Leprosy and AI: Do the pros outweigh the cons?

Opinion from Timothy Burton, Global Communications Lead at TLM International

Leprosy and AI: Do the pros outweigh the cons?

People in the tech industry say that the sector likes to ‘move fast and break things,' and you could argue that no technology has moved faster or broken more things over the last few years than AI. Almost no sector has been left untouched by this rapid rise. Leprosy and NTDs are no different.

The power of Artificial Intelligence means that it could turbocharge our drive towards the interruption of leprosy transmission, making our work more accurate, more efficient, and more effective. But it also has the potential to put up barriers and create setbacks. In this article, we consider whether the advantages of AI within our sector outweigh the disadvantages.

This chart from Stanford University reflects how quickly AI has advanced in being able to engage with complex scientific and mathematical problems, as well as image classification and other issues that will support the fight to defeat leprosy. (Click image to expand)

Three ways AI can help us in our fight to defeat leprosy

Recognising leprosy skin lesions and other symptoms to aid diagnosis

Supporting data processing as we register new cases

Categorising data to help us understand transmission, those at risk, and how we can improve diagnostics

How AI can help recognise skin lesions

The challenge: Leprosy is most commonly spotted and diagnosed when health workers recognise the signs and symptoms. We do not yet have a fast, field-friendly diagnostic test that can be used around the world.

However, it can be challenging for health workers to recognise the signs and symptoms of leprosy, especially if they are not leprosy specialists. This is because these signs and symptoms can have lots of different appearances. AI can help us to overcome this.

Watch how the WHO Skin NTDs app can help a clinician with a leprosy diagnosis

The WHO Skin NTDs app

The App operates offline and employs a logical algorithm to assist health workers in making accurate diagnoses and treatment decisions. Watch the video on the left to see how that works.

However, according to the WHO website, AI functionality could strengthen the app even further, “In addition to the public version, WHO is collaborating with Universal Doctor and Belle.ai to develop a beta version of the App. This version incorporates two online artificial intelligence (AI)-based algorithms that can classify photos of skin lesions instantly.” This would allow field workers to upload a photo and receive a suggested diagnosis. Field tests are planned in selected countries to evaluate the usability and performance of these AI algorithms in real-world conditions.

For anyone without access to a diagnostic test, this app and its AI functionality could be a significant help in diagnosing leprosy and other NTDs.

Supporting data processing in leprosy case detection

The challenge: Field workers are often over stretched and working on paper-based systems when they are in the field diagnosing cases of leprosy. Sometimes a case will be registered in a paper system, but the information about that case does not reach the centralised level at the Ministry of Health - or perhaps if it does, the data is incomplete.

A new system has been proposed called HANNA, which could allow us to streamline everyday tasks so that notes about new cases at the local level are received and recorded centrally, allowing us to better track the spread of leprosy and other diseases. The next step forward for this proposal would be feasibility testing to understand how it would work in practice.

Earlier this year, ILEP hosted a webinar on the uses of AI in leprosy, which included a presentation on this proposed new tool, HANNA, and the associated CREW system. This extract from the webinar captures how these systems could support our leprosy control work. Click the video above to play.

AI explains how AI can contribute to the fight to defeat leprosy

There are a number of research projects that have explored how AI can contribute towards the fight to defeat leprosy, many of which are captured in this article. The summaries in the article are very technical, as is much of the science surrounding leprosy. However, one of the ways that AI can help us in our leprosy work is by making scientific knowledge understandable to non-technical people. Below is how ChatGPT has explained some of this AI reserach.

Using AI to identify groups at risk of leprosy

This paper from 2018 used a type of artificial intelligence to look for patterns among people with leprosy and their household contacts, with the aim of identifying groups of people who might have a higher risk of leprosy transmission.

Supporting health workers to diagnose PB or MB leprosy

Back in 2021, this paper outlined how researchers trained an AI system using thousands of Brazilian leprosy records so that it could help health workers decide whether a person with leprosy should be classified as paucibacillary (PB) or multibacillary (MB), and then put the AI into a web-based app. The AI tool was a decision tree known as Random Forest, that allows decisions to be made across multiple parameters.

Understanding who is at risk of developing leprosy

Another paper from 2021 helps us understand why some people are at risk of developing leprosy and others are not. Again using Random Forest technology, this study helps us to see how host genetics and immune response appear to influence what happens after exposure to M.leprae, with potential implications for PEP and diagnostics.

How could AI be doing harm to our cause?

This was a topic covered in TLM International's Duck Drop blog series, which is a series of opinion pieces from our Global Advocacy Lead, Mathias Duck. One of Mathias' biggest concerns was around how leprosy and other disabilities are presented through generative AI, which creates text, images, and videos for us based on our prompts.

“AI leans towards what is most common and what might be considered typical. That is usually not persons with disabilities and if you ask specifically for persons with disabilities to be represented, you don’t get much diversity represented within that.”

