Rethinking Turing: My Take on Computing Machinery and Intelligence

The Paper is more of a philosophical thought process than an actual Research paper, it attempts to answer the question "Can Machines Think?"  

To answer this question Turing introduces the concept of The Imitation Game to replace the question “Can Machines Think"  due to the fact that the concept of "Thinking" or “Machine”  can be sometimes ambiguous. 

The imitation game involves three participants, a machine, a human and an interrogator. The idea is quite simple, the interrogator has a conversation with both the Machine and Human, if the Interrogator can't differentiate between the two then the  Machine is considered to be "Thinking". This idea builds on the argument  that "Thinking" should be measured by outputs and behavior rather than internal make up.

Consider your colleague. You can't  sufficiently prove that they are actually thinking internally by looking at them, For  you to do this you have to somehow be in their mind. However you can verify a certain level of thought process by looking at their behaviours, so Turing suggests that we focus on just the behaviour and neglect the internal make up.

The  idea of the imitation game  can be extended beyond just conversations, so more generally, Let's  Imagine a hypothetical body that  is identical to the human body such that this body can not be differentiated from a Human Body, a Computer Program(Christopher) lives inside this hypothetical Body, if Christopher can live amongst Humans for a certain period without being noticed then Christopher is “Thinking” or rather intelligent. But for how long does  Christopher  have to go unnoticed to be considered intelligent or  capable of “thinking”? In the paper the imitation game is played for 5 mins, for this Hypothetical situation how long should this game last for? Because for an infinite amount of time I argue that the Interrogator will always  be able to differentiate between the Machine and the Human. 

I give two illustrations to this claim

1. Image Generation: Let's look at two look a-like images created by a camera and SOTA diffusion model. At first glance you can't differentiate the two images but over time you begin to notice certain differences like the focal point these  differences usually stem from how these images were generated Internally 

2. LLMs : While having conversations with LLMs, for a while it seems and feel like it's another person that's at the other end of your computer, however after a long time  you begin to realize the difference, you begin to see the repetitions, the absence of creativity and uniqueness.

In the Paper, Turing went on to disprove contrary opinions like  The Machine will only do what it has been programmed to do by Ada Lovelace. This can be easily confirmed false as AI models are considered black boxes, there exists cases of goal misalignment, so if at certain times the LLMs doesn’t do exactly what we tell it, there is a chance it can do something we never anticipated, however the extent of this deviation might be little.

Turing went further to explain how a thinking machine could be birthed, he proposed that the task could be subdivided into two, which he modeled against an adult's mind. The Adult mind which was initially  a child’s mind that passed through some education. So he proposed that there should be creation of a child’s Mind and that mind should be passed through some Education that uses some form of reinforcement Learning through rewards, Punishments and a supreme source of truth. This idea can be seen in SOTA models, which usually involve some form of General training then some finetuning. 

More importantly he  proposed that there should be a random variable as this will aid in the faster solution of some problems. This i agree with as Randomness plays a vital  role in Learning, it is part of a group of factors which I would like to refer to as Natural factors, they include randomness, Uniqueness, Adaptation, reproduction and time, i think alot of researchers don’t pay attention to these factors yet, except Adaptation which is a recent development by Adaption Labs. I personally think the development of these factors will improve learning, however i hope we don’t end up creating a HUMAN while aiming for super intelligence 

The impact of this paper is truly tremendous as it gave a blueprint to how AGI should be developed, this paper gave rise to the TURING TEST, which is  a scenario where a machine plays the Imitation Game.

Thanks for reading this far. I would love to hear your thoughts in the comment section.

I used an LLM to help draft this post and it likely contains >10% AI-generated text, but I’ve edited/rewritten it extensively and endorse it.