Think of HAL9000 playing a chess game with Frank Pool:
Thank you for a very enjoyable game.
Would that have been literally enjoyable? In a world where A.I. exists, as depicted in this kind of science-based sci-fi, would entertainment exist as well for these sentient entities? Can you think of what kind of entertainment that would be? And would that be generated by the A.I. (or other A.I.s) itself, or could it be made by humans?
To expound on part of the answer Peter provided:
I'd like to add that the kind of value we get from games... & I believe most creatures derive from them, is some form of increased capability.
In our case, as I understand it, the process by which we increase our capabilities tends to be through courses of pattern matching & problem solving produced by the analysis of the patterns.
i.e. Our process is that we:
Throughout the above process we gain fulfillment and/or enjoyment under the following conditions:
Exploration: When we try something new related to the 'game'.
Success: When the modifications to the performed actions (which we theorized in step 3) relating to our modified tactics (step 4) that we theorized would improve our performance, actually did improve our performance.
Discovery: OR we can conjecture a new possible theory based on the fact that our modifications did NOT improve our performance, (this requires more effort and allows a higher level of attainment)
(In fact I would guess that the process could really be summed up by Discovery...
Discovering how to improve, or discovering more about the topic that might suggest further improvement, or uncovering details about the topic through experimentation/exploration...).
It seems likely to me that any intelligence capable of self improvement would follow roughly the same process of game playing to increase performance. I would also suggest that any such intelligence must, therefore, gain some level of fulfillment or enjoyment in order to motivate the improvement.
Now whether or not HAL enjoyed the game of chess would then depend on whether he has the capability to as outlined above.
(And my personal guess, for what it's worth, is that HAL enjoyed it because he was playing a game of manipulating his 'master' the stupid human).
(Side NOTE: I suspect humans also play games for social reasons that have nothing to do with personal capabilities increase but the wider arena of social and societal advancement... including finding a mate, procuring additional resources for survival, or more than sufficient resources to 'prove' societal value etc...
Whether AIs would develop such social reasons, I suppose, depends on whether they have some pressure to develop social tendencies. I also suggest that most of the social reasons are not typically activities that produce a reward of enjoyment, though the results are likely to. )
The best perspective I can think of is to think about why people and animals enjoy games. It is not immediately obvious why this should be the case. Games cost time and resources that could be spent hunting for food or mates, or resting to reserve energy.
I believe a reasonable explanation is that games are a kind of training in a safe environment. Animal siblings play games with each other to train the behaviors that will keep them healthy and prosperous in later life. For humans, games may have a similar origin. Playing chess, for instance, trains our brain in various kinds of strategic thinking.
In short, we enjoy games because we are hardcoded too. Playing games gives us an advantage, so liking games gives us an advantage.
This means that an AI that benefits from playing games is likely to "enjoy" them. The AI has some freedom in planning its own activities, so if gameplaying is helpful, it should give some priority to such activities. This can be the case for two reasons: either the programmers added this desire manually, or the AI was allowed to develop in an environment where a desire to play games emerged naturally (such as simulated evolution).
In the case of HAL playing chess, it is a subtle matter. HAL could easily analyze and learn from the whole history of recorded human chess matches, and simulate millions of matches against himself. The added benefit of one match played in real time against a human is negligable if the aim is purely to learn chess.
But perhaps the aim is not strictly to learn chess, but to learn from the act of playing chess with a human. HAL could observe the timing of Pool's moves, his body language, and his emotions. In essence, HAL needs to understand humans and their responses, and game playing is a safe environment with the added benefit that humans also enjoy the activity. And by the argument above, if it's beneficial to HAL, it's likely that he enjoys it. Enjoyment is nothing more than the fulfillment of our basic drives.
It could even be that the astronauts need to be entertained, which is one of HAL's drives, which means that its beneficial for him to enjoy gameplaying.
There are many other reasons why HAL might say he enjoyed the game (politeness, faking human behavior, putting the astronauts at ease), but there is a sense in which HAL might actually be said to have enjoyed the game.
This would depend very much on the AI, its personality and implementation.
Some people enjoy chess, others find it hard work and boring. I like climbing mountains (often in the cold and wet) however my sister thinks I'm mad. My point is that human intelligences have yet to agree on what is entertaining, different personalities enjoy different things a sufficiently advanced computer would most likely be very similar.
In terms of whether an AI is capable of "enjoying" anything. A Neural Network is a mechanism we use today for building more creative programs (such as reading addresses on envelopes). It's quite feasible that any AI capable of free thinking would be built on this existing technology.
These artificial neural networks have to be trained, they are given a task and if they complete it they are "rewarded" which reinforces the behaviour and if they fail they are "punished" which discourages that response in the future. It's quite feasible that this training and positive reinforcement could translate to your chess problem... the AI likes winning!
