Papers of the day   All papers

Rethinking the maturity of artificial intelligence in safety-critical settings

Comments

Mar 15 2020 Mark Taflinger

While you are quarantined, watch 60 Minutes tonight. Driverless semi-trucks on American roads with real loads by 2021. #YangGang
42 replies, 1820 likes


Feb 06 2020 Missy Cummings

The entire self-driving community has failed to grasp that they are trying to field deeply flawed, immature systems that cannot cope with uncertainty, when defense companies have gone down this path, such premature deployments ended in failures, see http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf
15 replies, 322 likes


Feb 05 2020 Gary Marcus

A measure of just how hard AI in the real world is https://www.theinformation.com/articles/money-pit-self-driving-cars-16-billion-cash-burn
15 replies, 215 likes


Feb 07 2020 Matt Farah

This is the word. Uncertainty. Driving is dealing with uncertainty, in real time, at 70 mph. A lot of just being ready for "what if." Exactly the thing computers are fucking terrible at.
22 replies, 200 likes


Feb 23 2020 Missy Cummings

@AnimaAnandkumar @GaryMarcus Deep learning and symbolic reasoning both fall very short in replicating the reasoning needed by safety-critical systems w/ embedded AI. What we really need is for *famous* AI academics and practitioners to stop posturing and solve a very real problem http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf
1 replies, 48 likes


Feb 15 2020 Missy Cummings

AVs will only be safer in operational domains low in uncertainty (slow, geo-fenced, additional infrastructure for monitoring & control, no bad weather) The technology - both software & hardware - do not exist to get us to safe full self-driving, see http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf
4 replies, 43 likes


Jan 11 2020 Jarno M. Koponen

On deep learning and reasoning 💎: “The reason you can’t count on deep learning to do inference and abstract reasoning is that it’s not geared toward representing precise factual knowledge in the first place.” -@GaryMarcus, Rebooting AI #ml #ai
5 replies, 41 likes


Mar 15 2020 Missy Cummings

OMG, maybe it's cabin fever but @60Minutes is desperate for ratings in their gushing love of driverless trucks. Seriously, where is their supposed critical insight? NO WAY is this tech going to be ready for nationwide unrestricted release by 2021, see http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf
1 replies, 29 likes


Feb 14 2020 E.W. Niedermeyer

Interesting topic. The truth is probably in the middle ground here: AVs will absolutely be safer than humans (that bar is quite low) but they won't be perfect, and we shouldn't let the assumption that they will be justify bad development practices. https://jalopnik.com/everyone-needs-to-stop-assuming-autonomous-vehicles-are-1841604880
11 replies, 27 likes


Feb 07 2020 Missy Cummings

Let me gently suggest that you carefully read this http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf AGI will NEVER happen until we figure out how to approximate top-down reasoning which DOES NOT EXIST in any fashion in known AI techniques today. Connectionist/deep learning & symbolic AI can't do this.
1 replies, 26 likes


Feb 07 2020 Noel Sharkey

SELF-DRIVING into 16 billion dollar spending and the CARS are still not fit for purpose until billions more spent. We’ve been steered into a hype cycle See who is driving all the spending.
3 replies, 26 likes


Jan 28 2020 Missy Cummings

Love it! Perfect illustration of the insurmountable problems of computer vision that I detail here http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf People who think all camera-vision driverless cars like what @Mobileye proposes is a good idea need to have their heads examined.
1 replies, 21 likes


Jan 12 2020 Missy Cummings

Deep learning nor symbolic reasoning can provide expert-based judgment under uncertainty like that needed for the conflicting traffic sign conundrum below. This is critical for AI embedded in safety-critical systems, but very far from reality. See http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf https://t.co/RtqOHjRtQU
0 replies, 10 likes


Feb 06 2020 Ches | Frank

Traditional driving tests don't apply to AVs because humans are bad at the kinds of tasks machines are good at, & vice/versa. However, @missy_cummings describes a crash resulting from a lack of basic functionality that a driving test would've revealed. 🤔http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf https://t.co/qxIhN5lmOK
2 replies, 9 likes


