Better Images of AI: Research-Informed Diversification of Stock Imagery of Artificial Intelligence
Lead Research Organisation:
University of Cambridge
Department Name: Future of Intelligence
Abstract
The stock images currently used to illustrate news stories, reports, and other publications about artificial intelligence (AI) are often misleading and threaten to distort the public understanding of new technologies. There is effectively a visual barrier to getting the informed input of a greater diversity of people into the design, development and use of AI systems, which are currently facing increasing criticism for being biased and untransparent. 'Better Images of AI' is a project that aims to develop a repository of stock images of AI that more accurately represent contemporary AI technologies and their underlying concepts. Through building an international network that brings together AI experts, media representatives, artists, activists, and the professionals who use these stock images, the project aims to investigate how best to identify existing and develop new stock images that both meet the needs of the users and accurately shape public perceptions.
| Title | 30 images in 30 days |
| Description | Follow-on funding from AI4Media, a consortium funded by a Horizon 2020 grant (see 'Further Funding') secured funding to commission artworks from three artists; collaboration with the BBC secured funding for a collaboration with another artist collective. The team used the report that came out of the AHRC-funded research, 'Better Images of AI: A Guide for Users and Creators', to give guidelines to the artists regarding the kinds of images the database needed, and regarding the kinds of ideas and tropes to avoid. In total, the project released 30 new images in November 2023, publicising it widely as '30 images in 30 days', with a daily spotlight on each new image. |
| Type Of Art | Image |
| Year Produced | 2023 |
| Impact | The new images have been used just as widely as the existing ones, with a total estimated (based on Google News searches) 261,000 uses. The new images have been used in publications by organizations ranging from the German news outlet Deutsche Welle to AlgorithmWatch and the UK's National Cybersecurity Centre. |
| URL | https://blog.betterimagesofai.org/a-new-better-image-of-ai-every-day-for-november/ |
| Description | Responding to a dire lack of representative, inclusive, and accurate stock images of artificial intelligence (AI), the Better Images of AI project and database launched in 2021 to curate, commission, and advocate for alternative images that better represent the current state of technology. Securing AHRC funding in 2022 enabled its project team to achieve the following objectives, as set out in our original proposal: 1. Establishing a network. We established an international network of over a hundred stakeholders who work with, inform, commission, or produce stock images of AI, including journalists, academics, marketers, communicators, artists and those in the technical community. Noteworthy additions to this network include the UK Government; an existing specialist network of Human-Centered AI (HCAI) researchers; and news outlets ranging from AI Business to the Daily Mirror. 2. Hosting relationship-building events. Groups of up to 20 of these stakeholders at a time were invited to four online roundtable events organised with the aim to investigate the needs and wants of image users. These roundtables revealed motivations for using certain kinds of images; audience responses that image users intend to invoke; and decision-making processes that lead to certain images being used. 3. Testing an existing dataset in a series of stakeholder workshops. These workshops focused on the 26 images available in the Better Images of AI database at the time, and revealed which images met the needs of image users best; what kinds of images need to be added to the database; and what kinds of purposes the existing images could best fulfil. 4. Developing artists' briefs and guidelines. The new knowledge generated by these events informed our writing of the guide 'Better Images of AI: A Guide for Users and Creators'. This guide, aimed at a non-academic audience, serves as a concise reference tool filled with key considerations that should be front of mind for anyone using or creating images of AI. The findings from these roundtables and workshops are also currently being written up as an academic paper. |
| Exploitation Route | 'Better Images of AI: A Guide for Users and Creators' is intended to inspire further work by artists, editors, and academics a like. Its main goal is to show people why many of the existing images of AI that are currently most often used are insufficient. We have already seen the research findings that are communicated in this guide being put to use by others, as detailed in the Impact section. The sectors to which this work is most relevant are Government, Democracy and Justice (for helping to inform constituents accurately about the current state of AI to stimulate informed democratic decision-making); Creative Economy (showing artists the areas in which their creative interventions are most needed); and Media/Journalism (helping editors make informed decisions about their visual messaging). |
| Sectors | Communities and Social Services/Policy Creative Economy Digital/Communication/Information Technologies (including Software) Education Electronics Financial Services and Management Consultancy Government Democracy and Justice Culture Heritage Museums and Collections Other |
| URL | https://blog.betterimagesofai.org/better-images-of-ai-guide/ |
