2

I am trying to build consciousness in artificial intelligence. The general mechanism is that agent (randomly or based on some preexisting knowledge) performs some actions (gets some knowledge, makes some plan, performs physical action or mental action, communicates with other agents, invents new actions, improves itself, etc., etc.) and then gets rewarded for its action. The reward can be immediate. The reward can be delayed and sparse (e.g. only after completing a very long, nondeterministally long sequence of actions), but still - there is exploration-action-reward-improvement (learning) cycle that is repeated again and again. All is very simple if the agent has a clear set of preferences and the agent acts in a utilitarian framework. Then such agents can achieve anything that is possible in the Universe. There is one (out of many examples), how a robot with such a cycle http://robotics.sciencemag.org/content/4/26/eaau9354 got self-knowledge.

My question is - is there anything in philosophy that is going beyond this exploration+reward cycle? Can all the epistemology be cast in this exploration+RL approach? If it is not possible, then, please, mention the concepts in epistemology that could not be modeled in exploration+RL paradigm?

Of course, this applies both to individuals and societies, and this paradigm can capture the emotional, non-rational behavior as well and the changing, undeterministic sets of preferences as well. So - I am seeking something really, really apart! Even the religions use this reward concept.

closed as unclear what you're asking by Eliran, Gordon, Mark Andrews, Jishin Noben, Conifold Mar 13 at 7:42

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • I made some edits which you may roll back or continue editing. – Frank Hubeny Mar 11 at 2:07
  • "All is very simple if the agent has a clear set of preferences and the agent acts in a utilitarian framework". Maybe not so simple. This could burn through all the electricity we have. A good deal of information and energy is required to use the "utilitarian framework". Ie to try to prevent some fact that is not considered and the unintended consequences that could follow therefrom. – Gordon Mar 11 at 4:26
  • Why exactly can such agents achieve "anything that is possible in the Universe"? And if they can what exactly is stopping non-such agents from doing the same thing, on a whim, say? The exploration-reward presupposes an already pre-conceptualized reality with clear true/false on what is to be rewarded. Since in discovery situations the requisite concepts are missing, this framework is epistemologically inadequate. One needs to account for new concept formation, which can not be faithfully modeled, because if it could they wouldn't be new, unknown unknowns. – Conifold Mar 11 at 23:53
2

The OP asks the following:

My question is - is there anything in philosophy that is going beyond this exploration+reward cycle? Can all the epistemology be cast in this exploration+RL approach? If it is not possible, then, please, mention the concepts in epistemology that could not be modeled in exploration+RL paradigm?

One aspect that may need to be considered in addition to knowledge of facts is understanding of those facts especially if one wants to consider the knowledge to have something to do with consciousness. As a simple example, is an old style printed phone book containing a lot of facts conscious?

Emma C. Gordon addresses issues of understanding and epistemology:

Epistemology is often defined as the theory of knowledge, and talk of propositional knowledge (that is, “S knows that p”) has dominated the bulk of modern literature in epistemology. However, epistemologists have recently started to turn more attention to the epistemic state or states of understanding, asking questions about its nature, relationship to knowledge, connection with explanation, and potential status as a special type of cognitive achievement. There is a common and plausible intuition that understanding might be at least as epistemically valuable as knowledge—if not more so—and relatedly that it demands more intellectual sophistication than other closely related epistemic states. For example, while it is easy to imagine a person who knows a lot yet seems to understand very little, think of the student who merely memorizes a stack of facts from a textbook; it is considerably harder to imagine someone who understands plenty yet knows hardly anything at all.

Add to this John Searle's Chinese Room Argument that strong AI cannot understand the language it simulates in Turing tests and there may be a disconnect between consciousness and automated logic machines that centers around what understanding is all about. Wikipedia describes the Chinese Room as follows:

The Chinese room argument holds that a program cannot give a computer a "mind", "understanding" or "consciousness", regardless of how intelligently or human-like the program may make the computer behave.


Gordon, E. C. "Understanding in Epistemology" Internet Encyclopedia of Philosophy Retrieved March 10, 2019.

Searle, J. R. (1980). Minds, brains, and programs. Behavioral and brain sciences, 3(3), 417-424.

Wikipedia contributors. (2019, March 2). Chinese room. In Wikipedia, The Free Encyclopedia. Retrieved 01:53, March 11, 2019, from https://en.wikipedia.org/w/index.php?title=Chinese_room&oldid=885860786

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