Nine out of 10 car accidents are due to driver error. That’s why many people in Virginia and across the U.S. are hoping self-driving cars will eventually prevent these errors from arising. However, the Insurance Institute for Highway Safety says that the ability of self-driving cars to prevent crashes is being overestimated.

After analyzing more than 5,000 crashes involving driver error, researchers divided the types of error into five categories. They concluded that self-driving cars, in their current state, would only be able to prevent errors in two of those categories. Specifically, these cars could prevent sensing and perceiving errors and errors stemming from incapacitation. The former were behind 24% of the crashes analyzed, and the latter were behind 10%.

Sensing and perceiving errors include distracted operation and driving in low-visibility conditions. Incapacitation refers to drug, alcohol and medication impairment as well as falling asleep behind the wheel. The other three categories were predicting errors, planning and deciding errors and execution and performance errors.

Safety advocates say automakers need to program self-driving cars in such a way that they prioritize safety over speed or a driver’s preferences: in other words, not drive too much like people. They must be able to adapt to road conditions and react to unforeseeable actions from other drivers as well as pedestrians.

Before such vehicles come out, though, drivers are expected to maintain control of their cars at all times. If the failure to do this leads to an accident, victims may, in certain circumstances, pursue a personal injury case. Virginia follows the rule of pure contributory negligence, which means that any amount of contributory negligence bars a plaintiff from recovering damages. It’s a strict rule, so a victim may want a lawyer to assess the case before filing the claim and seeking a settlement.