| Description | The project has had significant impact in two ways. 1. Significant behaviour change enabled by the image database. Over the course of the project, the Better Images of AI (BIoAI) stock image database grew to 36 images. Although still a limited number, the impact of these images has been extraordinary. Major press outlets have started to regularly use these images instead of the standard white plastic androids and Terminators (e.g. Time, The Independent and The Washington Post), meaning that millions of readers have been exposed to the BIoAI images rather than the misleading, inaccurate ones. Beyond news media, the images have also been incorporated by research institutions including The Alan Turing Institute and the universities of Cambridge and Edinburgh. 2. Significant behaviour change informed by the report. Although the report was released only a month and a half ago, on 24 January, we have identified significant change made to the practice of policymakers, artists, and media representatives. While in some cases readers decided to specifically use images from the BIoAI database, others decided to look for alternatives more widely, or commission or design their own images - a similar image creation project has now been instigated by DeepMind for their communications. For example, the Pulitzer Center on Crisis Reporting in Washington D.C. removed and replaced an anthropomorphic image of AI informed by the Guide; a student at Edinburgh University noted that "It has been super useful for academics too - I shared with my uni press office and they now use it as a go-to source!"; and news producer Joe Amditis detailed in a series of Tweets: "awesome resource about the types and styles of images we use alongside stuff about AI and technology went through and changed the entire design/style of my ChatGPT ebook because of it. worth a read!" |
| First Year Of Impact | 2021 |
| Sector | Communities and Social Services/Policy,Creative Economy,Digital/Communication/Information Technologies (including Software),Education,Financial Services, and Management Consultancy,Government, Democracy and Justice,Culture, Heritage, Museums and Collections,Retail,Other |
| Impact Types | Cultural Societal Economic Policy & public services |
| Description | House of Lords |
| Geographic Reach | National |
| Policy Influence Type | Contribution to new or improved professional practice |
| Impact | The representative who invited Dr Dihal attended the launch of the project report and states that "The guide has remained on my desk and has sparked a couple of conversations. I spoke with our comms person who'd developed an animation launching an inquiry into autonomous weapons systems - sadly too late, she said it would have been very helpful!" This strongly suggests that for the next project this comms person is involved in, her behaviour will be changed. |
| Description | UKRI |
| Geographic Reach | National |
| Policy Influence Type | Contribution to new or improved professional practice |
| Impact | An AHRC AI & Design representative stated in an email to the project team, "I noted the impressive list of publications that are using the Better Images stock photography. We used it for the launch of our Responsible AI Ecosystem programme, as well." |
| Title | Better Images of AI Image Library |
| Description | The Better Images of AI image library serves as a repository of stock images, all under a Creative Commons licence, that showcase approaches to depicting AI that are representative, inclusive, and accurate. Over the course of the research project, the database grew to 36 images, most of which are original artworks commissioned by the project team or donated by artists. Follow-on funding secured in 2023 allowed the project team to commission a further 30 images, which were all added in November 2023. |
| Type Of Material | Database/Collection of data |
| Year Produced | 2021 |
| Provided To Others? | Yes |
| Impact | Major press outlets regularly use these images instead of the standard white plastic androids and Terminators (e.g. Time, The Independent and The Washington Post), meaning that millions of readers have been exposed to the BIoAI images rather than the misleading, inaccurate ones. Beyond news media, the images have also been incorporated by research institutions including The Alan Turing Institute and the universities of Cambridge and Edinburgh, and third-sector organizations such as AlgorithmWatch. A Google news search for "better images of AI" returns 261,000 results as of March 2024, all of which display at least one image from the database. |
| URL | https://betterimagesofai.org/images |
| Description | BBC R&D |
| Organisation | British Broadcasting Corporation (BBC) |
| Country | United Kingdom |
| Sector | Public |
| PI Contribution | We have created a resource that is being actively used within the BBC to inform the ways in which it reports on AI. The Guide is used by journalists and editors alike. At the same time, this collaboration with the BBC has led to more people within the corporation to be aware of and use the Project's image database. |
| Collaborator Contribution | The BBC has provided support in the form of allowing BBC R&D staff members to spend working hours on the Better Images of AI project, helping organise the roundtables and workshops. We were able to access the extensive network of journalists within the BBC and beyond that made these workshops and roundtables a success. Our main collaborator at the BBC co-chaired several of these sessions. BBC R&D also continued to provide technical support for the development of the Better Images of AI website. |
| Impact | 'Better Images of AI: A Guide for Users and Creators' presents the outcomes of four research workshops and four roundtables, several key attendees of which were directly identified through the BBC partnership. At least five key attendees who are quoted in the Guide would not have attended if we had not had this collaboration with the BBC; although it is difficult to prove this, we suspect that having the BBC name listed also convinced several other invited stakeholders to attend. The BBC has also contributed over a dozen images to the database through partnerships with artists. |
| Start Year | 2021 |
| Description | BBC 'Get Curious' event at Manchester Science Festival |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Schools |
