Autism Spectrum Disorder (ASD) is a neurodevelopmental disorder characterized mainly by deficits in social communication and stereotyped and restricted behavior and interests with a male to female bias of 4.2/1. Social behavior in ASD animal models is commonly analyzed in males, and seldomly in females, using the widely implemented three-chambers test procedure. Here, we implemented a novel procedure, the Live Mouse Tracker (LMT), combining artificial intelligence, machine learning procedures and behavioral measures. We used it on mice that were prenatally exposed to valproic acid (VPA) (450 mg/kg) at embryonic day 12.5, a widely recognized and potent ASD model that we had previously extensively characterized. We focused on female mice offspring, in which social deficits have been rarely documented when using the 3-CT procedure. We recorded several parameters related to social behavior in these mice, continuously for three days in groups of four female mice. Comparisons were made on groups of 4 female mice with the same treatment (4 saline or 4 VPA) or with different treatments (3 saline and 1 VPA). We report that VPA females show several types of social deficits, which are different in nature and magnitude in relation with time. When VPA mice were placed in the LMT alongside saline mice, their social deficits showed significant improvement as early as 1 h from the start of the experiment, lasting up to 3 days throughout the duration of the experiment. Our findings suggest that ASD may be underdiagnosed in females. They also imply that ASD-related social deficits can be ameliorated by the presence of typical individuals.