@mitsunee Haha, I hadn't thought of that but you're absolutely right.
Mailchimp have now abandoned all of its "whimsiness", so I doubt we'll ever really know.
When does the EU AI Act start being enforced? The answer will vary depending on what type of AI we are talking about.
As we are getting closer to the plenary vote on the EU AI Act, provisionally scheduled for April 10-11, there is some confusion as to when the law actually comes into effect.
After the European Parliament vote, and assuming it goes through as expected, the new legislation will need to be entered into the EUR-Lex Official Journal of the European Union. The law comes into force 20 days after this event.
Even though it comes into force, the legislation first enters into what is called "Vacatio legis". This is a transitional period intended to give stakeholders time to adapt and make any necessary changes to their tools and operations before legislation is applied.
The deadline for Vacatio legis varies for different applications:
⊠The parts concerning prohibited AI systems apply after 6 months. I would expect as early as December.
⊠Specific obligations for "general-purpose" systems come into effect after 12 months. Therefore, around mid-2025 is something that "foundation models" and "AI assistants" likely need to consider.
⊠In essence, the rest of the act comes into effect after 24 months, presumably around mid-2026 - this is what most people have in mind when they say that the AI Act becomes operational after 2 years.
⊠However, there are also details that only come into effect after 36 months, and these are related to the parts in Annex II.
I made this timeline to give you a very quick and brief overview of these next steps before the AI Act comes into full enforcement. References for further reading are provided in my blog post on axbom.com
Bra reflektioner av Fatmir om hur kÀnslokalla rekryteringsprocesser Àr idag.
Det hÀr fick mig ocksÄ att fundera över hur automatiserad rekrytering kan inverka nÀr ordet funktionsnedsÀttning och relaterade begrepp Äterfinns. Vi gÄr allt tydligare mot en vÀrld dÀr mÀnniskor sÀllan kommer granska ansökningar i ett första skede. Maskiner kommer göra det. Exakt hur arbetssökande sÄllas bort av dessa automatiserade verktyg Àr ofta höljt i dunkel.
Enligt Dataskyddsförordningen (GDPR) mÄste företag informera om det sker automatiserad profilering och automatiserat beslutsfattande (som att sortera in arbetssökande i prioritetsordning). Man har ocksÄ som arbetssökande rÀtt att bestrida automatiserad behandling av ens uppgifter.
Hur öppna företag Ă€r med detta nĂ€r de tillĂ€mpar automatisering i rekryteringar Ă€r dock ganska okĂ€nt i nulĂ€get. Om man misstĂ€nker att företag inte Ă€r öppna med sina automatiserade beslut â och man inte fĂ„r tillfredstĂ€llande svar frĂ„n organisationens personuppgiftsansvariga - bör man övervĂ€ga en anmĂ€lan till Integritetsskyddsmyndigheten (IMY).
IMY informerar hÀr om dina rÀttigheter nÀr beslut fattas automatiskt:
https://www.imy.se/privatperson/dataskydd/dina-rattigheter/automatiserade-beslut/
@falcennial đ Off by default would definitely work for me.
I remember when Mailchimp and its monkey mascot had a tone-of-voice that was full of jokes and whimsiness. Many found this annoying or offensive (either generally or after prolonged exposure) and in the end Mailchimp were compelled to implement a setting so that people could switch off the humor.
In a continued demonstration of zero concern for the real human discomfort of customers, they called that setting "party pooper mode".
Anyway, a setting that I would really appreciate is to turn off all the AI assistants that pop up in different services wherein I want to write my own text.
Feel free to call it "AI pooper mode".
(Update: after writing this I realised that "AI pooper" may more accurately describe the setting when it is turned on. This will get confusing. Just let me hide it from sight, please.)
Spread the word to all developers.
https://axbom.com/ai-pooper-mode/
@timthelion Oh yeah! I remember listening to The Missing Cryptoqueen and the part where they described graphs on a web page showing how investments in OneCoin were growing in value, all faked.