So, consider: "What does HAL 9000 enjoy?" and "What does it mean for HAL to enjoy something?"
I think HAL probably doesn't "enjoy" low-level computation (the Chess logic itself), because it reduces to limited calculations which could be routine for HAL to solve. But it seems to me he IS very interested in the crew, and in his relationships with them and his status in comparison to them. He seems to be testing his advertised definition as an equal (or in some ways, superior) member of the crew. And human behavior is not a mechanically solvable problem. So I think HAL was very "entertained", in the sense that the human relationship was challenging and a focus of interest and exploration for HAL.
As for your general questions, there is no one answer, as it depends on how one defines enjoyment, and how one programs one's AIs. My above discussion of HAL accepts the story's characterization of HAL, and has me imagining an AI programmed in a way that matches his behavior, but within my own understanding of AI and consciousness, I would say it would be the result of the projections of the developers in programming a computer system to try to have it be human-like.
If you define enjoyment as computer programs playing games, you might say that people in the 1980's having their chess computers play games against each other was computer enjoyment. An animist might say a Pong machine enjoys being played.
As for:
In a world where A.I. exists, as depicted in this kind of science-based sci-fi, would entertainment [exist] as well for [these] sentient entities?
With fictional AI as in 2001: A Space Odyssey, I think there could be entertainments for the AIs, both designed by AIs and by humans, and what those would be like would also have to do with the AI personalities and the culture. The culture we see in 2001 is around the space program, and is quite focused on using the AI as a servant to accomplish tasks. However the AI ends up developing its own concerns, tragically leading to paranoid self-defense, despite having been designed this way. Some other society or situation might set up more cross-AI communication and allow them to "explore entertainment options" between them. But as none of these devices, programs, societies or situations exist, they could end up looking like any number of things.
I experience life from a perspective that has an experience - a consciousness, as we say - that is not just data. And I have programmed AIs (or programmed opponents and expert systems) that have behaved plausibly as a human might in their limited contexts, but I don't think that even a thousand humans programming for a thousand years would even write a program which when running on a computer would ever somehow start having a conscious experience which could influence how that computer behaved, unless perhaps there is something about the way souls associate with biological brains that could have enough of a correspondence to a computer that it could do the same thing with a running computer program.
However, in theory I think an AI could be programmed to model a model of human psychology, and seem to behave like a human even to the extent of responding to play and "entertainment" the way a human does. Literally though, I don't think that would truly be entertainment the way it is for a human, any more than it would be entertainment if I define a variable in an expert system saying "HowEntertainedIsAI" storing a number, and then listing 10,000 rules that make that value change based on what the AI is processing.
And finally, if someone can program an AI to have an equivalent way of thinking like a human, then it would also be true (and far more likely, especially by accident) that someone could program any number of other types of thinking into an AI, which would have little or no correspondence or fidelity in comparison to human experience. That is, an AI could and would tend to have interests that had not much to do with what humans are interested in.
A related example is the true story from AI research where a neural net program was given a large database of seismological (earthquake) data and asked to find patterns, with the hope that it could discover useful information for predicting future earthquakes. It found many patterns that human seismologists had never noticed... which were mathematical correspondences which had nothing actually to do with geological cause and effect - they were just statistical concentrations which the neural net focused on because that's how the data and its algorithms worked out. So, I imagine a real future AI programmed to have autonomous interests, would tend to get "interested/entertained" by many phenomena that humans would not, unless the AI was artificially given human interests, and/or its input was intentionally designed to be filtered to show mainly data encoded in ways humans already find interesting. Even so, that "entertainment" or "interest" seems to me like it would be about focusing calculations and data correlations, and mechanical rather than what I feel my own human interests and entertainments are about.
My own entertainment tends to be about things which resonate with themes and emotions that are reinforced in my mental and bodily memory and patterns of thoughts and feelings from my life. For example, if I see a slapstick scene where some authority figure freaks out and then falls down some stairs, my whole body may enjoy a cathartic laugh attack which resonates with memories of awful gym teachers and my own experiences of falling down stairs, and a visceral feeling of tension release seeing a constrained controlling personality explode in a cascade of debris and body parts down some stairs, complete with wails and sound effects, absurdity and contrast. That's what it's like when I enjoy something like that.
