Three Observations about Justifying AI
Written by: Anantharaman Muralidharan, G Owen Schaefer, Julian Savulescu
Cross-posted with the Journal of Medical Ethics blog
Consider the following kind of medical AI. It consists of 2 parts. The first part consists of a core deep machine learning algorithm. These blackbox algorithms may be more accurate than human judgment or interpretable algorithms, but are notoriously opaque in terms of telling us on what basis the decision was made. The second part consists of an algorithm that generates a post-hoc medical justification for the core algorithm. Algorithms like this are already available for visual classification. When the primary algorithm identifies a given bird as a Western Grebe, the secondary algorithm provides a justification for this decision: “because the bird has a long white neck, pointy yellow beak and red eyes”. The justification goes beyond just a description of the provided image or a definition of the bird in question, and is able to provide a justification that links the information provided in the image to the features that distinguish the bird. The justification is also sufficiently fine grained as to account for why the bird in the picture is not a similar bird like the Laysan Albatross. It is not hard to imagine that such an algorithm would soon be available for medical decisions if not already so. Let us call this type of AI “justifying AI” to distinguish it from algorithms which try, to some degree or other, to wear their inner workings on their sleeves.
Possibly, it might turn out that the medical justification given by the justifying AI sounds like pure nonsense. Rich Caruana et al present a case whereby asthmatics were deemed less at risk of dying by pneumonia. As a result, it prescribed less aggressive treatments for asthmatics who contracted pneumonia. The key mistake the primary algorithm made was that it failed to account for the fact that asthmatics who contracted pneumonia had better outcomes only because they tended to receive more aggressive treatment in the first place. Even though the algorithm was more accurate on average, it was systematically mistaken about one subgroup. When incidents like these occur, one option here is to disregard the primary AI’s recommendation. The rationale here is that we could hope to do better than by relying on the blackbox alone by intervening in cases where the blackbox gives an implausible recommendation/prediction. The aim of having justifying AI is to make it easier to identify when the primary AI is misfiring. After all, we can expect trained physicians to recognise a good medical justification when they see one and likewise recognise bad justifications. The thought here is that the secondary algorithm generating a bad justification is good evidence that the primary AI has misfired.
The worry here is that our existing medical knowledge is notoriously incomplete in places. It is to be expected that there will be cases where the optimal decision vis a vis patient welfare does not have a plausible medical justification at least based on our current medical knowledge. For instance, Lithium is used as a mood stabilizer but the reason why this works is poorly understood. This means that ignoring the blackbox whenever a plausible justification in terms of our current medical knowledge is unavailable will tend to lead to less optimal decisions. Below are three observations that we might make about this type of justifying AI.
Cognitive snobbery: The Unacceptable Bias in Favour of the Conscious
There are many corrosive forms of discrimination. But one of the most dangerous is the bias in favour of consciousness, and the consequent denigration of the unconscious.
We see it everywhere. It’s not surprising. For when we’re unreflective – which is most of the time – we tend to suppose that we are our conscious selves, and that the unconscious is a lower, cruder part of us; a seething atavistic sea full of monsters, from which we have mercifully crawled, making our way ultimately to the sunlit uplands of the neocortex, there to gaze gratefully and dismissively back at what we once were. It’s a picture encoded in our self-congratulatory language: ‘Higher cognitive function’; ‘She’s not to be blamed: she wasn’t fully conscious of the consequences.’: ‘In the Enlightenment we struck off the shackles of superstition and freed our minds to roam.’ Continue reading
Shaming unvaccinated people has to stop. We’ve turned into an angry mob and it’s getting ugly
Julian Savulescu, University of Oxford and Alberto Giubilini, University of Oxford
Unvaccinated mother, 27, dies with coronavirus as her father calls for fines for people who refuse jab.
This is the kind of headline you may have seen over the past year, an example highlighting public shaming of unvaccinated people who die of COVID-19.
One news outlet compiled a list of “notable anti-vaxxers who have died from COVID-19”.
There’s shaming on social media, too. For instance, a whole Reddit channel is devoted to mocking people who die after refusing the vaccine.
COVID-19 vaccinations save lives and reduce the need for hospitalisation. This is all important public health information.
Telling relatable stories and using emotive language about vaccination sends a message: getting vaccinated is good.
But the problem with the examples above is their tone and the way unvaccinated people are singled out. There’s also a murkier reason behind this shaming.
Paying for the Flu Vaccine
By Ben Davies
As I do every winter, I recently booked an appointment for a flu vaccine. I get it for free in the UK. If I didn’t have asthma, I’d still get vaccinated, but it would cost me between £9 and £14.99. That is both an ethical error on the part of the government, and may be a pragmatic one too.
The double ethical mistake of vaccinating children against COVID-19
Alberto Giubilini
Oxford Uehiro Centre for Practical Ethics
University of Oxford
Against the Joint Committee on Vaccination and Immunisation (JCVI)’s advice that did not recommend COVID-19 vaccination for children, the four Chief Medical Officers in the UK have just recommended that all children aged 12-15 should be vaccinated with the mRNA Pfizer/BioNTech vaccine.
This is a double ethical mistake, given our current state of knowledge.
Philosophical Fiddling While the World Burns
By Charles Foster
An unprecedented editorial has just appeared in many health journals across the world. It relates to climate change.
The authors say that they are ‘united in recognising that only fundamental and equitable changes to societies will reverse our current trajectory.’
