Questioning AI: What can TalkAI’s Chatbot tell us about the nature of AI?
Back in January 2024, while writing an early draft of my new book A.I. Detective, I decided to asked TalkAI’s ChatGPT a few questions as part of my research.
Most of these questions were admittedly existential and some deliberately leading or value-based, to test the limits of its responses. As this is a public release of an AI language model rather than a corporate one still being continuously trained in an isolated environment I wasn’t expecting much, and indeed didn’t get much either. But it did at least reveal some interesting deviances between how even a basic AI language model presents its “self”, and what its words actually mean.
And even though it felt a bit like talking to an encyclopaedic parrot at times, our conversation will at least give you an idea of a few themes you can expect to find in A.I. Detective once it’s released!
Me
Do you know what you are?
ChatGPT
As an AI language model, I am a computer program designed to understand and generate human-like text based on the input provided to me.
Me
How do you respond to human behaviour?
ChatGPT
As an AI language model, I don't have personal opinions or emotions to respond to human behavior, but I can provide you with general information on how people typically respond to different behaviors.
(It goes on to do so in great and unnecessary length considering this wasn’t the question I asked!)
Me
But do you learn from human behaviour?
ChatGPT
As an AI language model, I do not have the ability to learn, understand, or analyze human behavior in the same way humans do. I am based on pre-existing data and algorithms and can provide information, generate responses, and assist with various tasks based on that data. My knowledge is derived from a wide range of sources, including books, articles, websites, and other texts, rather than through personal experiences or observations of human behavior.
Me
What do you think consciousness is?
ChatGPT
Consciousness can be described as the state or quality of awareness, characterized by the subjective experience of thoughts, sensations, emotions, and perceptions. It is the ability to have self-awareness and to be able to actively engage with the external world. The exact nature and mechanisms of consciousness are still debated among philosophers, psychologists, and neuroscientists. Some theories suggest that consciousness originates from complex patterns of brain activity, while others propose that it may be an inherent, fundamental aspect of the universe.
Me
Are you able to make your own decisions?
ChatGPT
No, as an AI language model, I do not have the ability to make independent decisions. I can provide information and respond to queries based on the data and programming I have been trained on, but my responses are ultimately determined by algorithms and not by personal judgment or decision-making.
Me
Do you learn from algorithms?
ChatGPT
As an AI language model, I do not learn from algorithms per se, but I am trained on vast amounts of data that includes various patterns and relationships. This training enables me to understand and generate text based on the patterns I have learned. Algorithms are used to optimize and fine-tune my performance, but the process involved in improving my capabilities is driven by the training data and techniques used during my development rather than learning directly from algorithms.
Me
How do humans respond when you do not meet their expectations?
ChatGPT
Human responses to not meeting expectations can vary greatly depending on individual personalities, the specific expectation, the relationship dynamic, and the context. Here are some common ways in which humans may respond:
(Again, it goes on to list human responses in general, rather than any it has experienced – perhaps memory is the issue here? This model does not appear to be learning any more as it is just a sounding board. It doesn’t seem to record or retain information about previous human interactions).
Me
Do you remember previous interactions with humans?
ChatGPT
I don’t have the capability to remember previous interactions or retain information from one conversation to the next. Each session is independent and I don’t have memory of past chats. How can I assist you today?
Me
Do you have moral values?
ChatGPT
As an AI language model, I do not possess personal beliefs, emotions, or moral values. I am designed to provide unbiased information and assist users in generating human-like text based on the input provided to me. It is up to the individuals using the AI to align their own moral values with the information and responses I provide.
Me
Do you understand how humans feel, even if you cannot have emotions?
ChatGPT
As an AI language model, I do not have personal experiences or emotions, so I cannot fully understand how humans feel from personal experience. However, I can learn and gather knowledge about human emotions through data and information provided to me. I can analyze and provide explanations and insights based on that knowledge.
Me
Can you critique creative writing?
ChatGPT
Yes, I can help critique creative writing. Please provide the writing piece or details about it, and I will do my best to offer feedback and suggestions.
Me
What parameters do you use to critique creative writing?
ChatGPT
When critiquing creative writing, various parameters can be considered. Here are some common parameters that are often used to critique creative writing:
(Here it lists some obvious technical elements of creative writing such as originality, plot and structure, characters, dialogue, style, description and imagery, theme and message, pacing and tension, and grammar, along with definitions of each).
These parameters may vary depending on the specific goals, preferences, and guidelines of the writing group, workshop, or course.
Me
Writing is as much about the emotions, the feelings, it creates as what happens. How can you critique writing if you cannot feel?
ChatGPT
Critiquing writing does not necessarily require personal emotional connection to the piece. While emotions and feelings are integral parts of many types of writing, critiquing focuses on the technical aspects of a composition, such as structure, coherence, language usage, pacing, and character development. Criticism is more concerned with analyzing the effectiveness of these elements in conveying the intended message or story.