Mar 18 2020 William Wallace

“It’s baffling that NHTSA is using its limited resources on changing the rules for the vehicles of the future, when there's so much the agency can do—and hasn’t done—to make cars safer today." @CRAdvocacy in @ConsumerReports cc @Ctr4AutoSafety https://www.consumerreports.org/autonomous-driving/us-regulators-propose-first-ever-self-driving-car-rules/
1 replies, 8 likes


Jan 28 2020 Tim Almond

One for @missy_cummings, and anyone who thinks self driving is months away. https://arstechnica.com/cars/2020/01/how-a-300-projector-can-fool-teslas-autopilot/
1 replies, 6 likes


Feb 06 2020 Filip Piekniewski

This article http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf by @missy_cummings is so much needed voice of reason in this AI circus. Such a short and great summary of the elephant in the AI room.
1 replies, 6 likes


Feb 07 2020 Tim Miller

I would double tweet this if there was a double tweet option.
0 replies, 5 likes


Feb 11 2020 Andrew

This is very, very good from @missy_cummings talking about the inherent problems of using AI in safety critical situation, y’know, like driving a car http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf Have you read it @kentindell? As the UK pushes into AVs more we need to stop and talk about this!
1 replies, 4 likes


Mar 18 2020 Missy Cummings

It would be nice for @NHTSAgov to add some important rules like driverless cars should not be tricked by tape on signs, they should be able to work in low sun angle conditions, they should be able to correctly "see" partially obscured signs, see more at http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf
2 replies, 4 likes


Feb 10 2020 Missy Cummings

@willknight @geoffreyhinton what's especially funny is that both approaches fall very short - neither deep learning proponents nor symbolists can get us across the finish line, see http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf
1 replies, 2 likes


Feb 15 2020 HotComputerScience

Most popular computer science paper of the day: "Rethinking the maturity of artificial intelligence in safety-critical settings" https://hotcomputerscience.com/paper/rethinking-the-maturity-of-artificial-intelligence-in-safety-critical-settings https://twitter.com/missy_cummings/status/1225423463550046208
1 replies, 2 likes


Feb 14 2020 Arthur Holland Michel

Nobody cuts through the AI hype as incisively and knowledgeably as @missy_cummings. http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf
0 replies, 2 likes


Feb 07 2020 Mark Cannon

@missy_cummings Only a developed AGI will solve this issue. Unfortunately none of the big investment is in true GI.
2 replies, 2 likes


Feb 08 2020 Dr. S.A. Applin

See Applin and Fischer 2012: http://posr.org/w/images/a/a5/Applin_Fischer_AutoUI_2012_DRAFT.pdf 2015: https://ieeexplore.ieee.org/document/7064858?arnumber=7064858&filter%3DAND(p_IS_Number:7064851)= with Riener 2015: https://ieeexplore.ieee.org/document/7310907?arnumber=7310907 2017: https://ieeexplore.ieee.org/document/7948873 draft: http://posr.org/w/images/c/cb/Applin_S_AV_Ethics.pdf
0 replies, 1 likes


Feb 07 2020 KillingMyCareer

Amen! From Waymo to Tesla, autonomous accidents are underreported & hidden abusively behind false claims of proprietary rights. Consumers have to pass a drivers test, self-driving cars do not! #TheSociopathicBusinessModel #FraudFormula
1 replies, 1 likes


Feb 07 2020 Marci Baranski

Why we should be skeptical of AI claims coming from Silicon Valley... more hype than bite. Complex systems are complex, who knew?
0 replies, 1 likes


Feb 06 2020 Liza Dixon ⚡️

🚨 Required reading for self-driving Twitter 🚨 "Rethinking the maturity of artificial intelligence in safety-critical settings" by @missy_cummings 💯 I will add that TRL 6 & TRL7 are prime levels for #autonowashing 🙃 http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf https://t.co/RLddqDTRqh
0 replies, 1 likes


Mar 16 2020 Missy Cummings

@pfmgmt @60Minutes http://hal.pratt.duke.edu/sites/hal.pratt.duke.edu/files/u39/2020-min.pdf
0 replies, 1 likes


Content

PDF content of a computer science paper: Rethinking the maturity of artificial intelligence in safety-critical settings"