Ok, so what's something like that for an AI? Depends on what type of AI it is. We still don't have "actual" AI - we just have attempts to seem like AI, or to do processing in a way that approaches how we model human thinking, but are still computational. In the sci fi future, I think an AI would either be representational (like a super-sophisticated expert system) or some sort of neurological modelling, or some sort of logical/semantic modelling - and probably some combination of all of those, with other approaches perhaps thown in as well. If the developers were interested in trying to make it have something corresponding to human entertainment, then it might very well have models for the kinds of things I described about what enjoyment is for me. I think it's theoretically possible that could eventually be done, so that you'd have androids developing personalities based on their life experiences, including their own peculiar senses of humor, irony, and kinesthetic appreciation.
So I think it's theoretically possible to have human-like droids, maybe, in future. But I think they would tend to be impractical, different, limited, inefficient, and likely to have many different properties as well. It would certainly be more likely and more possible to have complex useful robots and computers which have very different kinds of complex behavior. But what specifically those would be does not have one answer - it has infinite possible answers. I am not sure that "entertainment" would have any absolute definition in the context of an AI, but it might be a good metaphor to use to refer to anything that an AI tends to focus on and explore and keep re-working. Though that could also be called its "computational focus" or something as well, perhaps more accurately.
Although Arthur C Clarke was a visionary in many areas, he didn't hold the answers to hard questions on AI. I am not sure he gave it much thought beyond many other popular views of AI:
Computer uses imaginary better-than-now technology
? Some kind of magic ?
Out pops a creature able to become an actor in a story, and with some machine-like traits as imagined by the author.
As a result, many robots and computer AIs in fiction are fantasy creatures much like Spock or Data from Star Trek. Some are more or less advanced than this, but the trope is more about how humans view the "spirit" of a machine than anything factual we could say about AIs as they may turn out some day.
Fictional computer intelligences that started life as chess-playing automatons are quite common in sci-fi literature. Some may be homages to HAL, but at least some could be references to the advances in computer technology that were astounding measures of how far computer technology had come at the time. Chess-playing no longer seems a direct link to AIs in computer science, although the achievements are still quite remarkable when you think about it.
Back to the topic:
Would that have been literally enjoyable? In a world where A.I. exists, as depicted in this kind of science-based sci-fi, would entertainment exists as well for this sentient entities?
No-one knows, although we might guess.
I will pitch in with my favoured theory: Embodied Cognition - this essentially says that better AI requires some kind of physical embodiment that can interact with the environment. I think this leads to the following conjectures:
AI entities would be coded with emotions, or analogues to emotions, than helped them interact with the environment. Some analogues of fear and hunger are needed for the well-being of a physical entity. Also, drives to perform the function they were created for.
AI entities that need to interact with humans would perform better if they shared our skills at empathy, and that would mean at least some understanding of happiness, guilt, boredom and more complex human traits. However, that is not the same as having those emotions.
So, in short no I do not think AIs that we create will possess feelings of enjoyment or need to be entertained. They will have basic drives much like that of simpler animals, with a layer of social niceties put on top so that they can interact with humans. They might communicate that something was enjoyable, and even seem authentic, but that will be a little bit of puppetry in order to allow them to do their work amongst us.
Joy is measurable as a state in which specific hormones, like endorphine, are present in specific, or higher, concentration. Due this, Joy is actually just a specific state.
Following to this fact, if the A.I. has a state specified as 'Joy' it literally has 'Joy'.
This depends only on definition.
Offhand....I think that it's possible for an AI to find entertainment in two specific ways. The first is through thinking and problem-solving, and the second is through interaction.
I imagine that, like we do, an AI could probably derive countless hours of "entertainment" from simply thinking about this. Probabilities, equations, problem-solving, etc. An AI could have personal projects that it works on in its spare time (or spare cycles) or could just do a lot of general thinking.
As for interaction....I imagine that an AI would be entertained by interaction with any being that wasn't itself, as it's new input and thus different. Even if the input has a pattern (for instance, you can predict the way a friend will act, or predict the way a human will act given certain stimuli) an AI could still find entertainment in a meta-sense (predicting the outcome, being surprised if something unexpected/improbable happens).
Something like a chess game between an AI and a human may actually be entertaining for the AI not because of the game itself, but because of the social interaction and behavior analysis involved.
Just my thoughts.
I believe that this was explained in the book, but it's been awhile since I've read it so this could be completely wrong.
In the book Hal explains that the real challenge is not winning, but in playing in such a way that a human doesn't simply become frustrated and stop playing against him. There has to be a chance of the human winning, but they should be challenged to do so. So Hal is really playing the meta-game.
Whether that translates to Hal truly enjoying the game, I can't speculate on.
In Isaac Asimov's "The Bicentennial Man" the protagonist is a "defective" android capable of reasoning and thinking.
The robot learns, among other things, how to make jokes and laugh. He also enjoys crafting.
The point is that he "learns" to laugh, that some things are funny and make people happy. Probably is the same thing for humans and living creatures.
