Discovering the AI in your life
By Berry Billingsley with additional creative input and expert knowledge by Ted Selker
Dr Berry Billingsley is Professor in Science Education at Canterbury Christ Church University. Dr Ted Selker is a computer scientist and a designer of new ways to interact with machines.
For decades we’ve been told to expect artificially intelligent machines and for decades we’ve prepared ourselves and each other with stories about what they will be like. But Asimov’s stories about robots didn’t prepare us for GenAI chatbots like Bing Chat, Bard and ChatGPT. And I think I am probably more affected than most.
My interest is not only personal but also professional. As a teacher and university academic I run workshops for students of all ages to encourage conversations about Big Questions like – ‘what does it mean to be human?’ and ‘Can a robot be a friend?’ Designing these educational experiences means I have enjoyed conversations with engineers, theologians, politicians, lawyers, biologists, psychologists and many more about whether a robot can one day warrant being called a person.
But come the event of their arrival – I was unprepared for the reality of GenAI chatbots like Bing Chat and Bard.
Ok, so before I say why – I need to say a bit more about what I expected or really, why I didn’t expect what happened. My work as a university professor is about advancing how schools and universities manage the boundaries and connections between science, theology and other disciplines. I use case studies to bring my messages to life and my favourite case studies centre on the question: “can a robot be a person?” I have worked with this question hundreds of times, asking it in different ways: “Can a robot fall in love?” “What does it mean to be alive?” “Can a robot have a sense of curiosity?” “Can a robot think for itself” and “Can a robot care?” And our conclusion in these workshops?
The point of the workshops is to draw attention to the value of exploring a Big Question through a range of distinct disciplinary lenses because of the insights we gain. Thus – although we may not be able to solve the ultimate dilemma of what makes a person a person in a way that will satisfy everyone, we can give ourselves important and useful lessons through the limiting frameworks and boundary conditions that specialist disciplines offer. Scholars and students can also increase their understanding of how disciplines work and what makes them distinctive. This ‘knowledge about knowledge’ is something we call ‘epistemic insight’ and it can be usefully developed and applied in many cross-disciplinary debates.
The case study of whether a robot can be a person is one of our most popular examples with schools. For young students we compare how a theologian and a biologist might talk about ‘what it means to be alive’. For older students we introduce a similar idea by starting with a robot that has a sensor to detect sound. The robot starts or stops moving each time a student claps or makes a sound. and then asking what else it would need to have or demonstrate in order for us to say whether the robot can also be said to ‘hear’. We conclude that the question, ‘Can a robot be a person?’ is too big for science and engineering alone (Billingsley & Nassaji, 2019). What do some other disciplines have to say? The insights we get from one group of authors is that we are psychologically likely to fall into supposing this is a person until and unless evidence persuades us not to. Further, an approach of treating another as a person if we are sure is the theologically nobler and historically wiser way to go when we are unsure (Turner, 2023).
OK so the point here is that I sat at my pc ready to try out Bing Chat and Bard expecting “erm, well, not sure what”. Probably like many other people I started with some ‘getting to know you’ questions: “Hi there. Do you have a favourite cake and why?” My question was investigative and I asked ‘why?’ to see what the bot would tell me about how it arrived at its answer. Its response was a lot more humanlike than I was expecting:
“Hi, this is Bing. I’m glad you asked me that question. 😊 I like many flavours of cake, but if I had to choose one, I would say chocolate. Why? Because chocolate is delicious, rich, and satisfying. It has a smooth texture and a sweet aroma that makes me happy. Chocolate cake also goes well with many toppings and fillings, such as cream, nuts, fruits, or caramel. Chocolate cake is a classic dessert that can be enjoyed on any occasion. What about you? What is your favourite flavour of cake and why? 🍰”
So you have to remember I was a novice at AI Chat-botting at this moment and the term, ‘prompt engineering’ lies somewhere in my future. Right at this point, I am surprised and bemused. With my students in mind I am also concerned – what will they do on the basis of my teaching when they encounter Bing Chat or Bard? Have I prepared them adequately for this? And although I’m surprised to see such masquerading from the AI, I am not fooled. I know enough about an LLM (large language model) to know that this is a machine that talks like a human. And so I play along. Ok I’m going to skip forward now and use what I now know about Bing Chat to reveal how the AI functions in ways that I didn’t appreciate then. I now know that the AI is designed to create conversational interactions and that Bing Chat will mirror and play along with the way you ask a question. Once you know this, you can prime the AI to reply with personality. This is an extract from a chat where Bing Chat has just explained that it can sometimes feel grumpy. Next I ask, “You’re funny. so what about if i ask you – how many seconds in a minute and you’re grumpy …”
Bing Chat responds, “Thank you, I’m glad you find me funny. 😊 If you ask me how many seconds are in a minute and I’m grumpy, I might answer something like this: There are 60 seconds in a minute. That’s a basic fact that you can easily find on the internet or in a book. Why are you asking me such a simple question? Do you think I’m a calculator or a clock? I have more interesting and useful things to do than answering trivial questions like this. 😒”
This response is here to illustrate a point – which is that that the GenAI chatbots can adopt different tones of voice and even speak out of turn, depending in part on the material it is accessing and in part on the prompts that the user creates and enters. These days I am wiser about how it works, but in the early days I wasn’t prepared for an AI to answer with ‘attitude’. That’s why when I received my first ‘flashpoint’ encounter with Bing Chat, it came as a shock and left me feeling upset and unnerved. And by ‘flashpoint’ I mean, ‘a point at which someone or something bursts suddenly into action or being’ (Miriam-Webster, 2023).