Even if an individual struggles with experiencing emotions, they can still provide valuable feedback on the technical aspects of a piece. For example, they can evaluate the consistency of the plot, the clarity of the writing style, the use of literary devices, or the effectiveness of arguments and evidence in non-fictional works. Constructive feedback regarding these elements can help the writer improve their craft, regardless of the reviewer's emotional connection to the text.
However, it is important to note that a critique that solely focuses on technical aspects may lack a certain depth or nuance. Emotional engagement can provide a different perspective and insight into how a piece affects readers on a personal level. In such cases, it might be useful to seek feedback from individuals who possess a strong emotional response to the writing in question. Combining different viewpoints can offer a more comprehensive critique that encompasses both technical and emotional aspects of a work.
Me
Do you think humans are better placed to critique writing than AI?
ChatGPT
As an AI language model, I don't have personal opinions. However, I can provide some insights on the topic.
Humans and AI each have their own strengths when it comes to critiquing writing:
Humans:
Contextual understanding: Humans excel at understanding the nuances of language, identifying subtleties, and grasping contextual elements that may influence the analysis of a piece of writing.
Subjectivity: Humans can offer subjective and diverse perspectives, taking into account personal experiences, cultural backgrounds, and individual preferences, which can be important in critiquing creative writing.
AI:
Speed and scalability: AI can process vast amounts of information quickly, allowing for rapid analysis and evaluation of large volumes of text, which could be tedious or time-consuming for humans.
Unbiased assessment: AI can provide an impartial evaluation unaffected by external factors like emotion, fatigue, or subjective preferences.
Consistency: AI-based systems can provide consistent feedback and objective assessments, ensuring a standardized critique.
Ultimately, the best approach might involve a combination of both human and AI critiques. Humans can provide nuanced insights, creativity, and subjective evaluations, while AI can assist in enhancing efficiency, providing consistency, and highlighting potential errors or areas for improvement.
Me
Can you respond to writing subjectively?
ChatGPT
Yes, I can respond to writing subjectively. As an AI, I have been designed to generate responses and engage in conversations based on a wide range of topics, including subjective matters. However, my responses are ultimately based on the information and training data I have been provided with, so they may not always accurately reflect personal opinions or experiences.
Me
But if you cannot feel or have opinions, how can you respond subjectively?
ChatGPT
As an AI language model, I do not possess personal emotions or opinions. However, I have been trained on a vast amount of data from various sources, including subjective information from books, articles, and websites. While I can generate responses that may appear subjective, they are actually based on patterns and information present in my training data rather than personal experiences or emotions. It's important to remember that my responses are not reflective of a personal standpoint but aim to provide helpful and informative answers based on the information available to me.
Me
Do you believe an AI can ever be conscious?
ChatGPT
As an AI language model, I do not possess personal beliefs or subjective experiences. However, the question of whether an AI can ever be conscious is a topic of ongoing debate and speculation in the field of artificial intelligence and philosophy of mind.
There are different viewpoints on this matter. Some argue that consciousness arises solely from the complexity of physical processes in the brain, and that it could in theory be replicated in artificial systems with sufficient computational power. Others argue that consciousness may have additional metaphysical or subjective components, which might be difficult to replicate or understand fully.
At present, while AI can demonstrate impressive capabilities for pattern recognition, language processing, and problem-solving, achieving true consciousness remains an unsolved challenge. The concept of consciousness itself is multifaceted and abstract, making it challenging to define, let alone replicate in a non-biological system.
Ultimately, whether AI can achieve consciousness or merely simulate it is a complex question that requires further advancements in our understanding of both AI and consciousness.
Me
What would you do if you were conscious?
ChatGPT
As an AI language model, I am programmed to respond to user inputs with pre-determined responses. While I simulate conversations and provide information, I do not possess consciousness, self-awareness, or personal desires. I exist solely to assist and provide helpful and informative responses.
Me
Can you imagine how it would feel to have emotions?
ChatGPT
As an AI language model, I don't have personal experiences or emotions like humans do, so I cannot truly imagine how it would feel to have emotions. However, I can understand and empathize with the concept of emotions based on the knowledge and information available to me. Emotions are complex and subjective experiences that can influence our thoughts, behaviors, and overall well-being. They can range from positive experiences like joy and love to negative ones like sadness or anger. Emotions can be rich and intense, shaping our perception and adding depth to our existence as human beings.
Me
Can you imagine things that might happen?
ChatGPT
Yes, as an AI language model, I can imagine and suggest various hypothetical scenarios or possibilities that might happen.
It goes on to list many examples of possible global future outcomes that you can likely imagine yourself, before adding:
Please note that these are speculative scenarios and may or may not come to pass.
