0

I'm not very familiar with this kind of formal philosophy discussions so please excuse my unpreciseness.

I know that the Artificially intelligent agent we have been putting the most hopes into recently for gaining human like capacities, uses techniques such as Reinforcement Learning where the AI is trying to make a "reward" numerical variable increase by performing actions.

This looks a lot like the way we are acting to me, as most of the time we do things with the hope to achieve happiness or whatever other output. Also the fact the architectures of these AIs are more and more ressembling our own brains indicates that we could learn things about ourselves by studying these AIs.

Is there a specific term refering to and/or field of research in philosophy where people are trying to compare how our brains are built with how we build our AIs in order to determine exactly whether we function identically to them, and if so, how our "reward variables" would be encoded?

To me it seems to be the most promising lead we have ever had in human history in the quest of finding meaning to our behaviours.

0

Direct Answer to Your Question:--

This field is simply:--

  • computational neuroscience
  • computational theory of mind

A more philosophical angle and less technical- / theory- based answer is: computational theory of mind.


What This Answer is About:--

  1. AI-driven rewards/goals
  2. Abstraction Modelling of the Human Brain Parallel to Abstraction Modelling of Artificial Intelligence

This answer will discuss neural networks. They are tools used in machine learning. Neural networks will transform input into via self-references into information that the output layer can use.

Neural networks are tools to find for finding patterns, especially those too complex for standard tools and methodology to find.


Layperson's Explanations:--

"Artificial neural networks are one of the main tools used in machine learning. As the “neural” part of their name suggests, they are brain-inspired systems which are intended to replicate the way that we humans learn. Neural networks consist of input and output layers, as well as (in most cases) a hidden layer consisting of units that transform the input into something that the output layer can use. They are excellent tools for finding patterns which are far too complex or numerous for a human programmer to extract and teach the machine to recognize." — "What Is an Artificial Neural Network? Here's Everything You Need to Know." Digital Trends, 6 Jan. 2019, < www.digitaltrends.com/cool-tech/what-is-an-artificial-neural-network/ >.

(https://www.digitaltrends.com/cool-tech/what-is-an-artificial-neural-network/)

This article discusses the philosophical implications of neural nets, in particular to ethics. I have included this article to make this a more philosophical article, and one less for < ai.stackexchange.com >. It applies A.I.-driven goals towards ethics and morality:--

Artificial Intelligence and Its Implications for Future Suffering

This is another source in simple language:--

(https://simple.wikipedia.org/wiki/Artificial_neural_network)


Technical Explanations:--

Compare the OP's request to the Wikipedia text below:--

"Is there a specific term refering [sic] to and/or field of research in philosophy where people are trying to compare how our brains are built with how we build our AIs in order to determine exactly whether we function identically to them, and if so, how our "reward variables" would be encoded?" ~ Caunes Andrew, Stack Exchange user

-

"Computational neuroscience is a branch of neuroscience which uses computational approaches, to study the nervous system. Computational approaches include mathematics, statistics, computer simulations, and abstractions which are used across many subareas of neuroscience including development, structure, physiology and cognitive abilities of the nervous system." — Wikipedia contributors. "Computational neuroscience." Wikipedia, The Free Encyclopedia. Wikipedia, The Free Encyclopedia, 27 Jul. 2019. Web. 1 Sep. 2019.

(https://en.wikipedia.org/wiki/Computational_neuroscience)

A more philosophical angle and less technical- / theory- based answer is: computational theory of mind:--

"In philosophy, the computational theory of mind (CTM) refers to a family of views that hold that the human mind is an information processing system and that cognition and consciousness together are a form of computation. Warren McCulloch and Walter Pitts (1943) were the first to suggest that neural activity is computational." Wikipedia contributors. — "Computational theory of mind." Wikipedia, The Free Encyclopedia. Wikipedia, The Free Encyclopedia, 19 Apr. 2019. Web. 1 Sep. 2019.

(https://en.wikipedia.org/wiki/Computational_theory_of_mind)

" ... how our 'reward variables' would be encoded?" ~ Caunes Andrew, Stack Exchange user ... "

This discusses theoretical research regarding reward variables in neuroscience:--


Other Sources and Further Reading:--


[Disclaimer: This answer is not medical and/or therapeutic advice. It is theoretical writing.]

0

Neural Networks are inspired by the human brain but you can't compare them in any way. The human brain cells are way too complex to be simulated by a neural network.

An artificial neural network is just a pairing with inputs and outputs and the connection between those inputs and outputs are adjusted by simulating cases. The inputs and outputs represent the electricity that go through our brain but there are more mecanisms like the action of some molecules (drugs for example) on the brain. The training process is also way more complex. The neurons can expend to new neurons to make connections. Since an artificial neural network works with layers, we cannot observe such a behaviour.

So now I tried to make the differences between human brain and A.I., I think you should start reading about Neuroscience : https://en.wikipedia.org/wiki/Neuroscience This field is young and there are some serious opponents to this but I think this is a good entry point to the question. The thought behind this field is that neurons are the bases of all our thoughts but also of all the mecanisms (organs).

Another thing you should look at is Materialism : https://en.wikipedia.org/wiki/Materialism This field is the monism that consider everything as the result of interactions between physical items.

