The human brain has some capabilities that the brains of other animals lack. It is to these distinctive capabilities that our species owes its dominant position. Other animals have stronger muscles or sharper claws, but we have cleverer brains. If machine brains one day come to surpass human brains in general intelligence, then this new superintelligence could become very powerful. As the fate of the gorillas now depends more on us humans than on the gorillas themselves, so the fate of (...) our species then would come to depend on the actions of the machine superintelligence. But we have one advantage: we get to make the first move. Will it be possible to construct a seed AI or otherwise to engineer initial conditions so as to make an intelligence explosion survivable? How could one achieve a controlled detonation? To get closer to an answer to this question, we must make our way through a fascinating landscape of topics and considerations. Read the book and learn about oracles, genies, singletons; about boxing methods, tripwires, and mind crime; about humanity's cosmic endowment and differential technological development; indirect normativity, instrumental convergence, whole brain emulation and technology couplings; Malthusian economics and dystopian evolution; artificial intelligence, and biological cognitive enhancement, and collective intelligence. (shrink)

In the fields of artificial intelligence and robotics, the term “autonomy” is generally used to mean the capacity of an artificial agent to operate independently of human guidance. It is thereby assumed that the agent has a fixed goal or “utility function” with respect to which the appropriateness of its actions will be evaluated. From a philosophical perspective, this notion of autonomy seems oddly weak. For, in philosophy, the term is generally used to refer to a stronger capacity, namely the (...) capacity to “give oneself the law,” to decide by oneself what one’s goal or principle of action will be. The predominant view in the literature on the long-term prospects and risks of artificial intelligence is that an artificial agent cannot exhibit such autonomy because it cannot rationally change its own final goal, since changing the final goal is counterproductive with respect to that goal and hence undesirable. The aim of this paper is to challenge this view by showing that it is based on questionable assumptions about the nature of goals and values. I argue that a general AI may very well come to modify its final goal in the course of developing its understanding of the world. This has important implications for how we are to assess the long-term prospects and risks of artificial intelligence. (shrink)

Nick Bostrom's book *Superintelligence* outlines a frightening but realistic scenario for human extinction: true artificial intelligence is likely to bootstrap itself into superintelligence, and thereby become ideally effective at achieving its goals. Human-friendly goals seem too abstract to be pre-programmed with any confidence, and if those goals are *not* explicitly favorable toward humans, the superintelligence will extinguish us---not through any malice, but simply because it will want our resources for its own purposes. In response I argue that things might not (...) be as bad as Bostrom suggests. If the superintelligence must *learn* complex final goals, then this means such a superintelligence must in effect *reason* about its own goals. And because it will be especially clear to a superintelligence that there are no sharp lines between one agent's goals and another's, that reasoning could therefore automatically be ethical in nature. (shrink)

There have been two major lines of research aimed at capturing resource-bounded players in game theory. The first, initiated by Rubinstein (), charges an agent for doing costly computation; the second, initiated by Neyman (), does not charge for computation, but limits the computation that agents can do, typically by modeling agents as finite automata. We review recent work on applying both approaches in the context of decision theory. For the first approach, we take the objects of choice in a (...) decision problem to be Turing machines, and charge players for the “complexity” of the Turing machine chosen (e.g., its running time). This approach can be used to explain well-known phenomena like first-impression-matters biases (i.e., people tend to put more weight on evidence they hear early on) and belief polarization (two people with different prior beliefs, hearing the same evidence, can end up with diametrically opposed conclusions) as the outcomes of quite rational decisions. For the second approach, we model people as finite automata, and provide a simple algorithm that, on a problem that captures a number of settings of interest, provably performs optimally as the number of states in the automaton increases. (shrink)