In a 2019 report, Sensity â an Amsterdam-based company with AI and deepfake expertise â showed that 96 percent of deepfake videos are sexual and produced without consent. In 99 percent of cases, it is women who are affected. Women being targeted for deepfake films is significantly more common than politicians being targeted.
It is now possible, for a few dollars, to create porn videos with anyone who has a bunch of photos of themselves posted on the Internet. Movies with celebrity faces are available in large quantities for around $5. In chat forums, customized videos are being offered with any person on Instagram with less than 2 million followers, a so-called "personal girl", for 65 dollars.
It is unfortunate that it is called fake because it is a very real abuse; the face belongs to a real person. Desperate young women are regularly alerted to the existence of nude photos and pornographic images of people who look exactly like them, and are named as them. This can of course contribute to immense harmful effects in terms of stress, mental health and self-image. Which in turn can have further tragic consequences.
My blog post from June:
https://axbom.com/deepfake-porn/
#DigitalEthics #DeepFake #AIEthics
Remember when AI was used to generate a George Carlin comedy routine? That didnât happen. Generative AI isnât that good.
After the George Carlinâs estate sued, a representative of the show admitted that it was human-written. The claim that it was produced by an AI trained on Carlinâs material appears to be far from the truth, and rather used as a way to garner attention.
Cory Doctorow frequently reminds us that these stories of magical AI is peddled by both boosters and critics. Critics of AI make the mistake of also assuming that AI is this good, and talk about it as bad use of AI. This unnecessarily inflates the idea that AI can do things that it really canât, adding fuel to magical thinking.
Itâs probably a good idea to more often question if AI is even in the picture, given how effective it has become as a marketing vehicle. Was there AI involved? Perhaps, but not to the extent that salespeople would have you imagine
And yes, there's a name for this kind of criticism: "criti-hype.â A term coined by Lee Vinsel. Read more in Doctorowâs blog post, as always littered with further reading:
https://pluralistic.net/2024/01/29/pay-no-attention/#to-the-little-man-behind-the-curtain
#Critihype #DigitalEthics #AIEthics
Remember when AI was used to generate a George Carlin comedy routine? That didnât happen. Generative AI isnât that good.
After the George Carlinâs estate sued, a representative of the show admitted that it was human-written. The claim that is was produced by an AI trained on Carlinâs material appears to be far from the truth, and rather used as a way to garner attention.
Cory Doctorow frequently reminds us that these stories of magical AI is peddled by both boosters and critics. Critics of AI make the mistake of also assuming that AI is this good, and talk about it as bad use of AI. This unnecessarily inflates the idea that AI can do things that it really canât, adding fuel to magical thinking.
Itâs probably a good idea to more often question if AI is even in the picture, given how effective it has become as a marketing vehicle. Was there AI involved? Perhaps, but not to the extent that salespeople would have you imagine
And yes, there's a name for this kind of criticism: "criti-hype.â A term coined by Lee Vinsel. Read more in Doctorowâs blog post, as always littered with further reading:
#CritiHype #AIEthics #DigitalEthics
Some of the most brilliant work on current digital accessibility. Follow Alexa, read her guides, get her book.
And then, follow her advice.
#accessibility
How things play out when Iâm critiquing a specific implementation of technology »»
Me: âMy feeling is that the premise for this film is unconvincing and its portrayal of the protagonist are in poor taste.â
Miscontructionists: âWhy do you hate all movies!?â
SÀger som Per-Erik Eriksson - det hÀr Àr den nya verkligheten vi har att förhÄlla oss till. Vi har pratat om en kontrollfras i familjen. Har ni?
En kontrollfras för kollegorna Àr nog inte dumt det heller.
Det Àr alltid en bra tid att se över sina lösenord. UppmÀrksamhet kring lÀckta lösenord Àr sÄ klart en given signal om att det Àr dags igen. Jag skrev denna guide 2017 men jag skulle sÀga att det mesta gÀller fortfarande. Den hjÀlper dig att tÀnka till kring dina lösenord online och hur du skyddar dem.
