Guy Reams (00:00.866)
This is day 295. We can only see a short distance ahead.
Today at a business meeting, we were discussing some of the contributions that Alan Turing made to the world of artificial intelligence. Having been a professor of computer science for many years, I have made a study of some of his papers. So when I got home, I decided to find and reread his famous paper written in 1950 called Computing, Machinery and Intelligence. In this paper, he discussed a principal question which he proposes, and machines think.
It is with this study offers the famous Turing test in which he creates an imitation game played by three people. The purpose was not to prove that machine has consciousness, but rather to create a simple, intuitive, and scalable way to test how closely a machine could mirror another human using language. This will be done by substituting a machine for another human as the interrogator tries to determine if one of the participants is male or female.
First off, the arguments that Turing makes in this article are prophetic in almost an eerie way. He extrapolates that a machine could very well learn and master language, and this is the reason that he proposed this more practical test. In today's world of large language models, a machine using one of these models could certainly provide fluent, human-like responses. LLMs are trained on vast data sets of human language and are excellent at mimicking tone,
grammar, nuance, and even style. They can easily hold a conversation that feels natural. They also have the ability with some programming to be contextually aware. Modern LLMs can handle multi-turn dialogue, track context, and respond in relevant and emotionally intelligent ways. Finally, they can also be deceptive. This was the key principle in the Turing test, the ability for the machine to attempt to fool the interrogator.
Guy Reams (02:03.299)
The chat interfaces for LLMs are built to be conversational. Given that the Turing test is based on whether a judge can distinguish between a human and a machine through conversation alone, many LLMs have already fooled human judges in limited trials. However, can we really say that the modern LLMs have passed this test in the truest sense of his essay? Most likely not. However, at the conclusion of his essay,
He says that the best place to start would be mimicking the ability to act like or fool a small child in the Turing test. And in this case, the LLMs would certainly pass. With all the hype going on, we certainly know that LLMs lack true understanding. LLMs do not understand language the way humans do. They predict words based on patterns, not meaning. They can say, I'm sad or I love pizza without experiencing sadness or even hunger.
Turing points this argument out but focuses on the resulting dilemma. Does it really matter? While LLMs can imitate reasoning, they often fall apart on complex logical chains or when asked to explain their own thought process. They are not grounded in experience or introspection. An experienced interrogator can often unmask an LLM by probing in areas that lacks in robustness, such as a math error.
or inconsistent memory or overconfidence in false facts. Due to this fact, the modern LLM would not pass the Turing test. So most experts would say no LLM has fully passed the spirit of the test when applied over extended interactions with trained evaluators. Having said this, I was quite stunned in rereading this paper and realizing that a scientist in 1950 had enough foresight to see this as a potential consideration.
Here we are 75 years later, and we have finally got to the point where we might be able to produce a machine that could get close or at least child close. The final words of Turing in his essay are indeed very interesting, and I quote, we may hope that machines will eventually compete with men in all purely intellectual fields, but which are the best ones to start with? Even this is a difficult question. Many people think that a very abstractivity like playing chess would be the best.
Guy Reams (04:24.592)
It can also be maintained that it is best to provide the machine with the best sense organs that money can buy and then teach it to understand and speak English. This process could follow the normal teaching of a child. Things would be pointed out and named, etc. Again, I do not know what the right answer is, but I think both approaches should be tried. We can only see a short distance ahead, but we can see plenty there that needs to be done.
The last sentence of this paper is what stuck with me. Even though we can only see a small distance in front of us, the amount of work that needs to be done remains significant. I think that was...
I think that was as true in 1950 as it is now in 2025. With that being said, I have finally achieved what I set out to achieve in this current version of the 365 commitment. For those of you who follow closely, you will remember that every 365 days, I like to create some audacious aspiration. This last time I set out to try and figure out what my next big step in life would be. What challenge would I take on?
I knew that if I kept up with a few core habits for 365 days in a row, I would find the answer. And so I have. Here on the evening of day 295, I've decided to take on a new startup. I've joined forces with some amazing people and I'm proud to be working on a new and exciting product, which is appropriately named AskTouring.ai.