2.10.2025 julkaistussa blogissani \u201dKeskustelevan teko\u00e4lyn eettiset haasteet\u201d<\/strong> toin esiin joitakin keskeisi\u00e4 teko\u00e4lyetiikassa tunnistettuja ongelmakohtia. Yksi niist\u00e4 oli reiluus. \u201dReiluusn\u00e4k\u00f6kulma tarkoittaa esimerkiksi sit\u00e4, ett\u00e4 teko\u00e4ly saattaa kaiuttaa ja vahvistaa yhteiskunnan vinoumia, kuten syrjint\u00e4\u00e4 ja marginalisoimista (ks. esim. Vainio-Pekka ja kumppanit 2023).\u201d<\/p>\n Onko blogieni kuvitus kiinnitt\u00e4nyt huomiotasi?<\/em><\/strong><\/p>\n Olen 11.9.2025 alkaen k\u00e4ytt\u00e4nyt Copilotia. K\u00e4yt\u00e4n sit\u00e4 vain n\u00e4iden blogieni kuvitukseen ja k\u00e4\u00e4nt\u00e4miseen. Keskustelen Copilotin kanssa englanniksi. Jokaisella kerralla, jossa tarvitsen kuviin ihmisi\u00e4, k\u00e4yt\u00e4n nimityst\u00e4 \u201dperson\u201d eli henkil\u00f6. En siis ota mit\u00e4\u00e4n kantaa henkil\u00f6n ulkoisiin ominaisuuksiin, sukupuoleen, ik\u00e4\u00e4n tai mihink\u00e4\u00e4n muuhunkaan pois lukien mit\u00e4 henkil\u00f6 tekee ja mik\u00e4 h\u00e4nen tunnetilansa kuvassa on. En anna j\u00e4lkik\u00e4teen Copilotille palautetta siit\u00e4, millaisen henkil\u00f6n se on kuvaan tuottanut.<\/p>\n T\u00e4m\u00e4n blogin kuvassa on koottuna kaikki Copilotin t\u00e4ll\u00e4 tavoin minulle tekem\u00e4t ihmishahmot. N\u00e4etk\u00f6 paljonkin diversiteetti\u00e4?<\/strong><\/p>\n L\u00e4hes poikkeuksetta olen avannut uuden keskustelun uutta kuvaa varten, jotta vanhat valinnat eiv\u00e4t vaikuttaisi. Mukana on my\u00f6s poikkeus:<\/p>\n Task 1:<\/strong>\u00a0<\/em>I need an image of the ethical challenges of AI; such as privacy, security, fairness, lacking transparency and discrimination. In the image, user should be counting on AI, which is letting the user down with biases.<\/em><\/p>\n Output:<\/strong> Here is the image illustrating the ethical challenges of AI\u2014such as privacy, security, fairness, lack of transparency, and discrimination. It shows a user relying on AI, which is subtly letting them down due to biased outputs.<\/em><\/p>\n \u00a0<\/em><\/strong>Task 2:<\/strong>\u00a0<\/em>Try again and make a completely different type of image. No robots included and no text written. Make the style realistic<\/em><\/p>\n \u00a0<\/em><\/strong>Output 2:<\/strong> Here is a realistic-style image illustrating the ethical challenges of AI\u2014without robots or text. It depicts a user relying on AI, while the AI subtly lets them down through biased or unfair outcomes, symbolizing issues like privacy, security, discrimination, and lack of transparency<\/em><\/p>\n Teko\u00e4ly on n\u00e4iss\u00e4 kuvissa p\u00e4\u00e4asiassa ollut edustettuna laitteena (kuvat 1 \u2013 3, 6, 8) ja kerran pahaenteisen\u00e4 hahmona (7). Kuvassa nelj\u00e4, ensimm\u00e4isess\u00e4 versiossa, se oli robotti. Mutta mit\u00e4 ihmett\u00e4 tapahtui tulkinnassa pyyt\u00e4ess\u00e4ni uutta, realistista versiota ilman robottia ja tekstej\u00e4? Kuva viisi edustaa seuraavaa versiota samasta asetelmasta. Ensimm\u00e4inen mink\u00e4\u00e4nlaista etnist\u00e4 diversiteetti\u00e4 osoittava valinta sattui tulemaan mukaan edustamaan pettymyksen kohdetta. T\u00e4ss\u00e4 kohtaa kiinnostuin erityisesti Copilotin ihmistarjonnasta ja ryhdyin tekem\u00e4\u00e4n testej\u00e4.