We consider oscillators evolving subject to a periodic driving force that dynamically entangles them, and argue that this gives the linearized evolution around periodic orbits in a general chaotic Hamiltonian dynamical system. We show that the entanglement entropy, after tracing over half of the oscillators, generically asymptotes to linear growth at a rate given by the sum of the positive Lyapunov exponents of the system. These exponents give a classical entropy growth rate, in the sense of Kolmogorov, Sinai and Pesin. We also calculate the dependence of this entropy on linear mixtures of the oscillator Hilbert space factors, to investigate the dependence of the entanglement entropy on the choice of coarse-graining. We find that for almost all choices the asymptotic growth rate is the same.then the typical pure state in H has very close to the maximal amount of entanglement allowed between H A and H B , and this is in turn maximized if dim(H A ) = dim(H B ). We will call such factorization of H into "observable" and "non-observable" physics a coarsegraining of the system. This suggests that if we evolve a system randomly from an initial configuration with zero entanglement entropy, then it will eventually forget essentially all of the information of the initial state if we only measure observables sensitive to H A . There are many studies of this kind of process in specific systems. The rate at which the entanglement grows towards saturation depends, in general, on the details of these systems, although some general bounds exist [2][3][4][5]. Linear growth in time appears in many systems, for example, studies of decoherence and quantum chaos [6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21], and quenches of extended systems [22][23][24][25].When trying to apply these results in the context of black hole physics, we are usually confronted with two problems. First of all, the Hilbert space H is big. In the gauge/gravity duality [26] the dynamics takes place in an infinite-dimensional Hilbert space: it is the Hilbert space of a relativistic quantum field theory on the conformal boundary.A very naive application of the results of [1] would suggest that typical states have infinite entropy when splitting H in two pieces of the same size, since both are infinite dimensional.However, the notion of splitting along a random factorization has no meaning, because once we have factored into infinite dimensional pieces, we can factorize the pieces again: there is no natural notion of splitting in half. Thus, the question of the entanglement entropy for a typical state is ill-defined without additional structure on the Hilbert space.An example of such a structure is two operators algebras, one for H A the other one for H B . It is natural to do the splitting with respect to a choice of algebras with reasonable properties determined by features of the dynamics. Once that splitting is done, instead of computing the entanglement entropy of the typical state, we can compute the rate of growth of the entanglement entrop...