Autonomous Autos – Cruise Control pt. II

Cruise gets real.

• Cruise has doubled down on its commitment to safety by effectively putting the whole company in a holding pattern as well as the secondment of the Head of Legal from GM to oversee the whole process.
• This is yet another sign of some of the difficulties that technology companies are continuing to have when entering an industry where the device can kill its users if it goes wrong.
• The incident involving a hit-and-run driver (see here) was an accident waiting to happen and it was such an unlikely event, that there is no way that Cruise could have reasonably been able to train its vehicle to deal with it.
• This problem is not going to go away as the road is a random haphazard place where so many miles are driven every year that the most unlikely and bizarre events occur often enough to create a problem.
• The other issue is that machines are held to a much higher standard than humans.
• In the USA in 2022, human error while driving killed more than 40,000 humans which equates to roughly one death for every 75m miles driven.
• However, the machines will not be allowed to drive vehicles until they can reliably demonstrate that they will reliably kill almost no one which will take a long time to achieve and even longer to prove to the regulators.
• Cruise’s problems have been further compounded by the fact that it seems to have made a mess of its communications meaning that the regulator believes that Cruise deliberately misled it by not making certain videos of the incident available.
• I think that this is a pretty simple misunderstanding, but when the stakes are this high (i.e. Cruise as a going concern), concerted action needs to be taken.
• This is why GM has sent its head of legal to be the CAO (Chief Administration Officer) of Cruise while this mess is sorted out.
• The net result is that this event is likely to be the first of several where a machine ends up injuring a human and it brings into sharp focus the problem of causality.
• The machine vision systems all use deep learning to turn the signals they receive from the sensors into a scene upon which the system can depend to drive the vehicle.
• The problem is that while the machines can create the scenes, they don’t understand them which is what prevents them from being able to affect the reasoning that is required to deal with an event that has not been explicitly taught.
• I have long argued that it is not until this problem is solved or largely mitigated that the machines are going to get close enough to the safety requirement that will allow them to be deployed commercially.
• This day looks increasingly far off and the problems that Cruise is having are very unlikely to be limited to just Cruise and so I suspect that everyone in the USA and Europe who works in this space will be considering how to make the necessary adjustments.
• Hence, I don’t think that autonomous driving is going anywhere beyond motorway driving with an alert driver for some time.
• I am sticking to my long-held target of 2028 for commercial deployment although the risk of this being too soon is growing with every passing incident.
• Consequently, falling valuations, shutdowns and dilutive consolidation look like they will continue for some time to come.
• I have no real desire to get involved here beyond the lidar companies, many of which have plenty of use cases outside of autonomous driving.