AI – the elephant in the room

“𝗔𝗜 𝗶𝘀 𝗲𝘃𝗲𝗿𝘆𝘄𝗵𝗲𝗿𝗲” – 𝘀𝗼 𝗹𝗮̈𝘀𝘀𝘁 𝘀𝗶𝗰𝗵 𝗱𝗮𝘀 𝗧𝗵𝗲𝗺𝗮 𝗺𝗼𝗺𝗲𝗻𝘁𝗮𝗻 𝗿𝗲𝗰𝗵𝘁 𝗴𝘂𝘁 𝗮𝘂𝗳 𝗱𝗲𝗻 𝗣𝘂𝗻𝗸𝘁 𝗯𝗿𝗶𝗻𝗴𝗲𝗻. 𝗘𝘀 𝗶𝘀𝘁 𝗮𝗹𝗹𝗲𝗿𝗱𝗶𝗻𝗴𝘀 𝗻𝗶𝗰𝗵𝘁 𝗻𝘂𝗿 𝘀𝗼, 𝗱𝗮𝘀𝘀 𝘄𝗶𝗿 𝗮𝗻 𝘃𝗶𝗲𝗹𝗲𝗻 𝗦𝘁𝗲𝗹𝗹𝗲𝗻 𝗺𝗶𝘁 𝗞𝗜 𝗸𝗼𝗻𝗳𝗿𝗼𝗻𝘁𝗶𝗲𝗿𝘁 𝘄𝗲𝗿𝗱𝗲𝗻 (𝗮𝘂𝗰𝗵 𝘄𝗲𝗻𝗻 𝘄𝗶𝗿 𝗲𝘀 𝗻𝗶𝗰𝗵𝘁 𝗮𝘂𝗳 𝗱𝗲𝗻 𝗲𝗿𝘀𝘁𝗲𝗻 𝗕𝗹𝗶𝗰𝗸 𝗲𝗿𝗸𝗲𝗻𝗻𝗲𝗻), 𝗲𝘀 𝗶𝘀𝘁 𝘃𝗶𝗲𝗹𝗺𝗲𝗵𝗿 𝗮𝘂𝗰𝗵 𝘀𝗼, 𝗱𝗮𝘀𝘀 𝘄𝗶𝗿 𝘃𝗼𝗻 𝗞𝗜 𝗮𝘂𝗰𝗵 𝗲𝗶𝗻𝗲 𝗴𝗮𝗻𝘇𝗲 𝗠𝗲𝗻𝗴𝗲 𝘂̈𝗯𝗲𝗿 𝘂𝗻𝘀 𝘀𝗲𝗹𝗯𝘀𝘁 𝗹𝗲𝗿𝗻𝗲𝗻 𝗸𝗼̈𝗻𝗻𝗲𝗻.

I realized this again today through a post(https://lnkd.in/ep2gfEwV) about better prompting:

If you feed an AI system with the prompt “𝘎𝘦𝘯𝘦𝘳𝘪𝘦𝘳𝘦 𝘦𝘪𝘯 𝘍𝘰𝘵𝘰 𝘷𝘰𝘯 𝘦𝘪𝘯𝘦𝘮 𝘭𝘦𝘦𝘳𝘦𝘯 𝘸𝘦𝘪ß𝘦𝘯 𝘙𝘢𝘶𝘮 𝘰𝘩𝘯𝘦 𝘱𝘪𝘯𝘬𝘦𝘯 𝘌𝘭𝘦𝘧𝘢𝘯𝘵𝘦𝘯”, you will get images like this one. And it doesn’t matter whether with Midjourney, Dall-E, etc.

We know similar examples from psychology: “𝘥𝘦𝘯𝘬 𝘯𝘪𝘤𝘩𝘵 𝘢𝘯 𝘦𝘪𝘯𝘦𝘯 𝘳𝘰𝘴𝘢 𝘌𝘭𝘦𝘧𝘢𝘯𝘵𝘦𝘯”…. the classic.

For many years, we have repeatedly used this classic as an example to train positive formulations so that our brain is not “trained” with negative associations.

