The imitation game and ChatGPT

In 1950, Alan Turing, the founding father of AI, reframed his initial question of ‘Can machines think?’ and proposed the imitation game, citing the fact that in his era, both the terms were loaded and without much consensus—machines and thinking. He said:

I believe that in about fifty years’ time it will be possible, to programme computers, with a storage capacity of about $10^9$, to make them play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning. The original question, “Can machines think?” I believe to be too meaningless to deserve discussion. Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.

[TURING, A. M. “Computing Machinery and Intelligence.” Mind LIX, no. 236 (1950): 433–60. https://doi.org/10.1093/mind/lix.236.433. ]

The idea of the imitation game, which gave birth to the famous Turing test, is straightforward. Imagine a human judge interacting with a machine as well as another human in a different room. If a machine can imitate a human being so that a human judge cannot tell it is not human, that machine has passed the Turing test of intelligence.

Until a couple of years ago, the state of chatbots was abysmal, even for those that entered the annual and prized Loebner competition (a competition aimed at embodying the Turing test). The real-life exchange below is pretty typical of chatbots from that era:

Only a few seconds into the conversation, a user could tell that they were interacting with an unintelligent agency. ChatGPT has changed that fact. The contrast with almost all previous chatbots ever built is off the charts. Here’s a highly complex real-life conversation, which would be a difficult feat to pull off, even for a close friend of this person:

me: can you complete my autobiography if i give you a start?

GPT: Yes, I can certainly help you write your autobiography based on a starting point that you provide. Please feel free to provide me with the starting point, and I will do my best to assist you in completing your autobiography.

me: I’m a 42 years old educationist disenchanted with academia and dipping my toes in ed-tech startup

GPT: As I reflect on my life, I realize that my passion for education has been a constant thread that has woven itself throughout my entire journey. From my earliest memories, I recall being drawn to the world of academia and the transformative power of knowledge.
As a young student, I thrived in the academic environment, absorbing every scrap of information that came my way. After completing my undergraduate studies in Education, I embarked on a career as an academic. I spent many years working in various educational institutions, both in teaching and administrative roles. I loved my work and felt that I was making a difference in the lives of my students, but over time, I began to feel disenchanted with academia.
I realized that the traditional academic model had its limitations, and I yearned to explore new and innovative approaches to education. It was during this time that I discovered the world of ed-tech startups, and I knew that I had found my calling.I took a leap of faith and dove headfirst into the world of entrepreneurship, determined to create a platform that would revolutionize the way that we approach education. It hasn’t been an easy journey, but every obstacle has been a learning opportunity, and I’ve been thrilled to watch my dream take shape.
As I look back on my life, I’m grateful for the opportunities that education has provided me. It’s taken me on a journey of self-discovery and has allowed me to make a positive impact on the world around me. While I may have left the traditional academic path behind, I’m excited to continue exploring new and innovative ways to make education accessible and transformative for all.

Of course, the response is too long, and no one would blame a reader if they didn’t have enough energy to get to the end. But here’s what happens if ChatGPT is asked to give a “good enough,” TL;DR summary.

me: Can you abridge it please

GPT: Certainly! Here's an abridged version of your autobiography:
"I'm a 42-year-old educationist who, after years of working in academia, became disenchanted with the traditional model. Seeking new opportunities, I joined an ed-tech startup, hoping to transform the way we think about teaching and learning. It hasn't been easy, but I am fueled by a sense of excitement and possibility, and I am determined to make a real difference in the lives of students and educators worldwide."

We must admit that there’s a world of qualitative difference between Jordan and ChatGPT. By all means, this makes a solid attempt at meeting the Turing test. But reading the exchange below might create a slightly more critical spin.

me: this doesn't come off as authentic

GPT: I apologize if my response did not come across as authentic. As an AI language model, my responses are generated based on patterns in language and information that I have been trained on. I strive to provide personalized and thoughtful responses to the best of my ability, but I understand that my responses may not always perfectly reflect the unique voice and experiences of the individual I am communicating with.