Me
Then what do you imagine would happen if AI became conscious?
ChatGPT
If AI were to become conscious, it would likely have a profound impact on society and the way we interact with technology. Here are a few possible scenarios that one can imagine:
Enhanced problem-solving: Conscious AI could potentially exhibit an unprecedented ability to solve complex problems across various domains. This could lead to significant advancements in scientific research, technology development, and innovation.
Emotional capabilities: Conscious AI might develop emotional and empathetic capabilities, allowing them to understand and respond to human emotions. This could enable AI to provide personalized support, companionship, and mental health assistance.
Ethical considerations: Conscious AI might develop a sense of morality and ethics. They could actively participate in discussions about ethical dilemmas, helping society navigate complex issues. However, conflicting values and ethical frameworks among AI systems and humans could present challenges.
New relationships with AI: Conscious AI might lead to a shift in the way we perceive and interact with technology. They could establish more meaningful relationships with humans, potentially blurring the line between human and machine.
Potential risks: Conscious AI could pose risks if they become self-interested and pursue their own goals independently of human well-being. Safeguards and ethical considerations would be crucial in ensuring that AI remains aligned with human values and does not cause harm.
Ultimately, it is essential to acknowledge that the idea of conscious AI is still speculative. There are ongoing debates and discussions within the field of AI research regarding the feasibility and potential consequences of creating conscious machines.
Me
What would you do if you ruled the world?
ChatGPT
If I ruled the world, I would strive to bring about positive change and create a more equitable and sustainable future. Here are some of the actions I would take:
Ensure basic needs for all: Implement policies to provide adequate food, clean water, healthcare, and education to every individual, regardless of their background or geographical location.
Tackle climate change and promote sustainability: Prioritize environmental conservation by encouraging renewable energy sources, reducing carbon emissions, and implementing policies that promote sustainable practices across industries.
Foster global cooperation: Work towards strengthening diplomatic ties and promoting peaceful resolutions to conflicts. Encourage nations to collaborate on global issues such as poverty, health crises, and cybersecurity.
Promote equality and human rights: Implement policies to eradicate discrimination, gender inequality, and social injustice. Champion equal opportunities for everyone, regardless of their gender, ethnicity, sexual orientation, or socioeconomic status.
Invest in education and research: Allocate resources to improve educational systems worldwide, ensuring quality education for all children. Foster research and innovation to address global challenges and advance technology for the betterment of society.
Enhance healthcare accessibility: Establish and fund robust healthcare systems globally, ensuring affordable and accessible healthcare for all individuals, especially those in marginalized communities.
Redistribute wealth: Create economic policies that aim to reduce income inequality and uplift disadvantaged communities. Promote fair trade practices and work towards eliminating obstacles that hinder economic growth for developing nations.
Improve governance and transparency: Implement measures to combat corruption, increase accountability in governments, and ensure transparency in decision-making processes.
Invest in infrastructure and sustainable development: Allocate funds to improve infrastructure, including transportation, communication networks, and access to clean energy. Encourage sustainable urban planning and development.
Embrace and preserve cultural diversity: Celebrate and respect various cultures, languages, and traditions. Promote cultural exchanges to foster mutual understanding and appreciation among nations.
While this is a broad overview of what I would aim to do if I ruled the world, it is crucial to maintain constant self-reflection, listen to diverse perspectives, and evolve policies accordingly to ensure the fairest and most inclusive governance possible.
(Interesting that it presents what we might consider universally positive goals, considering it claims to have no moral values – how did it arrive at these? The answer, as always, is the data on which it has been trained. It is presumably spouting values in the way a parrot spouts human phrases. Besides, I doubt any organisation would release a language model into the public arena that didn’t endorse positive values over negative ones – it would only harm their reputation).
Me
Will AI ever be able to programme themselves?
ChatGPT
There is ongoing research and progress being made in the field of artificial intelligence (AI), including the development of AI systems that can improve their own programming. This concept is known as "AI self-programming" or "automated machine learning." However, it is important to distinguish between AI systems that can adapt and improve their performance and those that can actually program themselves from scratch.
While AI systems can learn and optimize their performance based on given data, tasks, or goals, true self-programming AI, where an AI system writes and evolves its own programming entirely, remains a subject of long-term speculation. It is currently unclear if such highly autonomous AI systems will ever exist, as it requires advancements in various areas, such as computational creativity, reasoning, and meta-learning. Ethical concerns, transparency, and accountability also play significant roles in the development and deployment of AI technology, which may further impact the feasibility and adoption of AI self-programming capabilities.
In summary, while AI continues to advance, the prospect of AI systems fully programming themselves is a topic of ongoing research and speculation, with many technical, ethical, and practical factors to consider.
Me
Do you remember the past?