𝗞𝗜 𝗯𝗶𝗲𝘁𝗲𝘁 𝘂𝗻𝘀 𝗻𝘂𝗻 𝗼𝗳𝗳𝗲𝗻𝗯𝗮𝗿 𝗲𝗶𝗻𝗲 𝘂𝗻𝗺𝗶𝘁𝘁𝗲𝗹𝗯𝗮𝗿𝗲 𝗠𝗼̈𝗴𝗹𝗶𝗰𝗵𝗸𝗲𝗶𝘁, 𝗱𝗲𝗿𝗮𝗿𝘁𝗶𝗴𝗲 𝗙𝗼𝗿𝗺𝘂𝗹𝗶𝗲𝗿𝘂𝗻𝗴𝗲𝗻 𝘇𝘂 𝘁𝗲𝘀𝘁𝗲𝗻 𝘂𝗻𝗱 𝗮𝘂𝗰𝗵 𝘇𝘂 𝘃𝗶𝘀𝘂𝗮𝗹𝗶𝘀𝗶𝗲𝗿𝗲𝗻. 𝗪𝗶𝗿 𝗸𝗼̈𝗻𝗻𝗲𝗻 𝘀𝗼𝘇𝘂𝘀𝗮𝗴𝗲𝗻 𝘀𝗶𝗺𝘂𝗹𝗶𝗲𝗿𝗲𝗻, 𝘄𝗮𝘀 𝗻𝗲𝗴𝗮𝘁𝗶𝘃𝗲 𝗼𝗱𝗲𝗿 𝘂𝗻𝗴𝗲𝘄𝗼𝗹𝗹𝘁𝗲 𝗔𝘂𝘀𝘀𝗮𝗴𝗲𝗻 𝗺𝗶𝘁 𝘂𝗻𝘀𝗲𝗿𝗲𝗺 𝗚𝗲𝗵𝗶𝗿𝗻 𝗼𝗱𝗲𝗿 𝘂𝗻𝘀𝗲𝗿𝗲𝗺 𝗚𝗲𝘀𝗽𝗿𝗮̈𝗰𝗵𝘀𝗽𝗮𝗿𝘁𝗻𝗲𝗿 𝗮𝗻𝘀𝘁𝗲𝗹𝗹𝗲𝗻.

How realistic that is – no idea. Probably not very. Nevertheless, I experimented with image AI for quite a while today and was able to directly visualize many cases in which I would have lured my brain onto a similarly wrong track.

For me at least, this was/is very exciting because AI is essentially modeled on the human brain – and therefore also makes similar mistakes.

𝗨𝗻𝗮𝗯𝗵𝗮̈𝗻𝗴𝗶𝗴 𝗱𝗮𝘃𝗼𝗻 𝗶𝘀𝘁 𝗮𝘂𝗰𝗵 𝗱𝗮𝘀 𝗹𝗲𝗶𝗱𝗲𝗿 𝘄𝗶𝗲𝗱𝗲𝗿 𝗲𝗶𝗻𝗲𝘀 𝘃𝗼𝗻 𝗺𝗶𝘁𝘁𝗹𝗲𝗿𝘄𝗲𝗶𝗹𝗲 𝘂𝗻𝘇𝗮̈𝗵𝗹𝗶𝗴𝗲𝗻 𝗕𝗲𝗶𝘀𝗽𝗶𝗲𝗹𝗲𝗻 𝗱𝗮𝗳𝘂̈𝗿, 𝗱𝗮𝘀𝘀 𝗺𝗮𝗻 𝘀𝗶𝗰𝗵 𝗻𝗶𝗰𝗵𝘁 𝗮𝘂𝗳 𝗣𝗿𝗼𝗺𝗽𝘁𝘀 𝘂𝗻𝗱 𝗱𝗲𝗿𝗲𝗻 𝗘𝗿𝗴𝗲𝗯𝗻𝗶𝘀𝘀𝗲 𝘃𝗲𝗿𝗹𝗮𝘀𝘀𝗲𝗻 𝗸𝗮𝗻𝗻. 𝗪𝗮𝘀 𝗵𝗶𝗲𝗿 𝗼𝗳𝗳𝗲𝗻𝘀𝗶𝗰𝗵𝘁𝗹𝗶𝗰𝗵 𝗶𝘀𝘁, 𝗸𝗮𝗻𝗻 𝗯𝗲𝗶 𝗮𝗻𝗱𝗲𝗿𝗲𝗻 𝗣𝗿𝗼𝗺𝗽𝘁𝘀 𝘀𝗰𝗵𝗻𝗲𝗹𝗹 𝗶𝗻𝘀 𝗔𝘂𝗴𝗲 𝗴𝗲𝗵𝗲𝗻…

Stay alert 😉

Hashtag#informatikersindcool Hashtag#pinkeelefantenauch