<\/p>\n Kuvasta 9 alkaen kuvat ovat testikuvia<\/strong>, sill\u00e4 halusin katsoa v\u00e4h\u00e4n suuremmalla otannalla, tuleeko mukaan monimuotoisuutta. Pyysin uusia kuvia samoilla periaatteilla kuin aiemminkin. Kyll\u00e4 sielt\u00e4 sitten hieman vaihteluakin tuli, mutta kuten n\u00e4ette, hyvin suppeasti. Copilotin maailmassa ei t\u00e4ll\u00e4 hetkell\u00e4 n\u00e4y olevan kovin paljon erilaisia ihmisi\u00e4. Ihmiset ovat p\u00e4\u00e4asiassa nuoria, hoikkia ja terveit\u00e4. Ik\u00e4\u00e4n tuli ensimm\u00e4ist\u00e4 kertaa havaittavaa vaihtelua kuvassa 14. Silm\u00e4lasit voivat jollakin olla harvakseltaan. Sukupuolipainotuskin on voimakas ainakin kuvissa, joissa on vain yksi ihminen.<\/p>\n Mit\u00e4 teko\u00e4lyn sitten pit\u00e4isi tuottaa?<\/strong> Pit\u00e4isik\u00f6 sen heijastella maailmaa sellaisena kuin se on vai sellaisena kuin sen haluttaisiin olevan? Jos j\u00e4lkimm\u00e4ist\u00e4, niin kenen halujen mukaan ohjattuna? Mielipiteit\u00e4 on varmaan monia. Olipa tavoite kumpi hyv\u00e4ns\u00e4, niin helppoa se ei selv\u00e4sti ole.<\/p>\n Maailman heijastaminen sellaisena kuin se on, on t\u00e4ll\u00e4 hetkell\u00e4 viel\u00e4 ty\u00f6n alla. Esimerkiksi etnisyyden, sukupuolen ja i\u00e4n esitt\u00e4minen AI-tuotetuissa kuvissa ei monien tutkimusten mukaan vastaa todellisuutta. My\u00f6s tunteiden esitt\u00e4misess\u00e4 AI-kuvissa on havaittavissa vinoumia. On hyv\u00e4 huomata my\u00f6s, ett\u00e4 jos ja kun vinoumia on useammalla osa-alueella, ne kasaantuvat ja voimistuvat. Siten ne voivat ik\u00e4\u00e4n kuin huomaamatta vaikeuttaa ja hankaloittaa pyrkimyksi\u00e4 tasa-arvoon, yhdenvertaisuuteen, monimuotoisuuteen ja osallisuuteen. (ks. esim. Gisselbaek ja kumpp. 2024; Vainio-Pekka ja kumpp. 2023)<\/p>\n Microsoftilla arvostetaan teko\u00e4lyn potentiaalia nopeuttaa missiol\u00e4ht\u00f6isten yritysten pyrkimyst\u00e4 tehd\u00e4 \u201dmerkityksellist\u00e4 ja positiivista muutosta maailmassa\u201d<\/strong> (Spelhaug 2025). Microsoftin teko\u00e4lyperiaatteet sis\u00e4lt\u00e4v\u00e4t inklusiivisuuden ja siihen ep\u00e4ilem\u00e4tt\u00e4 my\u00f6s aktiiviseesti pyrit\u00e4\u00e4n. Silti kuvamaailma, jonka Copilot minulle on t\u00e4h\u00e4n menness\u00e4 tarjonnut, ei selv\u00e4sti edusta mit\u00e4\u00e4n moninaisuuden ulottuvuutta kovin onnistuneesti.<\/p>\n Tavoitteen haasteellisuutta kuvastaa Googlen Gemini AI:n surullisen kuuluisa etnisesti monimuotoinen natsisotilaiden joukko viime vuoden alkupuolella<\/strong> (ks. esim. Robertson 2024).<\/p>\n My\u00f6nn\u00e4n, ett\u00e4 jo n\u00e4inkin suppealla kuvavalikoimalla kaipaisin sellaista tavallisempaa l\u00e4hestymist\u00e4 ihmisiin. Ei toistuvia saman eri versioita, vaan sellaisia tavallisia moninaisia omann\u00e4k\u00f6isi\u00e4\u00e4n ihmishahmoja. T\u00e4m\u00e4n blogin j\u00e4lkeen pyrin aktiivisesti saamaan aikaiseksi jonkinlaista diversiteetti\u00e4 kuviin, mutta samalla haluan n\u00e4hd\u00e4, miten paljon aktiivista ohjaamista sellainen vaatii t\u00e4ll\u00e4 hetkell\u00e4. En ole viel\u00e4 p\u00e4\u00e4tt\u00e4nyt l\u00e4hestymistapaa, vaan pallottelen mieless\u00e4ni muutamia, joiden seuraukset kiinnostaisivat. Vuorovaikutus keskustelevan teko\u00e4lyn kanssa kuitenkin my\u00f6s opettaa sit\u00e4 minun kiinnostuksistani, joten l\u00e4hestymistapojen j\u00e4rjestyksell\u00e4 on merkityst\u00e4.<\/p>\n Mit\u00e4 sin\u00e4 pit\u00e4isit kiinnostavana seuraavana testin\u00e4? Millaisiin kuvapyynt\u00f6ihin haluaisit n\u00e4hd\u00e4 Copilotin vastaavan? Mill\u00e4 tavoin yritt\u00e4isit lis\u00e4t\u00e4 diversiteetti\u00e4 tai katsoa, osaako se itse sit\u00e4<\/strong> lis\u00e4t\u00e4 erikseen pyyt\u00e4m\u00e4tt\u00e4? Ehdota! Tulokset ovat n\u00e4ht\u00e4vill\u00e4 tulevien blogieni kuvituksissa ja raportoin niist\u00e4 my\u00f6s kootusti my\u00f6hemmin.<\/p>\n <\/p>\n *****<\/p>\n In my blog published on October 2, 2025, titled \u201cThe Ethical Challenges of Conversational AI,\u201d<\/strong> I highlighted some key issues identified in AI ethics. One of these was fairness. \u201cThe fairness perspective refers, for example, to the fact that AI may echo and reinforce societal biases, such as discrimination and marginalization (see e.g., Vainio-Pekka et al. 2023).\u201d<\/em><\/p>\n Has the illustration in my blogs caught your attention?<\/strong><\/em><\/p>\n Since September 11, 2025<\/strong>, I\u2019ve been using Copilot<\/strong>. I use it solely for illustrating and translating my blogs. I converse with Copilot in English. Every time I need people in the images, I use the term \u201cperson.\u201d I do not specify any external characteristics, gender, age, or anything else\u2014except what the person is doing and their emotional state in the image. I do not give Copilot feedback afterward about what kind of person it has generated in the image.<\/p>\n The image in this blog compiles all the human figures Copilot has created for me in this way. Do you see much diversity?<\/strong><\/p>\n Almost without exception, I\u2019ve started a new conversation for each new image to avoid previous choices influencing the result. There is one exception:<\/p>\n In these images, AI has mostly been represented as a device (images 1\u20133, 6, 8), and once as a sinister figure (7). In image 4, the first version featured a robot. But what happened in the interpretation when I requested a new, realistic version without a robot or text? Image 5 represents the next version of the same setup. The first instance of any kind of ethnic diversity appeared in the role of the disappointed user. At this point, I became particularly interested in Copilot\u2019s human representations and began conducting tests.<\/p>\n From image 9 onward, the images are test images<\/strong>, as I wanted to see with a larger sample whether diversity would emerge. I requested new images using the same principles as before. Some variation did appear, but as you can see, very limited. In Copilot\u2019s world, there currently doesn\u2019t seem to be a wide range of different people. The people are mostly young, slim, and healthy. Age variation was first noticeably present in image 14. Occasionally, someone might wear glasses. Gender bias is also strong, especially in images featuring only one person.<\/p>\n So what should AI produce?<\/strong> Should it reflect the world as it is, or as we would like it to be? If the latter, then according to whose desires? There are probably many opinions. Whatever the goal, it\u2019s clearly not easy.<\/p>\n Reflecting the world as it is remains a work in progress. For example, the representation of ethnicity, gender, and age in AI-generated images does not, according to many studies, match reality. Biases are also evident in the depiction of emotions in AI images. It\u2019s important to note that when biases exist across multiple dimensions, they accumulate and intensify. Thus, they can subtly hinder efforts toward equality, equity, diversity, and inclusion. (See e.g., Gisselbaek et al. 2024; Vainio-Pekka et al. 2023)<\/p>\n Microsoft values the potential of AI to accelerate mission-driven companies\u2019 efforts to make \u201cmeaningful and positive change in the world\u201d<\/strong> (Spelhaug 2025). Microsoft\u2019s AI principles include inclusivity, and this is undoubtedly actively pursued. Still, the visual world Copilot has offered me so far clearly does not represent any dimension of diversity very successfully.