| Results and Impact | Hundreds of children participated in an event organised by project collaborator Tristan Ferne at our partner organisation BBC R&D. Children were asked to draw what AI looks like to them, following a public engagement activity as part of the BBC's Get Curious event at the Manchester Science Festival. Over a hundred images were collected, which helped inform the researchers how children's perceptions of AI are shaped and influenced, and where to intervene in order to ensure children do not grow up with misleading ideas about the technology that is changing their lives. |
| Year(s) Of Engagement Activity | 2022 |
| URL | https://blog.betterimagesofai.org/what-do-children-think-ai-looks-like/ |
| Description | Co-creating Better Images of AI at London Data Week |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Policymakers/politicians |
| Results and Impact | Better Images of AI aims to increase AI literacy and inclusion by providing alternatives to misleading images. One of the project's core tenets is that everyone should get a say in what AI looks like and how they want to make it work for them. No one perspective or group should dominate how Al is conceptualised and imagined. We were therefore delighted to be able to run the workshop 'Co-creating Better Images of AI' during London Data Week in July 2023. We convened over 50 people, including creative artists, technologists, and local government representatives, to each make their own images of AI. Informed by our report, 'Better Images for AI: A Guide for Users and Creators', and led by Tania Duarte and artist Yasmine Boudiaf, all participants created an image that best represented AI from their own personal perspective and experience. Although not made by professional artists, several of these images proved to be so outstanding that they have been added to the Better Images of AI database, as well as Yasmine Boudiaf's own image. It is difficult to pick from the pre-determined list below what the most significant output was of this activity; we would say it is the ideas about AI and empowerment we have given the participants, as well as gathering new images for the database. |
| Year(s) Of Engagement Activity | 2023 |
| URL | https://blog.betterimagesofai.org/co-creating-better-images-of-ai/ |
| Description | LOTI website coverage: Better Images of AI: An Important Discussion for Society |
| Form Of Engagement Activity | A magazine, newsletter or online publication |
| Part Of Official Scheme? | No |
| Geographic Reach | Local |
| Primary Audience | Schools |
| Results and Impact | Elo Esaloni, a Year 13 student who attended our workshop during London Data Week in July 2023, wrote up the workshop for the website of the London Office of Technology and Innovation (LOTI). It was great for us to see a young Black woman describe the workshop as follows: "The first thing I noticed when I arrived at the Gallery was the diversity of people who had decided to attend this workshop on a rainy Tuesday. Students, tech experts, artists and those simply interested in AI and its social impacts had congregated outside the venue; a range of races, ages, and genders converging around this increasingly important topic. This gave me a good feeling coming into the workshop, as I knew that there would be a great diversity of perspectives, which would enable more thought-provoking discussions. " |
| Year(s) Of Engagement Activity | 2023 |
| URL | https://loti.london/blog/better-images-of-ai-an-important-discussion-for-society/ |
| Description | LSE Impact Blog coverage |
| Form Of Engagement Activity | Engagement focused website, blog or social media channel |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | Better Images of AI network members Jenn Chubb and Raziye Buse Çetin wrote an article on the project for the LSE Impact Blog (58,000 Twitter followers). |
| Year(s) Of Engagement Activity | 2022 |
| URL | https://blogs.lse.ac.uk/impactofsocialsciences/2022/07/25/we-need-better-ai-imagery-for-better-scien... |
| Description | Project coverage in Columbia Journalism Review |
| Form Of Engagement Activity | A press release, press conference or response to a media enquiry/interview |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | In a July 2023 article for the Columbia Journalism Review, three leading voices in AI ethics (Sayash Kapoor, Hilke Schellman and Ari Sen) referred to the Better Images of AI project as an example of good practice in coverage of AI in journalism. Practising what they preached, the article is illustrated with an image from the Better Images database. |
| Year(s) Of Engagement Activity | 2023 |
| URL | https://www.cjr.org/analysis/how-to-report-better-on-artificial-intelligence.php |
| Description | Report launch event |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | 50 invited stakeholders attended the launch event for the 'Better Images of AI' guide in person at the Alan Turing Institution on 24 January 2023. Two news outlets reported on the event afterwards, and the PI and Co-I received invitations for further collaborations and presentations including for London Data Week; AI UK; and the House of Lords. The project's Twitter engagement about the event (@ImagesofAI) reached a further 10,000 users, and the series of Tweets explaining the contents of the report reached 15,000 views. |
| Year(s) Of Engagement Activity | 2023 |
| URL | https://blog.betterimagesofai.org/launch-of-a-guide-for-users-and-creators-of-images-of-ai/ |
| Description | Report on event in national news |
| Form Of Engagement Activity | A press release, press conference or response to a media enquiry/interview |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Media (as a channel to the public) |
| Results and Impact | A reporter from AI Business (10,800 Twitter followers) attended the launch event for the 'Better Images of AI' guide and interviewed the PI and Co-I, writing this up in an article that appeared the next week. |
| Year(s) Of Engagement Activity | 2023 |
| URL | https://aibusiness.com/about-us |