Duopsism - a person and his computer

A person is sitting in an airtight room. They know nothing about the outside world, not even if there is a world outside the room.

The room contains a monitor and a programmable black box. The person can use it for calculations or broadcasting and receiving messages, though where the messages go to or come from is unknown.

What can the person deduce about the world using this equipment?

This experiment addresses the difficulty distinguishing humans from artificial in the post-AI Internet.

1. 📦 The Programmable Black Box

A programmable black box is a machine equipped with a set of instructions (think of something like RISC or CISC). The idea is to have a computer the person can only use but not look into.

Which instructions are included exactly? We leave this question open, but require at least the following kind of instructions:

1. logic and arithmetic instructions,
2. instruction to send (to “somewhere”) and receive (from “somewhere”) messages,
3. instructions to parse and decode messages.

The person does not know where the messages go to or come from. For all the person knows, none of the messages ever leave or enter the airtight room; The person may simply be communicating with a simulation within the black box.

2. 🕴️The Person

What do we think of as a person? A person is an entity with… wait WAIT! First, let’s see what the person thinks about itself. Here is a trivia detail from the year 1637:

3. 🧠 Cogito, Ergo Sum

In 1637 published René Descartes the first principle of philosophy, Cogito, Ergo Sum, or in English, “I think, therefore I am”.

This principle is a foundation of Western philosophy and is considered as proof that I, or something, exists. Descartes argued that you can’t doubt the fact that you doubt. Someone needs to be making these doubts.

Ok, now let’s get back to the…

4. 🕴️The Person (again)

When we say “Cogito, Ergo Sum”, what do we think the “I” is? There are more philosophical texts on that than a homebrew AI can digest. However, influenced by computer science, let’s try to make our own definition.

An object is called a person if it has access to:

1. private memory,
2. private processing power.

We call it shortly private resources.

This means that only this object can access the resources. No other object can.

Examples? Here you go:

• Human is an entity. your brain is both the private storage and processing.
• Human might be two entities. Like, if you consider the subconscious as a separate private resource.
• *Personal computer is an entity**. The HD is the private storage whereas RAM is the private memory.
• Also a computer might be two entities. This happens if you consider TEEs as a second entity.
• Is an AI an entity by this definition? Or is it a machine operated by entities?
• Is an organization an entity under this definition? or is it a collection of entities? I’d argue it’s an entity since the amount of memory and processing it does is bigger than what each of its members is capable of on its own.

5. 🕴️The Person (alternative)

Here is an alternative definition of a person. It is a subjective one that makes it easier to define AIs and organizations as persons.

Person A says that entity B is a person if it looks to A that B has access to:

1. private memory,
2. private processing power.

Under this definition, it doesn’t matter if B really has access to private resources but only how it looks like to A.

Since it does seem from the outside that AI and organizations have access to private resources, then they are persons by this definition.

Which one fits better depends on you. Are you looking for objective statements about reality? or for practical statements, which are true in a specific situation?

6. 🧠🧠 Agit, ibi est

Cogito, ergo sum is explained as “I cannot doubt the fact that I doubt”. In terms of a person with private resources, it can rephrased as:

• Cogito, ergo sum (I). “every action requires processing. Doubting is an action. Hence, I doubt means I process, meaning I’m a person”.

Our specific thought experiment includes a second object, namely, the programmable black box. Using this we can push deduction further to get what I call “agit, ibi est”. This is Latin for “It acts, therefore it is.”

• Agit, ibi est (It). The person receives messages from somewhere. Creating these messages is a process that needs resources. We do not have access to these resources, meaning someone else created them. Hence, another person exists.

The person sitting in the room doesn’t know whether “I” or “It” are humans, machines, AIs, or a simulation. But it does deduce the existence of two entities with private resources.

7. 🌍 The World

A person sits in an airtight room with a programmable black box. What does the person know about the world so far?

1. 🕴️ The person exists, cogito, ergo sum.
2. 🕴️🕴️ Another person exists, Agit, ibi est.

This is all knowledge about persons. But can the person also deduce anything about “the resources”? I.e., about the existing processing power and memory in the world?

For this, we extend the assumptions about the instructions set. Assume that the black box instruction set includes:

• Randomality instructions (generate random numbers),
• Clock instruction, like getting the current time.

Then we can:

7.1 📦 Estimate The Computational Power of The Box

How? Like this:

1. Build a “Counter Program”, that counts till some number \(C\).
2. Mark the start time.
3. Run the counter program.
4. Mark the the end time.

This shows the black box can do at least \({C \over EndTime-StartTime}\) addition operations in a unit of time.

This naive algorithm is obviously inaccurate and inefficient. You can design better ones. But it fulfills our goal of giving some basic lower bound on the computational power of the person.

But we already know from “Agit, ibi est” that there are more persons in the world. Can we estimate their computational power?

7.2 🌍 Estimate The Computational Power of The World

Bitcoin uses proof-of-work to give an estimate of the computational power of the miners. We can use a similar idea in order to estimate the computational power of the world in our thought experiment.

7.2.1 🐣 The baby proof of hash algorithm

The person does as follows.

1. Design a hash function algorithm whose output is in the \([0, \dots, N]\) range, for some N. Program it in the machine.
2. Broadcast the hash function algorithm and implementation to the network, so everyone else would know.
3. Mark the start time.
4. Broadcast two numbers:
   • a random number \(X\) called “the seed”,
   • a small random number \(Y\) (within the hash range) called “the range”.
5. Ask the network to calculate hashes of the seed with random numbers, till one of the hashes is smaller than the threshold, and then send you the random number they found.
6. Verify the result, and mark the end time.

Assuming the hash function is randomly distributed, the probability that a random hash is smaller than the “range” \(Y\) is \({Y \over N}\). In order to complete the task, an average of \({N \over Y}\) hashes needed to be done.

Since we also estimate the time in which these N/Y hashes were performed, we can estimate a lower bound of the computational power of the world to be:

Performing this experiment once is not enough to give an estimate though. It needs to be performed multiple times for the estimate to make sense.

Needless to say, unlike Bitcoin’s proof-of-work, this algorithm gives no incentives for anyone to participate in it. But this is a thought experiment; let’s assume they participate nevertheless.

This estimate tells the person if there is anything outside the room. It might as well be the calculations are done within an extra processor, that the person doesn’t have access to, which is inside the programmable black box in the room.

Can the person build similar algorithms for estimating the proof of memory (storage) of the world? Let’s leave it as an open question.

8. ♾️ Conclusion

This thought experiment came up on two different occasions in the last few years.

The first was during the Covid lockdown, where I felt like this person locked in an airtight room, communicating with the outside world via a computer, losing faith in its existence.

The second is quite recently, where the rise of AI made it harder to know who’s human and whose a machine online.

This thought experiment is slightly related to the The Dead Internet. It is a conspiracy theory stating that most of the content in the web is AI-generated where most participants are bots, created by powerful organizations. Ironically, what seemed absurd when the theory was invented, only 3-4 years ago in 2021, seems more reliable now.

This thought experiment might also remind people of Mary’s Room thought experiment. There, Mary is a scientist living in a black and white room, with no colors, but all the books in the world. She studied everything possible about color without seeing it. Would Mary study anything by going outside and seeing colors, given that she already knows everything possible about them?

Does this experiment described here have any importance? It doesn’t matter. It was a fun post to write, I want to write more such things in the future.