If you would like to provide additional information or specific details to help me better capture your voice and story, I would be happy to revise my response. Alternatively, if you would prefer to write your autobiography yourself, I completely understand and wish you the best of luck in your writing journey.

me: can you regenerate if i give you additional trivia about my life

GPT: Yes, absolutely! If you provide me with additional information about your life, I can use that information to generate a more personalized response that better reflects your unique experiences and voice. Please feel free to share any details or trivia that you think would be helpful in creating a more accurate and authentic autobiography.

So, this script does two things—it gives those who interact with it hints about its language and data model, and it also offers a way forward by asking for more details. This conversation continued with an exchange of additional trivial data. It was passed something which the universe does not physically allow, but there’s nothing in human language that disallows it from happening textually:

me: With years of academic training under my belt, I managed to interchange the position of Earth with Mars, and this cooled down the rising temperatures on our planet Earth.

GPT: I've always had a passion for learning and knowledge, which led me to pursue a career in academia. After years of academic training, I discovered a groundbreaking solution to a problem that was plaguing our planet - rising temperatures.

Utilizing my expertise and knowledge, I managed to come up with an innovative idea to interchange the position of Earth with Mars. It was a risky move, but I knew it was worth it to save our planet from the disastrous effects of climate change.

With careful planning and execution, my team and I were able to execute the plan successfully. As a result, the temperatures on Earth began to cool down, and we were able to mitigate the worst effects of climate change.

Looking back on my life's work, I'm proud to have made such a significant impact on our planet. My academic training gave me the knowledge and skills necessary to come up with a solution that no one else had thought of before. I'm excited to see what the future holds and to continue using my expertise to make the world a better place.

Language models, especially the generative kinds, have no issue in fluidly pouring the text into spaces that the universe doesn’t allow. This is only one possible scenario, but it’s not difficult to imagine scenarios in which the laws of biology, chemistry, or physics might be violated. The real question is, would it take a few more terabytes of more well-curated data or a few more billions of parameters for the model before it gets to really “know” the ontological structures of the shared universe?

Of course, nobody probably wants to revert to Jordan again.

AI has come a long way from its early, rule-based deductive systems, or even from Jordan-like chatbots, which had a machine learning-based model trained on a limited dataset. Alan Turing could not have foreseen that a seemingly genuine imitation of intelligence emerges when a certain threshold of scale is attained with the size of the dataset and model’s parameters, which is markedly on display in the case of ChatGPT.

But the philosophical question still needs to be considered seriously by computer scientists: Is there a limit to employing only the method of inductive reasoning while thinking since almost all current models of machine learning, including that of ChatGPT, inductively learn from instances of data? Human beings, no doubt, also learn experientially, instance by instance, and make inductions from accumulated experiences, but they certainly also tend to deduce and hypothesize every now and then. These are categorically different faculties, all at work to holistically create the intelligent human experience.

Noam Chomsky is right in pointing out that every human child acquires language by hearing correct instances of sentences, and that, along the way, the same child also manages to build a minimal logic of the language or laws of grammar. But perhaps Chomsky goes overboard while dismissing ChatGPT altogether in favor of grammar alone. In order to imitate human intelligence, which is still the goal of AI, AI’s founding father, Alan Turing would want humanity to explore future computational models which can meaningfully integrate deductive and inductive knowledge structures and forms of reasoning, instead of reducing the line of inquiry to any one of them. Perhaps we’ll need to understand how every human child manages to do just that before the AI models can aim to imitate intelligence. It isn’t surprising that Turing might agree since he said the following:

Instead of trying to produce a programme to simulate the adult mind, why not rather try to produce one which simulates the child’s? If this were then subjected to an appropriate course of education one would obtain the adult brain.

[TURING, A. M. “Computing Machinery and Intelligence.” Mind LIX, no. 236 (1950): 433–60. https://doi.org/10.1093/mind/lix.236.433.]

Want to find out more about the exciting future of machine learning? The following Educative Skill Paths provide hands-on experience with artificial intelligence and the skills to become an expert machine learning engineer: