The seven pillars of ethical AI in personalised nutrition

That’s the idea behind a white paper published by B2B personalised nutrition platform Qina outlining the principles of an ethical AI framework.

The EU reached an agreement on the AI Act, the world’s first comprehensive law regulating AI, in December 2023. However, conversations on ethics tend to focus on privacy and security. Nutrition is often seen as low risk, despite its wide reach and capacity to influence behaviour and, ultimately, health outcomes.

Building AI solutions on empathy is key, according to Qina CEO Mariette Abrahams, who said she hoped for the white paper, entitled “The ethics of AI at the intersection of nutrition and behaviour change”, to act as an “audit tool”.

“We have a lot of companies starting to employ AI solutions now, but it's more about the implementation and about that level of personalisation, rather than thinking actually, what are we trying to do in terms of behaviour, and who are we impacting, and who are we benefiting in the end?” she said.

And, while cultural changes such as these can take years to embed, she warned that at the speed that tech is moving, time is of the essence.

“I think it's important that at this stage, we need to really raise the bar… [W]e can't look back in even two years’ time and have all these solutions drawing on the same datasets, coming up with the same recommendations, and excluding swathes of people,” she told Vitafoods Insights.

Embedding ethics into personalised nutrition AI

The white paper lays out seven pillars that form the basis of its ethical AI framework. These are “not standalone principles but are integrated within a multi-layered structure”, which considers four “cardinal directions”: health, nutrition, society, and technology.

1. Data

Data is a crucial consideration for all sorts of reasons, not least in terms of where and from whom it is sourced, as many personalised nutrition tools draw on research carried out in specific populations.

“It needs to be transparent [regarding] who was included in that research,” said Abrahams. “You can't recommend a Mediterranean diet and apply it to the Chinese population, for example.”

Meanwhile, companies that make products designed to drive behavioural change need to ensure transparency if they are to build solutions that are truly inclusive, she said.

2. AI system

The white paper says human agency and oversight are “paramount” to ensure personalised nutrition AI systems “are not only built under supervision for expert review, but also remain under expert guidance once deployed for continuous evaluation purposes”.

Abrahams emphasised the importance of transparency not only in terms of how nutrition recommendations are decided upon, but when and how consumer behaviour is being influenced.

“We should also be alerting people on, ‘Hey, we are now kind of nudging you’… helping people to raise their self-awareness and also their levels of self-efficacy,” she added.

3. Human-centric

Affordability and access are “foundational pillars for AI solutions in personalised nutrition, and a determinant to adoption”, the white paper states.

Abrahams added: “Social determinants are one of the biggest influencers of health outcomes. And if we look at where you live, your education level, your digital literacy level, your income level, how much you spend on your health, percentage of your income, access to fresh food and vegetables – those are data points that are not being taken into account in personalised nutrition solutions, in most cases.”

If the endpoint of personalised nutrition is to improve health outcomes and healthspan, “we are missing the biggest chunk of data of information that can drive the personalisation”, she warned.

4. People and planet

While companies broadly acknowledge the importance of sustainability, Abrahams said it tends to be treated as “an area on its own. It's not built into the current AI system”.

Personalised nutrition could act as “the bridge between sustainability and health”, according to the white paper; however, professionals need guidance on how to incorporate sustainable diet principles, for example, in different cultures and regions.

© iStock/t:ipopba

“At the moment they are separate worlds, but they should be integrated,” said Abrahams.

5. Regulation

Personalised nutrition “is unique”, said Abrahams, “because we sit at the intersection of so many other things, like functional food and medical software”.

It means there are multiple grey areas to navigate and sometimes brands fail to realise that they “are dipping their toes in regulated territory”, she said.

“If you want people to increase more vegetables, then that's good, that's great – but if you're going to tell people to have specific supplements or functional foods to manage their hypertension… now you're on medical territory,” she explained.

Abrahams called for enhanced awareness but also for the regulators themselves to step up.

“At the moment, it's a grey zone,” she said.

6. Organisational commitment

Many AI ethics frameworks focus on the AI system itself, with “not enough focus” on the inherent biases that every person working on that system brings with them.

The white paper suggests that organisations ensure diversity among their teams, as well as offer bias detection and reduction training to all staff who work on AI systems.

Abrahams also drew attention to discrepancies in the language used by different teams, arguing for them to agree on a “common language” around solutions that may be impacting communities at a societal level.

7. Education and training

The white paper highlights a “readiness gap” among both consumers and professionals regarding the adoption of AI technologies, which it blames on a lack of understanding of the benefits that AI can offer as well as a perceived threat to job roles.

Abrahams said the “massive shift” taking place in terms of job roles would have an inevitable impact on education.

“It's changing the way we need to think about not only how we train people, but also what we train them in,” she said. “There [are] so many areas we need to think about that if we don't watch out, then AI can easily just pick up any kind of bad science… That's what we need to be very, very mindful of.”