Yes, but this is what I don't get: don't software bugs in humans drivers' brains kill all the time? My cousin and his wife were recently hit by a human-controlled vehicle*. They were crossing in a crossing at a designated crossing legally with the "green man" and were hit by a turning vehicle. The driver, a granny picking a kid from school and driving into the sun, treated them as artifacts or was suffering sensor failure. All software contains bugs. Software that is going to make judgement call on an array of complex data is absolutely going to get it wrong at times, just like people do. There's no absolute fix for this. The question for me is not whether AI cars are perfect. There will be failures and we can reasonably demand that a cause will be identifiable. My question is this: Would my cousin and wife be better off if the granny was driving or an AI system?
Jim * Both survived, a broken bone and some soft tissue damage. Near full recovery expected over time. On 8 May 2018 at 09:41, Roger Clarke <[email protected]> wrote: > Report: Software bug led to death in Uber's self-driving crash > Sensors detected Elaine Herzberg, but software reportedly decided to > ignore her. > TIMOTHY B. LEE > Ars Technica > MAY 7, 2018 10:12 PM UTC > https://arstechnica.com/tech-policy/2018/05/report- > software-bug-led-to-death-in-ubers-self-driving-crash/ > > The fatal crash that killed pedestrian Elaine Herzberg in Tempe, Arizona, > in March occurred because of a software bug in Uber's self-driving car > technology, The Information's Amir Efrati reported on Monday. According to > two anonymous sources who talked to Efrati, Uber's sensors did, in fact, > detect Herzberg as she crossed the street with her bicycle. Unfortunately, > the software classified her as a "false positive" and decided it didn't > need to stop for her. > > ... [nice article] ... > > > [In the postmodern world, a new survival trait has emerged: > > [Don't exhibit patterns that stand out from the crowd and are hard for > AI/ML to classify, because the patterns will either mark you for attention, > e.g. by law enforcement and national security agencies, or will be treated > as an artefact and ignored. Both kinds of positives are survival threats. > > [Elaine Herzberg's unclassifiability was presumably a walking human form > merged with a bike-profile, and the treatment by the software melange was > 'unclassifiable, hence treat as an artefact of the image collection > system'; so she wasn't detected as a pedestrian in the vehicle carriageway. > > > -- > Roger Clarke http://www.rogerclarke.com/ > > Xamax Consultancy Pty Ltd 78 Sidaway St, Chapman ACT 2611 AUSTRALIA > Tel: +61 2 6288 6916 http://about.me/roger.clarke > mailto:[email protected] http://www.xamax.com.au/ > > Visiting Professor in the Faculty of Law University of N.S.W. > Visiting Professor in Computer Science Australian National University > _______________________________________________ > Link mailing list > [email protected] > http://mailman.anu.edu.au/mailman/listinfo/link > _______________________________________________ Link mailing list [email protected] http://mailman.anu.edu.au/mailman/listinfo/link
