0
\$\begingroup\$

Wikipedia says:

In video games, this usually means a character controlled by the computer through artificial intelligence.
~https://en.wikipedia.org/wiki/Non-player_character

In video games, a bot is a type of weak AI expert system software which for each instance of the program controls a player [...]
~https://en.wikipedia.org/wiki/Video_game_bot

However as far as I know such bots & NPCs are way too hard-coded or explicitly coded to be accurately called "artificial intelligence".

Mainly most often they lack the ability to learn dynamically from the player. Arthur Samuel calls "artificial intelligence" The "field of study that gives computers the ability to learn without being explicitly programmed" (1959).

So to what extend can bots and/or NPCs be called "artificial intelligence" if they (or a subset of them) can be designated as such?

\$\endgroup\$
6
  • \$\begingroup\$ Alan Turing thought about this extensively, and proposed the Turing Test as the appropriate means for testing AI: en.wikipedia.org/wiki/Turing_test \$\endgroup\$ Commented Mar 8, 2015 at 21:37
  • 1
    \$\begingroup\$ Without local definitions for the fuzzy terms "artificial" and "intelligence", I think this question is unanswerable. That's philosophy, not game development. \$\endgroup\$ Commented Mar 9, 2015 at 13:10
  • \$\begingroup\$ What game development problem are you attempting to solve here? \$\endgroup\$ Commented Mar 9, 2015 at 22:24
  • \$\begingroup\$ @Pieter Geerkens: See Computer game bot Turing Test. However both this variant and the original Turing test are no proper measurements for AI. Basically it's just a test for how well a program's designed to make it seem indistinguishable from a human - that's not the same as testing for artificial intelligence and dynamic behaviour / learning abilities. \$\endgroup\$ Commented Mar 19, 2015 at 11:15
  • \$\begingroup\$ @Josh Petrie I was recommended to post this question here. Is there another StackExchange site better suited for this question? I thought from the sites that do exist game development is probably the best place to ask about it as it's in the meta realm of GD and certainly not appropriate for Philosophy Exchange. \$\endgroup\$ Commented Mar 19, 2015 at 11:15

4 Answers 4

Reset to default
1
\$\begingroup\$

Artificial Intelligence is a very fuzzy idea without clear boundaries. The Oxford English Dictionary just defines it as:

the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.

The definition of "normally requiring human intelligence" has changed in meaning over time. Obviously, during the birth of AI, everything from simple addition upwards required human intelligence. Since them, the goal post has moved to be more and more complex. I highlighted "decision making" because to me it is the clearest example of NPCs being artificially intelligent. In my opinion, almost all software today is artificially intelligent in the sense described by the OED.

Also note that in Arthur Samuel's time, "learning" meant something entirely different than it does today. Samuel was referring to "expert systems" that determine answers to queries by deducing them using fundamental logic. This was considered "learning" back then, because the answer was not hard coded into the machine. Rather, the computer had to learn the answers by generalizing existing facts. Today, "learning" is generally considered to mean "statistical inference from data" -- an entirely different concept; wheras symbolic manipulation using fundamental logic is considered "programming." In 30 years, "learning" will probably mean something totally different as well. Already, "deep learning" is re-defining what we mean when we say a program has "learned" a concept.

\$\endgroup\$
0
\$\begingroup\$

The earliest, broadest definition of artificial intelligence included (what we now call) calculators. This was refined to exclude calculators because they could not be "taught" how to perform new calculations. Then, (what we now call) computers were invented, which could be reprogrammed to perform any conceivable algorithm. Again, we excluded computers on the basis that they could not program themselves but had to be instructed. We therefore assume that once a system is created which can learn to perform a task on its own (like this program which can learn to play certain classic NES games), we will have conquered AI.

Machine learning is its own highly developed domain, but it's usually hard to implement or is very slow. For games where it can be easily run in real time, it often makes games so difficult that they become unbeatable.

So to protect our own ego, let's call it a problem of definition.

\$\endgroup\$
0
\$\begingroup\$

"Artificial Intelligence", as you have already guessed from the other answers, is only loosely defined, and does not have much to do with intelligence. Personally I would not count learning as a requirement to call something an AI.

In Artificial intelligence, the common goal is to produce rational agents. Rational, because rational can be defined and achieved, and intelligent cannot. Rationality is defined as making the optimal choice, given the available information, to achieve some goal. The model is one of an "agent", an acting entity that receives some kind of sensory input, and can act. What is in-between can be anything from a rule-based hard-coded system, to planning systems and machine learnings. An agent could be a robot (with cameras for sensors, lots of motors for actuators), a security system, a wikipedia bot, a poker AI, or a game NPC.

Yes, NPCs are artificial intelligences, but the focus is on artificial and not intelligence.

An excellent book on the topic is "Artificial Intelligence: A Modern Approach" by Russel and Norvig http://aima.cs.berkeley.edu/

\$\endgroup\$
0
\$\begingroup\$

As someone who has worked on artificial intelligence applications outside the domain of gaming, I can tell you that what Hollywood convinces people AI is supposed to be is completely wrong. Anything that makes decisions based on learned heuristics is an application of AI. So bots, who are taught some rudimentary gameplay rules, objectively qualify as artificial intelligence.

Traditionally the rulesets used by game AI have been very simple and have been devised through human means. However, with the emergence of cloud computing, we are starting to see sophisticated game AI that uses data collected from massive samples of real-world players to generate many novel rules that humans probably never would have thought of. Forza 5 and Forza Horizon 2's "drivatar" system is the only concrete example I can give at this time. However, EverQuest Next is said to be using "emergent AI," which will also use data mining on a per-server basis to establish independent AI behavior on every server.

\$\endgroup\$

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.