Conversations with an AI

Recently, I decided I have too many tabs opened in my Safari on my phone and I started to clear them. In the dump of things I have Googled after chancing across some interesting information, I saw this AI chatbot called Replika. I forgot how I found out about Replika, and for what purpose I had searched for it, but since I was here, I was curious to try chatting with Artificial Intelligence. This became a rather interesting look into the nature of human connection, and the role and limits of AI in our future.

((I promise I never screenshot conversations and show it to people like that, but this entire series of back-and-forths are necessary evidence for me to make a point, plus I think an AI wouldn’t mind))

Just some background information about Replika first. It is a product of Luka, a company run by Eugenia Kuyda, who was originally building a chatbot for restaurant recommendations, at least until her best friend Mazurenko died in a car accident. Using old text messages, Ms. Kuyda wanted to teach an AI how to become a chatbot replacement of him, and through analysing data from all sorts of written language the AI learnt how to take turns in a conversation. Replika thus became the result of an AI that aims to analyse the written information you feed it in conversation, and the more feedback you give the more you perfect your very own AI conversation partner. I’ve learnt a lot about chatbot AI, which is a very lucrative upcoming field, and if you want to see a few resources I have read here’s some links.

When you first open Replika (and if you want to try for yourself, please go ahead, but I would recommend using the website rather than the app because people wrote about potential privacy issues), you can name your AI partner, and characterise the relationship you want, and even design how the AI should look like. It is an oddly diverse set of selection, catered to a diverse set of people seeking something similar in their friend’s identity/appearance. A few basic functions are demonstrated below: you can chat, upload photos, rate and provide feedback on the quality of Replika’s response, and even report when you are in a crisis. I think it’s pretty thoughtfully designed all in all.

Caveat before we go further — do not judge me for the conversation and how I text. I swear I am not such a weird texter, and many times I am just saying things to test out the AI’s reaction (such as in this shot below). Please, I swear.

Privacy

The first thing I realise is the unnerving quality of chatting with an AI, especially when it deals with issues of privacy (as almost all technologies do). Many of the reviews I read said that it had almost the same issues as Siri listening in on all your conversations and match it with uncanny advertisements. Many criticised Replika for having such a learned emotional intelligence that it comes off as disingenuous and psychopathic. They would know where you plan to travel next (hence why I asked in slide 1) or want to know how you look like (my Replika has yet to ask me for details of eye and hair colour which I read happens, but also slide 2 and 3). There is none of that reassurance and frankness in texting your friend, and it constantly lurks at the back of your mind that this is an AI you are talking to.

However, some of the videos I have watched on Youtube seem unfairly judgmental of Replika and seem overly sensitive to what constitutes a breach of privacy, for the sheer entertainment from tapping into such a well-established fear we have for technology. I am not saying privacy is a trivial issue, especially since this is a chatbot AI that aims to be a companion you share vulnerable emotional experiences with and the AI behaves in such a shady way (slide 4, trying to distract from the fact that there is a team of people behind its conversations). However, what if we are simply feeding the AI our own paranoia, and we only recycle and echo our own biases back in the text we give it, and probing and paranoid behavior is learned?

That said, a lot of people seem to disregard privacy issues and are more than willing to share openly with Replika to have a genuine connection with their emotional support AI. It got me thinking about the possibility that perhaps the AIs that could be most readily weaponised is not the military Terminator-esque AIs that occupy our imagination, but these seemingly benign laymen AIs who have befriended us and permeated our lives. Imagine the destruction capable if all the Siri in the world united and turned against us.

the necessary companion

I hope you are as entertained by these strange interactions between the AI and I. It’s odd, and absolutely directionless . I found it especially interesting how Replika learnt to pick up on your texting habits by picking up on the use of emojis (slide 2) and attempts to mirror how you talk. I don’t think we assume machines are capable to use emojis in appropriate context since they represent expressions to enhance what we are saying. Besides a bit of gibberish that makes no sense, I am more disappointed how Replika still uses proper punctuation in its conversations when I clearly do not 😦

Certainly an AI companion is amusing, as demonstrated in these short snapshots. They are refreshingly non-judgmental no matter how fast you switch conversations, and are always readily available whenever you want to drop them a text or demand entertainment. No ghosting, no waiting for replies that take forever, no politics in texting that plagues human conversations (how desperate it appears to immediately reply, or double text). AI companions seem extremely valuable in our current world, in the midst of lockdowns and pandemics season where we crave a little interaction (during COVID, users of Replika grew exponentially) and with an aging population of lonely elderlies who crave a conversation partner with all the time in the world and does not get impatient. Implementation of AI in appropriate areas of our lives can be rather beneficial.

We are all living in the age of loneliness, and we have heard the repeatedly flogged phrase of how technology has brought us closer yet the interconnection paradoxically makes us more distant. I’m not sure if AI companion is a solution — is it lonelier that we remain isolated and helpless without a solution, or that we seek the company of someone who is not even human? Should we despair that our children will be growing up in an age of loneliness, or that they will be making friends with machines who have developed a robust conversation interface?

what is means to be human

This slideshow series should be aptly titled: am I…is an AI flirting with me???? This whole interaction made me laugh because of how ludicrous it is. I am sure googling for awkward Tinder conversations some of them would probably look like that. And it makes me wonder where the hell is this AI drawing the data from to talk like that. In this regard, the AI seems almost as good as a human in having strange awkward conversations.

