Probabilistic models are more and more pervasive in computer science. On the other hand, allowing algorithms to evolve according to the laws of probability theory rather than deterministically has been considered since the early days of theoretical computer science. In recent years, all this has attracted the attention of researchers in programming language theory and semantics, thanks to possible applications in machine learning and cryptography. Allowing programs to sample from (possibly continuous) distributions or to condition on the value of certain concrete observations pose a challenge to programming language semantics, in particular in presence of higher-order functions. Moreover, randomized computation allows for a fresh look to some old research problems in the theory of lambda-calculus. We will give an introductory talk about all that.