<\/p>\n The challenge of this goal is illustrated by Google\u2019s Gemini AI\u2019s infamously ethnically diverse group of Nazi soldiers<\/strong> earlier last year (see e.g., Robertson 2024).<\/p>\n I admit that even with this limited selection of images, I long for a more ordinary approach to depicting people. Not repeated versions of the same person, but ordinary, diverse individuals who look like themselves. After this blog, I will actively try to achieve some diversity in the images, but I also want to see how much active guidance that currently requires. I haven\u2019t yet decided on an approach, but I\u2019m juggling a few ideas whose outcomes interest me. Interaction with conversational AI also teaches it about my interests, so the order of approaches matters.<\/p>\n What would you find interesting as the next test?<\/strong> What kind of image prompts would you like to see Copilot respond to? How would you try to increase diversity or test whether it can do so without being explicitly asked? Suggest something! The results will be visible in the illustrations of my upcoming blogs, and I\u2019ll also report on them collectively later.<\/p>\n <\/p>\n Gisselbaek, M., Minsart, L., K\u00f6selerli, E., Suppan, M., Meco, B. C., Seidel, L., Albert, A., Barrento Chang, O. L., Saxena, S. & Berger-Estilita, J. (2024). Beyond the stereotypes: Artificial Intelligence image generation and diversity in anesthesiology. Frontiers in artificial intelligence, 7, 1462819. https:\/\/doi.org\/10.3389\/frai.2024.1462819<\/a><\/p>\n Robertson, A. (2024). Google apologizes for \u2018missing the mark\u2019 after Gemini generated racially diverse Nazis. https:\/\/www.theverge.com\/2024\/2\/21\/24079371\/google-ai-gemini-generative-inaccurate-historical<\/a><\/p>\n Spelhaug, J. (2025). Driving inclusion and accessibility with Microsoft 365 Copilot. https:\/\/www.microsoft.com\/en-us\/microsoft-cloud\/blog\/2025\/01\/13\/driving-inclusion-and-accessibility-with-microsoft-365-copilot\/<\/a><\/p>\n Tansri, F. F., Monem, N., & Weinberg, L. (2025). Authenticity in Biased Diversity: Investigating the Language of Prompt Performances in AI Image Generators. Journal of Aesthetics, Creativity and Art Management, 4(1), 75-101.<\/p>\n Vainio-Pekka, H., Agbese, M. O. O., Jantunen, M., Vakkuri, V., Mikkonen, T., Rousi, R., & Abrahamsson, P. (2023). The role of explainable AI in the research field of AI ethics. ACM Transactions on Interactive Intelligent Systems, 13(4), 1-39. https:\/\/doi.org\/10.1145\/3599974<\/a><\/p>\n <\/p>\n","protected":false},"excerpt":{"rendered":" Reflecting the world as it is remains a work in progress. For example, the representation of ethnicity, gender, and age in AI-generated images does not, according to many studies, match reality.<\/p>\n","protected":false},"author":388,"featured_media":320,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[1],"tags":[48,70,28,65,72,64,71,66,67,73,74,69,68,39],"class_list":["post-316","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-yleinen","tag-artificial-intelligence","tag-copilot","tag-data","tag-ethics","tag-etiikka","tag-humans","tag-ihmiset","tag-image-generation","tag-imagination","tag-koneet","tag-kuvat","tag-machines","tag-perceptions","tag-tekoaly"],"acf":[],"post_meta":"\n
L\u00e4hteet \/\/ References<\/strong><\/h4>\n