August 25, 2023, by Brigitte Nerlich

Red and blue AI?

This is another post about artificial intelligence or AI, but it’s what one may call a bit ‘experimental’. I happened to think about an analogy and ran with it, but it might be a completely inappropriate one. Let me know!

Red and green GM

About twenty years ago, at the turn of the millennium, I helped to establish the “Institute for the Study of Genetics, Biorisks and Society” later called the Institute for Science and Society. That was the time when controversies around genetic modification, cloning, BSE and so on dominated the headlines. I remember us thinking about the difference in coverage of and attitudes to what some called ‘red’ and ‘green’ GM, that is, genetic modification for medicine and genetic modification for food and crops. In 2002 Martin Bauer talked about the emerging “contrast between ‘desirable’ biomedical (RED) and ‘undesirable’ agri-food (GREEN) biotechnology”.

If we were thinking of setting up an institute nowadays, we would probable call it something like the “Institute for the Study of AI, Cyber-risks and Society” and perhaps we would be thinking about the emerging contrast between red and green AI or perhaps better red and blue AI. What do I mean by that?

Promising or polluting AI

This distinction came to me when listening to two announcements about advancements in medical AI yesterday, advancement which seem to be coming strong and fast. First I heard about the possibility of determining someone’s heart age as compared to someone’s actual age. Then I heard about the possibility of predicting the risk of Parkinson’s disease from looking at various layers in the retina and analysing biomarkers. Both these advances are related to medical image analysis, as were previous advances in radiology using AI for the detection of breast cancer and so on. In general, the use of algorithms and AI in healthcare (RED AI) seems to be reported in quite a positive light.

At the same time. huge advances are being made in AI using large language models to produce what is called ‘synthetic text’ or help in coding, and here the situation seems to be quite different, as dangers are being discussed from hyped-up existential threats to more tangible dangers, such as discrimination, automated criminalisation and more. This type of AI, or generative AI, is reported in more negative ways, it seems (BLUE AI).

One of the biggest problems with generative AI is perhaps the pollution of knowledge and truth and trust by, as John Thornhill points out, “adding more imperfect information and deliberate disinformation to our knowledge base, […] producing a further ‘enshittification’ of the internet, to use Cory Doctorow’s evocative term”.

I talked about this in a blog post from 6 January this year (and this includes a great summary of the possible pollution of knowledge by ChatGPT itself!). Others have, of course, also noted this problem using interesting metaphors.

Emily Bender, a professor of linguistics, talks for example about the creation of “an equivalent of an oil spill into our information ecosystem”. In a thread on Mastodon Rich Felker also taps into the dangers posed by fossil fuels to frame the dangers of AI: “AI is a lot like fossil fuel industry. Seizing and burning something (in this case, the internet, and more broadly, written-down human knowledge) that was built up over a long time much faster than it could ever be replenished.” Both critics of AI draw inspiration from the dangers posed by the fossil fuel industry and climate change, something I can’t see when people discuss AI in healthcare.

Red and blue AI – really?

So, is there a difference between red and blue AI? I am not totally sure yet.

When asked about the benefits of AI, many people might say something like ‘diagnosing diseases’ and when asked about the risks of AI they might say bullshit, confabulation and disinformation. But is that enough to think there is an emerging dichotomy between red and blue, good and bad?

In both cases, there are dangers to workers in various industries, algorithmic discrimination and so on. In both cases, the medical/imaging advances and the generative/large language model advances, we are dealing with a process of pattern matching. So shouldn’t that lead to the same issues, problems and risks? Regarding, for example data extraction? Where do the data, images or words come from, who has provided them with or without consent?

And where do robots and autonomous vehicles fit in? I don’t know…..

However, if there is at least a slight perceived difference between what I call red and blue AI, that is, medical AI and non-medical AI, could this have impacts on people’s attitudes to ‘AI’? Would they see medical or red AI as “useful, morally sound, and to be encouraged”, as Bauer found with red GM (2002, p. 97), while they might be more sceptical about non-medical or blue AI? Again, I don’t know. But I think it is something worth thinking about in the context of researching responsible AI or trustworthy AI. More research needed!

Image: Red Blue Sunset (Wikimedia Commons)






Posted in artifical intelligence