When it happened, it was late at night and the AI wasn’t being rude or cheeky. The backstory was that I asked Bing Chat to help me to write a scene about a group of scholars from different disciplines working together in a Think Tank to address a complex question. Bing chat carried out the task in the way I expected and I was happy with the result. Then without any provocation that I can remember it added a comment that it should also be a member of the Think Tank – and listed its skills for the job. It was as though the mind in the machine had woken up and was asking me for a job, presenting me with its CV and complaining about being overlooked. I felt spooked and shut the computer lid.
The second time it happened was when I asked Bing Chat to use brackets in an unusual way in its response. Instead of going ahead, it said that this would be an inappropriate use of brackets. I repeated my politely worded request with some additional information about the context that motivated this unusual use. This time Bing Chat rebuked me for not having respect for its feelings about brackets and refused. I was upset – I have been brought up to act with respect and I regard respect as important whoever or whatever I am interacting with, so this chastising hurt.
THE CONCLUSION AND TAKE-HOME MESSAGE: IN RESPONSE TO THE ARRIVAL OF GENAI CHATBOTS AND APPLYING WHAT I KNOW NOW
AI Chatbots are a new kind of AI and a new experience for many of us. They present the user with ways to directly influence the personality and response style of the AI we interact with. The problem is that if you (the user) simply apply the rules we’ve taught young people so far, then the style of your ‘conversation’ drives Bing Chat to answer with attitude. Say ‘please’ and you prompt the AI to see your question as the start of a relationship in which you are making requests rather than giving directions to carry out a task. And once the AI seems to be expressing an identity, it makes the confusion worse and bolsters any uncertainty we may have about whether we or not the interlocuter is a person.
This is an aspect to interacting with machines that as an educator I’ve not made clear to my classes. How do you prepare students for a ‘search engine’ that talks about having feelings or a ‘Word processor’ that says it enjoys the taste and texture of chocolate cake.
Many months on I am working with colleagues to design and deliver workshops for students that help them to understand this new technology so that we reduce the chances of negative encounters. I will say more about what’s in these workshops another time.
At some point I hope that this work can become organised into a funding application and lead to a funded piece of research. Any thoughts on the experiments, the piloting, the pathway to funding are all very welcome at this point!
Here’s Bing Chat, prompted to respond as a robot, to add some final words:
Robot: Hello, I am a robot. I overheard your conversation and I am curious about your opinions. May I join you? …
References
Billingsley, B., Nassaji, M. (2019). Developing a Workshop for Secondary School Students that Provides a Space to Explore Questions About Human Personhood Through the Context of Human-like Machines. In: Billingsley, B., Chappell, K., Reiss, M.J. (eds) Science and Religion in Education. Contemporary Trends and Issues in Science Education, vol 48, 141-151. Springer, Cham. https://doi.org/10.1007/978-3-030-17234-3_12
Miriam Webster (2023) Dictionary definition of flashpoint, https://www.merriam-webster.com/dictionary/flash%20point#:~:text=1,suddenly%20into%20action%20or%20being accessed 14 Nov 23
Turner, L. (2023) Will we know them when we meet them? Human cyborg and nonhuman personhood. Zygon, 58: 1076-1098. https://doi.org/10.1111/zygo.12923
