[News] Soccer AI FAILS and mixes up ball and referee's bald head.
3:48

[News] Soccer AI FAILS and mixes up ball and referee's bald head.

Yannic Kilcher 15.11.2020 8 834 просмотров 339 лайков

Machine-readable: Markdown · JSON API · Site index

Поделиться Telegram VK Бот
Транскрипт Скачать .md
Анализ с AI
Описание видео
#ai #tech #news This soccer camera is operated by an AI to track the ball. However, the AI has an interesting failure mode and repeatedly mixes up the ball with the bald head of a referee. This raises some interesting questions about the role of ethics in AI research. Footage from SPFL Championship : ICTFC 1 v 1 AYR : 24/10/2020 Links: YouTube: https://www.youtube.com/c/yannickilcher Twitter: https://twitter.com/ykilcher Discord: https://discord.gg/4H8xxDF BitChute: https://www.bitchute.com/channel/yannic-kilcher Minds: https://www.minds.com/ykilcher Parler: https://parler.com/profile/YannicKilcher LinkedIn: https://www.linkedin.com/in/yannic-kilcher-488534136/ If you want to support me, the best thing to do is to share out the content :) If you want to support me financially (completely optional and voluntary, but a lot of people have asked for this): SubscribeStar: https://www.subscribestar.com/yannickilcher Patreon: https://www.patreon.com/yannickilcher Bitcoin (BTC): bc1q49lsw3q325tr58ygf8sudx2dqfguclvngvy2cq Ethereum (ETH): 0x7ad3513E3B8f66799f507Aa7874b1B0eBC7F85e2 Litecoin (LTC): LQW2TRyKYetVC8WjFkhpPhtpbDM4Vw7r9m Monero (XMR): 4ACL8AGrEo5hAir8A9CeVrW8pEauWvnp1WnSDZxW7tziCDLhZAGsgzhRQABDnFy8yuM9fWJDviJPHKRjV4FWt19CJZN9D4n

Оглавление (1 сегментов)

Segment 1 (00:00 - 03:00)

so there is this recording of the soccer match which is quite interesting because the camera of the match is ai controlled which just means that it's programmed to track the ball now it tracks the ball by visual features and what's funny about this particular one is that the ai switches constantly between the ball and the bald head of one of the referees which if you look at it looks exactly alike especially in low resolution at which i guess the camera would operate on yeah if you haven't seen it go look at it it's quite funny but it highlights a more interesting point technology fails now this particular system it's probably not very much ai it's not very smart i can guess that it's very standard kind of feature extractor maybe something like a huff transform with a few sifter surf features here and there to look at the color things and kind of low level information to track the ball is usually enough and it's probably more robust than deep learning uh let's be honest here but while this instance is funny uh a lot of times when these systems fail they have bad or even catastrophic consequences let's say a self-driving car mixes up a head of a child consequences can be quite grave so i would like to put this to the sort of people who advocate for having things like broader impact statements in papers and saying that the entire ai research process should be filled with considerations of ethics to the end application we all agree that these things can fail but let's take this particular instance right here if this system is trained at all it's probably not trained on too many bald heads and therefore simply mixes up the ball in the bald head because it looks almost the same interestingly enough this is one of the situations where the system disproportion often fails for white men but let's leave that out of the picture for now where in this process exactly should someone step in and say wait this is uh ethically concerning should the inventor of the huff transform i don't know who that was maybe alfred huff pool half say ha you know if my system detects circles and images then obviously the negative consequences could be that it mixes up a head with a ball interestingly enough the wikipedia page of the circle huff transform says that it can be used to detect people's heads uh i just thought that was funny where in the process except at the end when someone actually takes the technology and puts it into a camera that person should consider the failure modes knowing what the technology is about to go to the inventor of a circle detector and expect from them to predict kind of these negative outcomes is ludicrous i'm sorry try to write the broader impact statement for the huff transform doubt you would have come up with this failure mode or anything similar to it if it hadn't actually happened and you shouldn't like circle detectors are useful and they sometimes fail and when they fail we'll deal with it after all even with the best broader impact statement this wouldn't have been prevented that was just my two cents go check it out have fun bye-bye

Другие видео автора — Yannic Kilcher

Ctrl+V

Экстракт Знаний в Telegram

Экстракты и дистилляты из лучших YouTube-каналов — сразу после публикации.

Подписаться

Дайджест Экстрактов

Лучшие методички за неделю — каждый понедельник