On the right are two images that ChatGPT produced when we asked it for an image of a person with disabilities in cartoon style. As Mathias said, they both lean into stereotypes and are not reflective of the diversity within disability.

AI reflects our world back to us, which perpetuates our stereotypes

AI could go on to dominate our culture and the materials that we see online, thereby shaping the way we see the world. As Mathias said, “If generative AI creates one-dimensional images that lack diversity, communities that are already marginalised and underrepresented become even more so. Persons affected by leprosy certainly fall within that category.”

On the right are two further images that ChatGPT created when we asked for cartoon images of leprosy. This is Mathias' analysis of them:

"There are things to be commended here. The health messages are good, and it has given us a context where leprosy is more likely to be seen; we’re in a rural place rather than downtown Manhattan or an Italian beach. You also cannot detect any stigma in the photos, which is what we want people to see in leprosy and I’m grateful for that.

"Beyond that, there are issues. My biggest issue: why does the healthcare worker have lighter skin than the patient in both images? This pattern may reflect bias or reinforce harmful stereotypes. That seems to be the impression AI has taken from its sources. Another issue is that leprosy skin patches do not look like this, but other skin conditions may do."

Climate, discrimination, and the impact on women and girls

Climate considerations

As AI has grown so dramatically, so too has our understanding of its environmental impact. As the UN has highlighted, “AI-related water consumption could equal the basic annual domestic needs of 1.3bn people by the end of the decade, while its land footprint may exceed 14,500 square kilometres - roughly twice the size of the Jarkarta metropolitan area.”

In Mathias' opinion “For AI leaders, growth is more important than care for our planet, something that will come at the cost of persons affected by leprosy.” There is now a wealth of research to show that persons affected by leprosy are among the climate vulnerable. If AI growth continues as it currently is, persons affected by leprosy will be exposed to greater risk of droughts, floods, and extreme heat.

Discrimination

The UN's Special Rapporteur on Disability has highlighted the impact AI can have on persons with disabilities.

The report shows how biases in AI can affect whether a person with disabilities can access social protection, healthcare, jobs or education. One of the Special Rapporteur's greatest concerns was that many AI tools are built without the consultation of persons with disabilities and as a consequence they are inaccessible and not inclusive.

As Mathias put it, “Right now, AI companies are steaming ahead with their development at lightning speed, prioritising progress and profit over ensuring that all communities are heard and represented.”

Women and girls

For vulnerable women in low-income countries, and for women with disabilities, AI bias strongly reinforces the belief that women, particularly minoritised women, do not belong in high-earning fields, in which they are already severely underrepresented. This issue has been extensively explored by organisations like UN Women.

For the same reason, AI can also come at a disadvantage to women’s health. As AI tools are mostly developed by men, when women use AI-powered systems to diagnose illnesses, they often receive inaccurate answers, as AI lacks an understanding of women’s bodies, and of how symptoms may present differently in women.

Do the pros of AI outweigh the cons in the fight against leprosy?

In recent weeks, the publicity around AI has been the stuff of nightmares, with at least one leading AI researcher claiming that there is “more than 10% chance AI 'could kill all humans.'”

Whether it is existential crises about the future of humanity, fears over the climate impact of AI data centres, or concerns about AI reliability, we are at a point when most people are aware that AI should be approached with some degree of caution. Even in developing the AI summaries on p.5 of this article, ChatGPT was guilty of hallucinations in its responses.

And so we should see our approach to AI in the field of leprosy with similar caution. As the data on p.1 showed us, AI is accelerating dramatically in its understanding of science and maths, as well as its ability to categorise images and data. These are tools that could be jet fuel for technical experts in their fight to understand and control leprosy transmission. When it comes to technical expertise, AI should be considered an amazing tool when wielded by technical experts.

However, when it comes to the social consequences of AI on a disease that already carries too many negative social consequences, we should be very careful. In its current forms, many generative AI tools will reflect the worst of humanity; stereotypes about minoritised communities will entrench the power imbalance between those with power and those without. Persons affected by leprosy - especially those with disabilities because of the disease and women affected by the disease - are among those most at risk.

If generative AI is to be used and if AI is to be used in non-technical contexts, we need to use it with a critical eye. Are the materials we are receiving from AI reflecting the world as we want it to be, or is it reflecting the worst of ourselves back to us? Are there biases implicit in what we're receiving? Is the information accurate and verified? Are we allowing the ease of AI tools to prevent the voices of real persons affected by leprosy from being heard?

So do the pros outweigh the cons? The world is rarely so black and white. In some instances, AI tools are overwhelmingly positive. In other ways, AI is overwhelmingly negative. For the most part, it sits somewhere in the middle. What remains true: our human judgement and compassion are essential if we want to prevent AI from doing harm to persons affected by leprosy. If we want the pros to outweight the cons, we have to take responsibility for how AI is used.