Climate change, they agree, is the major threat to public health. Here is an excerpt: there will be nothing surprising here:
‘The risks to health of increases above 1.5°C are now well established. Indeed, no temperature rise is “safe.” In the past 20 years, heat related mortality among people aged over 65 has increased by more than 50%.Hi gher temperatures have brought increased dehydration and renal function loss, dermatological malignancies, tropical infections, adverse mental health outcomes, pregnancy complications, allergies, and cardiovascular and pulmonary morbidity and mortality. Harms disproportionately affect the most vulnerable, including children, older populations, ethnic minorities, poorer communities, and those with underlying health problems.’ Continue reading
We Should Vaccinate Children in High-income Countries Against COVID-19, Too
Written by Lisa Forsberg, Anthony Skelton, Isra Black
In early September, children in England, Wales and Northern Ireland are set to return to school. (Scottish schoolchildren have already returned.) Most will not be vaccinated, and there will be few, if any, measures in place protecting them from COVID-19 infection. The Joint Committee on Vaccination and Immunisation (JCVI) have belatedly changed their minds about vaccinating 16- and 17-year olds against COVID-19, but they still oppose recommending vaccination for 12-15 year olds. This is despite considerable criticism from public health experts (here, here, and here), and despite the UK’s Medicines and Healthcare products Regulatory Agency (MHRA) declaring COVID-19 vaccines safe and effective for children aged 12 and up—Pfizer/BioNTech in the beginning of June, and Moderna the other week.
In Sweden, children returned to school in the middle of August. As in the UK, children under 16 will be unvaccinated, and there will be few or no protective measures, such as improved ventilation, systematic testing, isolation of confirmed cases, and masking. Like the JCVI in the UK, Sweden’s Folkhälsomyndigheten opposes vaccination against COVID-19 for the under-16s, despite Sweden’s medical regulatory authority, Läkemedelsverket, having approved the Pfizer and Moderna vaccines for children from the age of 12. The European Medicines Agency approved Pfizer and Moderna in May and July respectively, declaring that any risks of vaccine side-effects are outweighed by the benefits for this age group.
Healthcare, Responsibility, and Golden Opportunities
Written by Gabriel De Marco
This blog post is based on a co-authored paper (w. Tom Douglas and Julian Savulescu) recently published in Ethical Theory and Moral Practice.
When it comes to determining how healthcare resources should be allocated, there are many factors that could—and perhaps should—be taken into account. One such factor is a patient’s responsibility for his or her illness, or for the behavior that caused it; e.g. whether a lifetime smoker is responsible for developing his lung cancer, or whether someone is responsible for heart disease on the basis of having an unhealthy diet. Policies that take responsibility for the unhealthy lifestyle or its outcomes into account—responsibility-sensitive policies, or RSPs, for short—have been a matter of debate for some time.
Compromising On the Right Not to Know?
Written by Ben Davies
Personal autonomy is the guiding light of contemporary clinical and research practice, at least in the UK. Whether someone is a potential participant in a research trial, or a patient being treated by a medical professional, the gold standard, violated only in extremis, is that they should decide for themselves whether to go ahead with a particular intervention, on the basis of as much relevant information as possible.
Roger Crisp recently discussed Professor Gopal Sreenivasan’s New Cross seminar, which argued against a requirement for informational disclosure in consenting to research participation. Sreenivasan’s argument was, at least in its first part, based on a straightforward appeal to autonomy: if autonomy is what matters most, I should have the right to autonomously refuse information.
I have previously outlined a related argument in a clinical context, in which I sought to undermine arguments against a putative ‘Right Not to Know’ that are themselves based in autonomy. In brief, my argument is, firstly, that a decision can itself be autonomous without promoting the agent’s future or overall autonomy and, second, that even if there is an autonomy-based moral duty to hear relevant information (as scholars such as Rosamond Rhodes argue), we can still have a right that people not force us to hear such information.
In a recent paper, Julian Savulescu and I go further into the details of the Right Not to Know, setting out the scope for a degree of compromise between the two central camps.
Post-Normal Challenges of Covid
Written by Stephen Rainey
How to manage the inevitable disruptions to life brought about by the emergence of a viral pandemic – a question that for many seemed remote has now had us all preoccupied for well over a year. With our just-published article, entitled The Post-Normal Challenges of COVID-19: Constructing Effective and Legitimate Responses, in the Journal Science and Public Policy, Maru Mormina, Sapfo Lignou, Joseph Nguyen, Paula Larsson and I set out to investigate some of the perplexing difficulties especially relating to effectiveness and legitimacy. We examine these in the light of pandemics as wicked problems and lay out how ‘post-normal science’ can contribute to a sound pandemic response.
In any pandemic response, the measures undertaken by authorities must effective in the sense of actually addressing the viral threats. A strategy that didn’t slow the rate of viral spread, for instance, wouldn’t work and for that reason would be due criticism. The concept of legitimacy is one perhaps less easy to cash out. In any pandemic response, the measures undertaken by authorities must be legitimate in the sense of fairly and justifiably constraining liberties enjoyed prior to the viral outbreak. A strategy that placed undue or disproportionate burdens on societal sub-groups, for instance, wouldn’t be legitimate and for that reason would be due criticism. For effectiveness in a medical crisis particularly, science is an essential element of any response. Continue reading
Affiliations
Recent Comments