; How it's used, where and how it goes wrong, and how we can make sure that our 'trusted AI' can trully be trusted to have everyone's best interests at it's electronic heart.