Optimization of what? For-profit health apps as manipulative digital environments

Conversion refers to the principle of turning users into profitable users.

Because of the limited space at my disposal, I will not explicitly engage with related challenges posed by health apps, such as challenges of informational privacy (Floridi 2005; Patterson 2013; Lanzing 2016) and surveillance (Lupton 2012).

Thanks to Thomas Nys for pointing this out to me. See also Nys (2016).

See Fahy et al. (2018) for a more elaborate analysis of Apple’s and Google’s developer platforms and the different kinds of monetization strategies apps can use.

Apple, ‘Choosing a Business Model’: https://​developer.​apple.​com/​app-store/​business-models/​.

Apple, ‘Choosing a Business Model’: https://​developer.​apple.​com/​app-store/​business-models/​.

Google, ‘Earn more revenue with the right monetization options’: https://​developer.​android.​com/​distribute/​best-practices/​earn/​monetization-options.

Google, ‘Earn more revenue with the right monetization options’: https://​developer.​android.​com/​distribute/​best-practices/​earn/​monetization-options.

In the case of health apps this means advertising that is deliberately designed to ‘look and feel’ like health content, thereby obfuscating to the user the true commercial nature of the content.

This option is not explicitly mentioned by Apple or Google. It is, however, a serious option. MyFitnessPal, an immensely popular calorie counting app with millions of users, is a good example. On its jobs page, MyFitnessPal announces that “MyFitnessPal has the largest database of human eating habits in the world. The opportunities for a data scientist here are almost endless.” (https://​www.​myfitnesspal.​com/​jobs) Forbes reports that health care providers and researchers can access the database when they enter into a “formal partnership” with MyFitnessPal (Olson 2014).

Apple, ‘Using the Freemium Model’: https://​developer.​apple.​com/​app-store/​freemium-business-model/​.

Google, ‘Improve Conversion Using Google Analytics for Firebase’: https://​developer.​android.​com/​distribute/​best-practices/​earn/​improve-conversions.

Google, ‘Improve Conversion Using Google Analytics for Firebase’: https://​developer.​android.​com/​distribute/​best-practices/​earn/​improve-conversions.

Google, ‘Improve Conversion Using Google Analytics for Firebase’: https://​developer.​android.​com/​distribute/​best-practices/​earn/​improve-conversions.

Headspace, ‘About Headspace’: https://​www.​headspace.​com/​about-us.

This lack of access to the (business) operations of Big Tech companies could of course be criticized for a variety of reasons. For example, it makes it harder for investigative journalists and academics to scrutinize the practices of these companies. The same holds for policymakers and regulators who often have a hard time gaining access to Big Tech companies.

Headspace, ‘Senior Data Analyst,’ job description that has since been removed, screenshot available here: https://​imgur.​com/​a/​qtSe4Ii.

Headspace, ‘VP of Growth,’ job description that has since been removed, screenshot available here: https://​imgur.​com/​a/​qtSe4Ii.

Headspace, ‘VP of Growth,’ job description that has since been removed, screenshot available here: https://​imgur.​com/​a/​qtSe4Ii.

There is a rich literature on Foucauldian biopower and health and the role (digital) technologies can play in the exercise of biopower (e.g., Foucault 1975; Armstrong 1995; Petersen and Bunton 1997; Casper and Morrison 2010; Lupton 2012; Mayes 2015; Ajana 2017; Fotopoulou and O’Riordan 2017; Sanders 2017). Although this literature provides interesting and promising perspectives for my research, I do not have enough space at my disposal in this article to incorporate this complex literature into my argument.

NativeAdBuzz, ‘This Health and Wellness Boom Has Been Building for Years… And It’s Finally About to ERUPT (Urgent: Your Free VIP Christmas Gift Has Arrived)’: http://​www.​nativeadbuzz.​com/​blog/​this-health-and-wellness-boom-has-been-building-for-years-and-its-finally-about-to-erupt-urgent-your-free-vip-christmas-gift-has-arrived/​.

Fitbit, ‘Why Fitbit’: https://​www.​fitbit.​com/​whyfitbit.

MyFitnessPal, ‘These Playlists Were Built to Make You Better’: https://​blog.​myfitnesspal.​com/​these-playlists-were-built-to-make-you-better/​.

MyFitnessPal, ‘These On-Ear Headphones Can Actually Withstand Your Workouts’: https://​blog.​myfitnesspal.​com/​these-on-ear-headphones-can-actually-withstand-your-workouts/​.

MyFitnessPal, ‘Why and How You Should Nix an Alarm Clock’: https://​blog.​myfitnesspal.​com/​why-and-how-you-should-nix-an-alarm-clock/​.

MyFitnessPal, ‘A Day in the Life of a Yoga Teacher’: https://​blog.​myfitnesspal.​com/​day-life-yoga-teacher/​.

MyFitnessPal, ‘How a Nutritionist Spends $50 at Whole Foods’: https://​blog.​myfitnesspal.​com/​how-a-nutritionist-spends-50-at-whole-foods/​.

Contrary to authors like Buss (2005) and Wood (2014) who treat manipulation as a non-moralized term.

Susser et al. (2019a) speak of ‘vulnerabilities.’ I rather use the looser term ‘exploitable characteristics of a person’ because the term ‘vulnerabilities’ is sometimes associated (especially in legal discourse) with a fixed set of narrowly defined weakness, such as those that are the result of one’s age (‘the old and the young’) or of one’s psychical or mental infirmities (‘people with medically diagnosed handicaps’).

