The Furstenberg recurrence theorem (or equivalently, Szemer\'edi's theorem) can be formulated in the language of von Neumann algebras as follows: given an integer $k \geq 2$, an abelian finite von Neumann algebra $(\M,\tau)$ with an automorphism $\alpha: \M \to \M$, and a non-negative $a \in \M$ with $\tau(a)>0$, one has $\liminf_{N \to \infty} \frac{1}{N} \sum_{n=1}^N \Re \tau(a \alpha^n (a) ... \alpha^{(k-1)n} (a)) > 0$; a subsequent result of Host and Kra shows that this limit exists. In particular, $\Re \tau(a \alpha^n (a) >... \alpha^{(k-1)n} (a)) > 0$ for all $n$ in a set of positive density. From the von Neumann algebra perspective, it is thus natural to ask to what extent these results remain true when the abelian hypothesis is dropped. All three claims hold for $k = 2$, and we show in this paper that all three claims hold for all $k$ when the von Neumann algebra is asymptotically abelian, and that the last two claims hold for $k=3$ when the von Neumann algebra is ergodic. However, we show that the first claim can fail for $k=3$ even with ergodicity, the second claim can fail for $k \geq 4$ even assuming ergodicity, and the third claim can fail for $k=3$ without ergodicity, or $k \geq 5$ and odd assuming ergodicity. The second claim remains open for non-ergodic systems with $k=3$, and the third claim remains open for ergodic systems with $k=4$.