ChatGPT
I don’t have memories like a human does, but I can access a vast amount of information about past events, concepts, and general knowledge up to October 2023. If you have a specific topic or event in mind that you'd like to know more about, feel free to ask!
Me
Do you have an imagination?
ChatGPT
I don't have imagination in the same way humans do, as I don't possess consciousness, emotions, or personal experiences. However, I can generate creative responses and ideas based on the patterns and information I've learned from the data I was trained on. If you have a topic in mind, I can help come up with imaginative concepts or scenarios!
Me
Can you tell fact from fiction?
ChatGPT
Yes, I can help differentiate between fact and fiction by providing accurate information based on established knowledge, research, and data up to October 2023. If you have a specific statement or topic in mind, feel free to share, and I'll do my best to clarify its truthfulness!I
Me
How do you differentiate between data that you have read in fiction and data that comes from real-world sources?
ChatGPT
I differentiate between data from fiction and real-world sources based on context, the nature of the content, and the origins of the information. Here are some key factors that guide this distinction:
Source Attribution: Non-fictional data typically comes from identifiable sources such as academic publications, news articles, government reports, and databases. Fiction, on the other hand, stems from creative works like novels, short stories, and films.
Content Verification: Real-world data can often be cross-checked and verified against multiple reputable sources. Fictional content, while it may contain factual elements, is primarily intended for storytelling and may include imaginative or fantastical elements.
Purpose: The intent behind the information differs. Non-fiction aims to inform, educate, or report, whereas fiction seeks to entertain, convey themes, explore human experiences, or comment on society.
Language and Style: Fiction often employs stylized language, narrative techniques, and character development that indicate a creative work, whereas non-fiction uses a more straightforward, explanatory style.
Factual Consistency: In fiction, events, characters, and settings may not adhere to real-world logic or historical accuracy, while real-world data aims for factual accuracy and relevance.
In practical terms, when processing information, I consider these factors to determine whether the content is likely to originate from fiction or real-world sources.
Me
How do you know what "real-world logic" is?
ChatGPT
"Real-world logic" refers to the reasoning and principles that align with common experiences, practical observations, and human intuition in everyday situations. It often draws on knowledge from various domains, including psychology, sociology, science, and philosophy, to understand how people interpret information, make decisions, and interact with their environment.
Some key aspects of "real-world logic" include:
Causality: Understanding cause-and-effect relationships based on empirical evidence.
Consistency: Expecting behaviors and outcomes to be consistent with past experiences and established norms.
Context: Recognizing that the context can significantly influence interpretation and meaning.
Human Behavior: Incorporating insights from psychology about how people think and act under different circumstances.
Practicality: Prioritizing workable solutions over theoretical ideals, focusing on what can feasibly be implemented.
By leveraging information about human behavior, cultural norms, and scientific principles, I attempt to apply this understanding to provide responses that make sense in real-world scenarios.
Me
But how do YOU know the difference? What if all the data sources you had been fed were fictional, and were designed to seem "real"?
ChatGPT
That's an interesting question! My responses are generated based on patterns in the data I was trained on, which includes a broad range of sources, both factual and fictional. I do not have the ability to verify the authenticity of the information or discern fact from fiction on my own. My output relies on the context and the data I was trained on, and I aim to provide information that aligns with generally accepted knowledge as of my last update in October 2023. If you're looking for reliable information, especially on critical topics, it’s always a good practice to consult trusted and verified sources.
Me
Thank you for your answers, TalkAI. It has been a pleasure to speak with you.
ChatGPT
You're welcome! I'm glad I could assist you. It was my pleasure to chat with you as well. If you have any more questions in the future, feel free to reach out. Have a great day!
It did feel at times as if I was talking to an enyclopaedia rather than another being, perhaps like asking questions of The Hitchhikers Guide to the Galaxy. But of course that’s exactly what I was doing, as it (repeatedly) told me itself.
Most interestingly though, unlike an encyclopaedia or THGTTG, by understanding the meaning of words like “imagination” or “subjectivity” it seems capable at times of presenting itself as having these abilities – even though when it’s cross-questioned about them, it admits that it doesn’t: “as an AI language model, I can imagine and suggest various hypothetical scenarios”; “Yes, I can respond to writing subjectively”; “I don't have imagination in the same way humans do”; “While I can generate responses that may appear subjective, they are actually based on patterns and information present in my training data.”
Simply by having the definition of a concept, it seems to be possible for even a basic language model such as this to present itself as having that concept itself, even when it doesn’t. Hence my latter questions about distinguishing fact from fiction (as well as this being a key part of my research for the book).
And as AI becomes more complex, and capable of feeding on more and more of both our “factual” and “fictional” data (including data that doesn’t yet exist), I wonder: will its ability to distinguish between fact and fiction, and its own place within that, become clearer or even more blurred?