Given the level of network connectivity between gamers, will the Turing Test be redundant one day? Since gamers seem to prefer interactions with flesh-and-blood players through the mediated technology of games (e.g. MMORPG)
I disagree with the premise that players prefer interactions with flesh-and-blood players. While MMORPGs are popular, other games which rely heavily on their Multiplayer element (ex. COD, Starcraft) still have long storied single player campaigns.
The recent game Titanfall is a good example of perhaps where some games will move towards - a mix of player characters (pilots) and AI driven mooks (grunts and spectres). The pilots are the elite units - who can call down large robots - with a high degree of autonomy on the battlefield. The AI grunts are the normal GIs with squad based tactics.
You cannot have players, who want to feel special, play as expendable cannon fodder. And expecting the level of dedication in order to have massive scale battles, with regimented lines - sounds more like work than play. There is some good examples of this in MMORPG raids, clans. But there is a high degree of buy-in. You need to maintain these relationships, and go on at set times, and plan. Many casual or even hardcore gamers do not have the time or inclination to want to maintain this level but want to have an experience with large army tactics.
Thanks for the in-depth sharing! Yes, amongst the many features you discuss, the on-demand features that a well-developed AI is perhaps one of the most convincing argument for the continuing relevance of the Turing Test. Could we say that AI caters for the multiple intelligences and diversity that defines an engaging gameplay?
The Turing test has to evolve for various reasons. Recent (still very simple) chatbots have shown that they can trick people into believing they're humans (for a short period of time, at least) by using various tricks. But there is more to intelligence than just that...
I have attached links to an interesting blog post discussion, and also a workshop of AAAI that happened just last January and discussed precisely how to find a replacement for the Turing test.
I agree. Intelligence has to go beyond human mimicry. And perhaps the Turing test's strong assumption that human intelligence is at the apex of the intelligence hierarchy is misleading? Thanks for the links to the workshop. As I have an on-going interest in cities, W02, really piqued my interest and will be monitoring advances in that trajectory as well.
The issue is really in your case Dr. Lew. Is that you have three major questions rolled up into one.
Issue 1: Is the Turing test reverent in games?
Issue 2: Is a Turing test a necessary requirement for an immersible game?
Issue 3: Can we generate a similar player experience with Human actors instead of machines and is this process now more tractable due to increases in speed of the network?
Thanks for clarifying. Yes sometimes muddled thinking can only be sorted out by a third perspective. Truly value the constructive feedback. Of the three I think that issue 2 comes closest to the "spirit of my question".
Yes as an intelligence test it is hard to beat because A.I. would essentially be another "life form" on par at least with humans. But if it is about giving an "impression" that you are playing/interacting with another human being, then I think the test may already be outdated because game telemetry may allow the game play to be updated in such a way that players that come on board later may "feel that they are player with a flesh-and-blood" player. When in actual fact it is the culmulative experiences of players thus far brought to bear on the most recent player of the game that gave this impression. I don't know whether this counts for "meeting the Turing test"
I have been giving some more thought to this question. There is the issue of game balance. Having a set of well constructed human competitive agents with different skill levels would allow for a system to do automatic balance of games. We could simulate hundreds of games in the time it would take to watch one human based play-though and yield a level of balance before presenting them to human players.
Very interesting notion on the AI- simulated and human- unraveling aspects of games and their reflexive/mutual relations- is there a concept that distinguishes the real-time vs. a priori aspects of AI (simulation to iterate all possible "worlds")?
Depends on how you structure your game playing AI and the game. It may act as an intelligent agent responding to stimulation by having a rules set and not require a simulation beyond that.
Perhaps the best analogy would be instinct v. contemplating. Chess, Go, etc. AIs will use the game tree and contemplate. Whereas, a decision tree, ANN, or FSM, will respond to stimulus of the game state(s) and react with a defined action.
Some methods use both, or may contemplate to learn, like muscle memory. Good chess AI don't use a game tree at the start or end of the game, they have tables of actions ready for common starting and endings, they are the memories of device. The mid game is simulated, and later on memory comes back to the end.
Thanks for all the insights, especially the implementation details and comparisons of possibilities. The timing of the whole thing seems to be the key. I guess good chess AI mimics the "chunking" approach (instinctive) to pattern recognition and then fall back on lookup tables (thinking/contemplating) to enhance the speed of the "thinking" process. Thinking and instinct, very interesting pair in cognitive psychology and behavioral economics as well.