Eighteenth-century rural England was a place where people starved each spring as the winter stores ran out, where in bad years and poor districts long hours of agricultural labour—if it could be got—barely paid enough to keep body and soul together, and a place where the “putting-out” system of textile manufacture at home drove workers harder for lower pay than even the factories would. (Ask Zambians today why they take ill-paid jobs in Chinese-managed mines, or Vietnamese why they sew shirts in multinational-owned factories.) The industrial revolution caused a population explosion because it enabled more babies to survive—malnourished, perhaps, but at least alive.
Returning to hunter-gatherers, Mr LeBlanc argues (in his book “Constant Battles”) that all was not well in ecological terms, either. Homo sapiens wrought havoc on many ecosystems as Homo erectus had not. There is no longer much doubt that people were the cause of the extinction of the megafauna in North America 11,000 years ago and Australia 30,000 years before that. The mammoths and giant kangaroos never stood a chance against co-ordinated ambush with stone-tipped spears and relentless pursuit by endurance runners.
Does it stillburn like it did that wonderful night at the police station?"Beth could not believe the depths to which Christine would go to make herlife a living hell.
All Beth wanted to do was to attack, to claw, to bite, tosink her teeth into the arrogant egotistical flesh of this traitorous femmefatale until she hit bone.
Since Beth had to keep her eyeslooking down, the mark of a properly domesticated slave, she only saw thewoman's fashionable leather pumps and part of her long flowing gown.
A number of researchers are currently investigating applications ofTD() to other games such as chess and Go. SebastianThrun has obtained encouraging preliminary results with a TD-chesslearning system that learns by playing against a publicly availablechess program, Gnuchess (Thrun, personal communication).Schraudolph, et al. have also obtained encouraging early resultsusing TD() to learn to play Go .Finally, Jean-FrancoisIsabelle has obtained good results applying the TD self-learningprocedure to Othello . The best network reportedin that studywas able to defeat convincingly an "intermediate-advanced"conventional Othello program.
In particular, when the input variables encode the raw boardinformation such as blots and points at particular locations, alinear function of those variables would express simple conceptssuch as "blots are bad" and "points are good." Such concepts aresaid to be context-insensitive, in that the evaluation functionassigns a constant value to a particular feature, regardless of thecontext of the other features. For example, a constant value wouldbe assigned to owning the 7 point, independent of the configurationof the rest of the board. On the other hand, an example of acontext-sensitive concept that emerges later in learning is thenotion that the value of the 7 point depends on where one's othercheckers are located. Early in the game, when one has severalcheckers to be brought home, the 7 point is valuable as a blockingpoint and as a landing spot. On the other hand, if one is bearingin and all other checkers have been brought home, the 7 point thenbecomes a liability.
What kind of obligations are relevant when we wish to assess whether abelief, rather than an action, is justified or unjustified?Whereas when we evaluate an action, we are interested in assessing theaction from either a moral or a prudential point of view, when it comesto beliefs, what matters is the pursuit of truth. Therelevant kinds of obligations, then, are those that arise when we aimat having true beliefs. Exactly what, though, must we do in thepursuit of this aim? According to one answer, the one favored byevidentialists, we ought to believe in accord with our evidence. Forthis answer to be helpful, we need an account of what our evidenceconsists of. According to another answer, we ought to follow thecorrect epistemic norms. If this answer is going to help us figure outwhat obligations the truth-aim imposes on us, we need to be given anaccount of what the correct epistemic norms are.
Finally, there are also a number of potential applications of TDlearning outside the domain of games, for example, in areas such asrobot motor control and financial trading strategies. For thesesorts of applications, one may lose the important advantage one hasin games of being able to simulate the entire environment. Inlearning to play a game, it is not necessary to deal with thephysical constraints of a real board and pieces, as it is mucheasier and faster to simulate them. Furthermore, the "action" ofthe learning agent is just a selection of which state to move tonext, from the list of legal moves. It is not necessary to learna potentially complex "forward model" mapping control actions tostates. Finally, within the simulated environment, it is possible togenerate on-line a potentially unlimited amount of trainingexperience, whereas in, for example, a financial market one might belimited to a fixed amount of historical data. Such factors implythat the best way to make progress initially would be to studyapplications where most or all of the environment can be effectivelysimulated, e.g., in certain robotic navigation and path planningtasks, in which the complexity lies in planning the path, rather thanin mapping the motor commands to resulting motion in physical space.
We assume thereis a "random surfer" who is given a web page at random and keeps clickingon links, never hitting "back" but eventually gets bored and starts onanother random page.
First make up a thesis outline: several pages containing chapter headings, sub-headings, some figure titles (to indicate which results go where) and perhaps some other notes and comments.