Rudinow (1978, p. 346, emphasis added) explains that “the manipulator’s behavior is normally either deceptive or predicated on some privileged insight into the personality of his intended manipulee.”.

Mills (2014, p. 138) provides a similar argument, referring to Gorin (2014) and Barnhill (2014): “Both Gorin and Barnhill point out that manipulation does not need to involve deception or covertness; these are not defining features of manipulation necessarily present in all cases of what we could agree to be manipulation. But most manipulators seek to hide the degree to which they are angling to achieve their desired result and would find the success of their project seriously compromised if their manipulative intentions were revealed.”.

Susser et al. (2019a, b) provide a more elaborate discussion on the connection between autonomy and manipulation.

Christman (1991, p. 10) has formulated a popular, somewhat weaker alternative: “What matters is what the agent thinks about the process of coming to have the desire, and whether she resists that process when (or if) given the chance.”

I admit that the resulting picture can feel a bit messy or unclear. Someone can have a manipulative mindset, but, in the end, be drawn to techniques that are not manipulative in nature—An in principle manipulative mindset does not necessarily lead to manipulation. Although I agree that the resulting picture is messy, I see no way to avoid this. The empirical reality of data-driven dynamically adjustable choice architectures simply is very messy. The industry is constantly running (multiple, parallel) experiments to test a plethora of tweaks to their digital choice architectures to test whether those tweaks can successfully shape (patterns of) behavior. In this constant hunt for new behavior influence techniques, some will turn out to be manipulative, while some will turn out to be something else (e.g. coercive).

It could of course still be argued that quite some business-to-consumer practices rise to the level of manipulation. Take, for example, Santilli (1983) and Crisp (1987) who argue that nearly all advertising is—at least slightly—manipulative. If one really wants to stretch my concept of manipulation, once could even try to argue that a billboard showing advertising is manipulative. Such a billboard with an advertisement for company X is (1) put up intentionally by company X, (2) with the aim to further the ends of company X without a genuine regard for the interests of the people passing by the billboard, (3) is designed by company X in such a manner that it targets either particular people in the street, or particular desires or fears of people in the street, and (4) the billboard does not explicitly communicate that “company X is trying to target you in such a manner that company X’s earns as much as possible.” Even if we agree that a billboard can, strictly speaking, be interpreted to be manipulative, it does not follow that every instance of manipulation warrants the same level of scrutiny. There is a significant difference between, one the one hand, a billboard that displays one and the same message to every person at a fixed location, and, on the other hand, a digital health environment which builds a relationship with the user over time, offers a continuous communication channel to the user, and can be personalized in real time based on what the continuous experiments tell will leads to maximum engagement. Unlike billboards, digital technologies like the health apps discussed can get to know their users over time and can at any time they see fit (e.g. through push notifications) try to leverage that information to manipulate every user personally.

Headspace, ‘Senior Data Analyst,’ job description that has since been removed, screenshot available here: https://​imgur.​com/​a/​qtSe4Ii.

NativeAdBuzz, ‘This Health and Wellness Boom Has Been Building for Years… And It’s Finally About to ERUPT (Urgent: Your Free VIP Christmas Gift Has Arrived)’: http://​www.​nativeadbuzz.​com/​blog/​this-health-and-wellness-boom-has-been-building-for-years-and-its-finally-about-to-erupt-urgent-your-free-vip-christmas-gift-has-arrived/​.

Already in 1999, Hanson and Kysar used the concept of ‘market manipulation’ to identify such cases, a concept that was later updated by Calo (2014) who spoke of ‘digital market manipulation.’ Calo (2014, p. 1018) noted how “firms will increasingly be able to create suckers, rather than waiting for one to be born.” Spencer (2019, p. 34) has argued in a similar vein that “[r]ather than discovering existing vulnerabilities, marketers could exacerbate or even create vulnerabilities in individual subjects and then exploit those vulnerabilities.”

For example, Headspace was funded through four funding rounds, raising $75 million (https://​www.​crunchbase.​com/​organization/​headspace#section-investors). MyFitnessPal also received funding from venture capitalists (https://​www.​crunchbase.​com/​organization/​myfitnesspal#section-investors) and was later acquired by Under Armour for $475 million (Olson 2015). Fitbit also saw four funding rounds raising $66 million from venture capitalists (https://​www.​crunchbase.​com/​organization/​fitbit#section-investors).

Consider also Culbert et al.’s (2015) meta-analysis of what causes people’s problematic relation to food. They emphasize that especially for perfectly healthy adolescent and young adult females, (digital) media exposure, and more specifically health ideals portrayed in those media, “have all been shown to prospectively predict increased levels of disordered eating cognitions and behaviors (e.g., body dissatisfaction, dieting, bulimic symptoms)” (Culbert et al. 2015, p. 1145).

Here are a few examples. The blog post called ‘A Day in the Life of a Yoga Teacher’ (https://​blog.​myfitnesspal.​com/​day-life-yoga-teacher/​) is, in reality, native advertising for skincare products that are framed as being part of a healthy, mindful life. The blog post called ‘These Playlists Were Built to Make You Better’ (https://​blog.​myfitnesspal.​com/​these-playlists-were-built-to-make-you-better/​) is native advertising for a particular brand of headphones. The blog post called ‘Why and How You Should Nix an Alarm Clock’ (https://​blog.​myfitnesspal.​com/​why-and-how-you-should-nix-an-alarm-clock/​) is native advertising for a company offering “certified sleep coaches” and for a company selling a wide range of sleep products.