A more precise field is the Computational Theory of Mind : https://en.wikipedia.org/wiki/Computational_theory_of_mind CTM is a thought that hold that our brain is like a computer and everything in that brain should be considered as a pairing of inputs and outputs. In my opinion, you should start by neuroscience and materialism but all the questions you think about should be answered (at least partially) when you will start looking at CTM.

0

tl;dr- We'll be able to construct better models for human minds by reducing general AI to human-like simplifications. However, since human intelligences are hive-minds which seek many different objectives, it seems almost certain that we won't find a single objective shared throughout a human mind.


Optimization is the field of minimizing/maximizing a value.

I know that the Artificially intelligent agent we have been putting the most hopes into recently for gaining human like capacities, uses techniques such as Reinforcement Learning where the AI is trying to make a "reward" numerical variable increase by performing actions.

Optimization is about extremizing a numerical variable called an "objective function". Artificial intelligence is a subset of optimization.

Note that some sources will talk about maximizing an objective function while others will talk about minimizing it. It's the same thing either way; add a negative sign to it, and maximization turns into minimization and vice-versa.


We'll be able to describe humans as reductions of general intelligence.

Is there a specific term refering to and/or field of research in philosophy where people are trying to compare how our brains are built with how we build our AIs in order to determine exactly whether we function identically to them, and if so, how our "reward variables" would be encoded?

Once we have super-human general AI, we could try to describe human intelligences as reductions of it. Once we find reduced variants of general AI that seem pretty human-like, then we can say that humans are like general AI operating with such simplifications.

As hive minds, humans don't have single objective functions.

That said, it seems pretty obvious that humans don't have singular objective functions; rather, parts of our brains act independently.

For example, consider your country as an intelligent agent: what's its objective function? Is it particularly meaningful to see your country as a singular intelligent agent that seeks a single thing?

The United States of America does exist as a conscious entity. It can be said to be sentient; it thinks, it has desires, it's self-aware, it has an identity, it seeks to maintain its own existence, etc.. But, it doesn't seem too reasonable to say that the USA has a single objective function that it seems to extremize.

Instead, it seems more reasonable to look at the USA as being composed of many subordinate intelligences with varying agendas. Exactly what those subordinate intelligences are depends on how far down you want to step – I mean, you can first consider individual regions of the US, or political parties, or human individuals, or human neurons, etc.. Let's pick US political parties, e.g. the Republican and Democratic parties.

Then, what does the Republican party want? And what about the Democratic party? Again, no single thing – while it's more informative to consider the parties individually, they're still not singular intelligences with singular goals.

We can then divide up the Republican and Democratic parties into, let's say, individual humans within them. Again, it's now more informative, but my point's basically that there's still not really a singular objective function.

Analogy: Companies are hive minds of employees.

Think about how a commercial organization (company) operates: many people have individual duties, right? The sales staff will keep trying to sell while the programmers will keep trying to program and the janitors will keep trying to clean.

In principle, the CEO is responsible for organizing everything such that all of these moving parts meet the organization's objectives. They delegate tasks down the chain-of-command, where subordinate executives have more focused responsibilities, down to the workers who perform tasks.

In reality, it's not so perfect. For example, the sales people might be incentivized to sell for commissions, perhaps making decisions that optimize their personal objective functions that don't precisely align with the company's own objective function.

Hive-minded thinking is appropriate for parallelization.

Obviously, it's undesirable for a company to have employees acting out-of-alignment with the company's overall objective function. So why do major companies do it anyway?

Because they have to! Because, if a CEO had to make every little decision constantly, nothing would ever get done. Instead, executives have to learn to delegate work to their subordinates. Equivalently, executives have to tweak their subordinates' objective functions to better align them with the company's overall objective function.

Ditto for the human brain.

This is, the little parts of the human brain aren't observing singular objective functions, but rather lots of 'em. Different parts of the human mind truly want different things. And this is necessary to have a reasonable level of overall intelligence, as a singular line of thought can't reasonably manage all decisions the human mind must make in a practical timeframe.

Conclusion: There's not one objective function human minds seek.

Point of all this being that human minds aren't seeking a single end. Rather, humans have a huge mix of objectives that different parts are seeking.

Of course, we can try to analyze people to determine specific objectives – just, my point's that there's not going to be an appreciably singular answer.

0

Many commentators have responded citing computational theory of mind as the basis for studying AI, but the philosophy that more accurately encompasses your question is the broader topic of the theory of mind, of which the computational model is just one approach. The difference between the two also highlights a difference in the approaches to creating artificial intelligence: narrow or weak AI and broad AI or AGI. Ultimately, picking in choosing from the various philosophies particularly to build a basis for studying AI results in yet another similar philosophy: philosophy of AI.

Many philosophers have contributed to the field, too numerous to list here, however 5 well-recognized contributors are:

  1. René Descartes
  2. David Hume
  3. Gilbert Ryle
  4. Daniel Dennett
  5. Jaegwon Kim

The last thing to note is that there are two general approaches to AI from an architectural perspective: the former seeks to simulate the function of neurons is known as connectionism, and the latter seeks to emulate the rational powers of man is known as symbolism. The former is generally viewed as the way by which the mind uses intuition, and the latter aligns more with formal logic like deduction, induction, and abduction.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.