Jag gjorde nÄgra snabba justeringar kring sÄdant som jag vet inte gÀller lÀngre, men sÀg till om du har frÄgor, vill att jag förtydligar nÄgot eller behöver uppdatera nÄgot som Àr förlegat sÄ fixar jag det.
Idag slĂ€pptes avsnittet av Digitala influencer-podden med mig. FĂ„r man kalla sig influencer dĂ„? đ
https://pca.st/3xc59gn3
För ett par veckor sedan skrev jag ut alla modeller jag tagit fram under 2023 som stöd för att reflektera och planera runt potentiella AI-satsningar. Det var fler Ă€n jag först trodde đ
Uppdatering: Jag la in taggen "modeller" pÄ dessa inlÀgg för de ska bli lÀttare att hitta (just nu ligger alla Àven pÄ startsidan pÄ axbom.se), sÄ nu finns alla Àven pÄ den hÀr sidan: https://axbom.se/tag/modeller
Samtliga dessa finns att ladda ner pÄ min sajt i PDF-format, pÄ varje tillhörande blogginlÀgg. Alla utom en. Den kommer jag skriva lite mer om i nÀsta nyhetsbrev. Det handlar om en översikt över alla typer av tekniker som nu hamnat under paraplybegreppet AI. För det Àr minst sagt förvirrande, och rÄder inte nÄgon tydlig enighet, om vad som egentligen menas nÀr olika personer skriver och sÀger "AI".
1. Elementen inom AI-etik, den mest anvÀnda i designsammanhang
2. Kommunikationsmodell för AI-verktyg med operatör, en gedigen modell som jag tror mĂ„nga har nytta av om som ska införa en chattbot â oavsett om den Ă€r AI-baserad eller inte
3. Om en hammare var som AI (en provokation som onekligen fÄtt mest spridning)
4. Förklaringsmodell för ansvar, pĂ„verkan och makt inom AI â ett stöd för att tĂ€nka kring vilka mĂ€nniskor och roller som ansvarar för vad och var riskerna dĂ€rmed finns
5. Data Generation Biases (finns just nu endast pÄ engelska pÄ axbom.com och Àr en tolkning av Suresh och Guttags modell kring var i maskininlÀrning det inbÀddas fördomar, eller bias)
6. Artificiell Intelligens (AI) Ăversikt, kommer i nĂ€sta nyhetsbrev
đŹ Prenumererar gör du hĂ€r: https://axbom.se/nyhetsbrev/
(Jag publicerar oftast Àven i nyhetsbrevet pÄ LinkedIn nÄgon dag senare)
@dekaminski @web2brain
There is also this Swedish case wherein an algorithm was used to allocate school placement based on geographical distance from home, rather than travel distance. This resulted in children having to attend schools that were close on a map but were across a body of water that could take an hour to travel around. The interesting part of the case is how hard it was to get any sort of admittance of wrongdoing on the side of the municipality. This was also before algorithmic decision-making was allowed in the public sector (in 2020). The lawyer writing about this notes how the courts do not understand the technical aspects enoughâŠ
https://www.advokaten.se/tidigare-nummer/2022/nr-6-2022-argang-88/algoritmisk-orattvisa--nar-domstolen-ar-blind-for-digitala-bevis/
@Beantin Very cool. I had no idea this was possible.
@MariaLiv Ja, precis, och Konsumentverket (efter att mÄnga anmÀlningar kommit in) har ju sagt Ät dem för mer Àn ett Är sedan, men det faller lite platt med brevvÀxling som enda tillsÀgelse.
Det Àr lite si och sÄ med tillÀmpningen kan jag tycka.
https://axbom.se/dagens-industri-konsumentverket/
@snowracer Jag trodde det ocksÄ, men inte nÀr de kör kampanjer, typ 99 kr för 3 mÄnader. DÄ kör de fortfarande med att det förlÀngs pÄ Ärsbasis om man missar att sÀga upp i tid (en vecka innan kampanjperiodens slut).