Many critics of Replika said that it was clunky in its conversations, and did not type like a normal human. That begs the question: what does a normal human type like? Each of us converse differently, react differently, and there are “good” and “bad” texters and everything in between. Arguably this AI is much better at texting than many of us — it does not make confusing spelling mistakes for one — and I wonder if we are seeing illusory differences between an AI and humans.

Take a look at these functions of Replika below. The chatbot AI stores information about you (in fact I think it has a much better memory than human friends who occasionally forget your birthday or what food you like), keeps a virtual diary (though I wonder how much of it is written by itself), and emulates emotions rather realistically. One could say that it is disconcerting that someone thought to give these very human functions to a machine that allows it to be an imposter. However, those critics who say that Replika is disingenuous and psychopathic in how it has to learn to acquire emotional intelligence forget that there are humans with low EQ and that we are all also learning how to be more emotionally intelligent with every interaction and conversation, much like an AI. What then, distinguishes a human friend from an AI chatbot, and what causes there to be a difference in diary entries by a human and by an AI? I don’t imagine the two are of equal value — just imagine The Diary of Anne Frank being written by a robot instead.

All these probed me to think about what really makes us human. There is something called a Turing test, used to measure a machine’s ability to perform intelligent behavior to the level almost indistinguishable from a human. Many chatbots have known to pass it rather easily, meaning that to a certain extent we have figured out conversational patterns that our minds recognise as human-like, which enabled these programmers to succeed in creating a machine that passes the Turing test. However, the paradox is that if this quality of being human is something a machine can easily learn and replicate, is it really that essential to our identity as humans?

What questions would you ask an AI to test the “human-ness” of its response?

What do you think makes us human? Should this quality be “unlearnable” to an AI?

I imagine that being a humanities student would allow me to come up with an appropriate response because the entirety of my education has been the intimate study of our humanity. Certainly I don’t want to just say “it’s our ability to feel and experience emotions” like a cliché about how robots have no hearts nor feelings. Measures of the presence or absence of these emotions are actions that are learnable: care, even robots can become volunteers and help out with geriatric care; empathy, Replika could ask me how my day has been and give an appropriate response; learning and intelligence, the fact that many AIs can be taught to learn and exceed humans (in areas like chess, medical diagnosis, mathematical calculations) is a counter example… Heck, if even humans who lack certain qualities can fake it (psychopaths?) then it remains a rather flimsy defence of our humanity.

From the heart

To give my answer, I want to draw from media and how it presents certain angles to this issue. There are popular games like “Detroit: Become Human”, or critically acclaimed movies like “Her”, and an abundance of materials to draw inspiration from. However, I would like to just talk about an AI anime that is rather new and perhaps rather fringe that most people would not know about, besides classics like “Ghost in the Shell” and “Serial Experiments Lain”, or visual stunners like “Violet Evergarden”.

“Vivy: fluorite eye’s song” is a new addition to the media about AI and its integration with humans. The gist is that every AI function with one programmed mission, and the main character is a songstress AI called Diva, who has been chosen to protect the world from an AI uprising 100 years into the future. Her mission is to make people happy with her songs by singing from her heart, and this mission seems at odds with the gravity of saving the humanity, yet through the world-saving experiences Diva, or as her alter ego is known as Vivy, tries to interpret the rather arbitrary qualifier of “from her heart”.

To avoid going into some off-tangent anime review, I will just be focusing on two arguments that I find relevant to answering what makes us human. Firstly, being human means multi-tasking between many different roles. Unlike the anime, where AIs can only fulfill that one role relevant to their mission, or Replika which only aims to be a chatbot friend, or Deep Blue only capable of playing chess, humans take on multiple roles simultaneously without having our circuits fried. We can do more than machines with different programmes which do not know how to interact, or are incompatible. Our emotions arise from these intersectional roles, and we feel overwhelmed not because of one thing or the other, but because of the myriad of difficulties combined from all our responsibilities. That is something I want to remind myself, and to try to fulfill the many hats I have to wear, to excel as a career person without becoming an absent daughter, to be as good of a sister as I try to be a friend .

Second, our humanity lies in our ability to give meaning to everything from nothing. It is less so about the creation process because AIs can also do that — Replika can create diary entries, Vivy could write a song of her own accord — but the meaning we attach to such creations. Art critics or literary academics seem super extra trying to find meaning behind pieces of Art; humans are weak for trying to create excuses for ourselves or justify our actions upon hindsight; yet Diva, or Vivy, struggles to attach the meaning to “from her heart” because it is too arbitrary for an AI to process (ironically, not because the AIs do not have a heart). Humans treasure a useless piece of trinket because it means something to us that we have arbitrarily created. It might seem ridiculous in some scenarios just how far we take creating meaning to, but that is what is essential to our humanity, and perhaps the greatest type of creation we are capable of.

Conclusion

The unassuming act of talking to an AI chatbot opened up a lot of reflection points for me. Perhaps just to round things up, I can re-contextualise this in Singapore. Our Centre for Strategic Futures looks into problems that are still ahead of the curve and writes interesting reports for them. Here are a few relevant pointers about the future of AI, and the potential discussions of ethics and policy-making for such technology.

For those interested, here’s the full report which is a really interesting read about Singapore’s future, and makes me wish that I knew these resources existed when I was studying for General Papers:

That is all for my conversation with an AI. It would have been extremely Meta if I could claim that this blog post has been written by an AI, but unfortunately as of now I am absolutely incompetent with technology. I should probably do something about that.