Centralised facial recognition systems that cover the entire population have been identified as mass surveillance technologies, especially when they are used by the state. This is the emerging consensus among human rights activists across the globe. Even in the US, home territory for technological solutionists, there have been strong responses from regulators. In Portland, San Francisco, Oakland, and Boston, the city authorities have been banned from using facial recognition. In contrast, facial recognition on the edge, for example for unlocking phones – where the model is stored and matching happens on the device – are exempt from these bans. This logic seems to apply to other forms of biometrics as well, since unlocking your phone with your finger does not require a centralized national database of fingerprints.
Previously, I have argued that facial recognition on the edge could be scaled up in special cases if trusted intermediaries manage consent effectively. For example, a visually-impaired participant could use facial recognition to identify other participants in the hallways and during the social events of a big conference with the support of the conference organisers. In other words, it is possible to move beyond the simple dichotomy of bans or forbearance towards greater privacy by design, which can also provide for the rights of the disabled. What can the world of internet advertising learn from the regulatory and technological developments around biometrics like finger-prints and facial recognition?
Broadly it can be argued that there are two extremities to the spectrum of business models on the Internet – data maximisation and data minimisation. For example, messaging applications that use end-to-end encryption minimise data extraction and retention by the service provider. Web advertising networks, in contrast, are exemplars of the data maximisation paradigm. Apart from the networks themselves, most other stakeholders are unsatisfied with this approach, especially for internet advertising. For consumers, this is as Shoshana Zuboff calls it, “surveillance capitalism”
, wherein more data is extracted than the amount required to maximize consumer welfare. For advertisers, the numbers reported by the platform cannot be trusted because of widespread fraud. The lack of control for advertisers also results in brand dilution since they have little control over other ads served on the same page and on the content of the page. Publishers or aggregators of attention also feel less empowered because they can’t curate their customer experience. Even for advertising networks, given staff organising and disavowal of rights-infringing practices, the risks of insider attacks and whistle-blowing are growing day-by-day and can severely destroy user trust and credibility.
Is advertising universally detested by all just like the consensus on viruses - digital or otherwise? If this were true, then popular advertisements would not propagate over dark channels like WhatsApp and Telegram. Advertising, like many other aspects of modern culture, can be appreciated for aesthetic reasons and will be accepted by some consumers if there are respectful modalities of delivery. Particularly, if it is the alternative to a paywall when it comes to the consumption of news, sports and entertainment. It is quite clear that the existing rates for paid subscriptions will not work for the next billion users. Is there a data minimisation approach to advertising? After all, advertising today is personalised messaging and, therefore, many of the features are overlapping with traditional customer relationship management softwares.
Neighbourhood fishmongers in Bengaluru send WhatsApp pictures of fresh batches of fish to their full customer list over the mornings of the weekend. Since micro, small and medium businesses (MSMEs) do not invest in CRM softwares - could a messaging software be used for this purpose? The fishmonger's effectiveness could be improved if a client-side AI could help optimise and personalise consumer experience. For example, an AI agent could selectively send a picture of fish in limited stock to the oldest customers, send regional delicacies only to persons of the relevant ethnicity, avoid sending pictures of bloody fresh gills to nouveau fish eaters who only eat skinless and boneless fillet. What if consumers could book an order just by responding to a picture with an emoji? What if the booking transaction could trigger inventory actions? In other words, could encrypted messaging channels be reimagined as a lean customer relationship software for MSMEs?
Traditionally, AI solutions built on the principle of data maximisation assume there are long-term patterns that could be exposed through induction and that more data meant more accurate and all-encompassing predictions. This proposal makes exactly the opposite argument – can data minimisation be used as the central guiding design principle, since the data for this agent would necessarily reside on the entrepreneur’s phone? This reduces the ensuing regulatory burden for the MSME and the messaging platform under national data protection regimes. User control and choice should be the foundation of the design since nobody wants to get periodic form letters featuring corporate speak from their neighbourhood fishmonger. The AI would aim to minimise transactions and communication load and also replicate the errors, the banter and idiosyncratic style of each entrepreneur. After all, recognition of the “from number”, pre-existing relationship and personalisation have been shown to increase the response rate to campaigns.
Client-side AI agents that does not require breaking end-to-end encryption—this is a non-negotiable, since failing to protect the privacy of retail consumption of taboo products and services (alcohol, tobacco and contraceptives) that might result in stigma and discrimination. Small data-based CRMs will require a completely new vision, which will be in stark contrast to classic desktop-centric CRM software which are dominated by forms, reports and dashboards. The intelligence of the AI is in providing the user with actionable “fragments” of information which they can use to grow their business in an authentic and consensual fashion. For example, the AI agent could remind the entrepreneur to take an action or provide training in the following ways, “Today is Vesak, would you like to wish all your Buddhist customers and offer them a special discount on flowers? Yes or No or Later” or “Customer X purchased Product Y today, did she celebrate a religious festival? Yes or No or Later”. All inferences drawn from such interactions with the merchant will remain only on the merchant’s phone.
MSMEs have been using messaging platforms like WhatsApp, WeChat, Line, and KakaoTalk in different parts of the global south, but it is an under researched area in comparison with research focusing on the use of these technologies in the education and health sector. While in these two sectors there are clear benefits for MSMEs and consumers, there are also some risks and harms. For example, researchers from Newcastle University and Great North Pharmacy Research Collaborative studied the role of WhatsApp out-of-hours pharmacy services. Even though there was improved service delivery thanks to professional development and improved communication, there were possible negative impacts to work-life balance. Here again, it is for technology providers to learn from the critique of ahistorical and mono-disciplinary design. Small data-based AI could play an important role to prevent the entrepreneur, her team and her customers from getting overwhelmed by instant message-based communication. The client-side AI agent could learn the working hours of the entrepreneur and suggest auto responders at other times will simultaneously building a follow-up list for when the work-day begins. Through an analysis of read receipts, the AI agent could also identify the best times in the day to target customers with promotional messages.
No technological solution is free of the possibility of negative outcomes and unintended consequences. It is possible that small data-based CRM for MSMEs can be used by criminals selling illegal products. But then, that is already possible with greater inefficiency on any bare-bones end-to-end encrypted messaging service. There is also a possibility that some entrepreneurs will use the CRM for unsolicited commercial messages. This could be countered with more efficient blocking and reporting features. More detailed research would be required to determine if there are other risks. If these risks can be mitigated, then definitely a step towards small data-based CRMs as an alternative and complement to the current approach of ad networks can be considered by technology providers.
Free software developers working as global communities may build the software that I have described above. However, to take advantage of existing network effects, they need the government to enact interoperability mandates to ensure compatibility with proprietary messaging software. However, since free software communities don’t usually have large revenue streams, they will not be able to scale up enforcements of their terms of service, comply with lawful information requests from the police and intelligence agencies or incrementally build necessary friction to create a healthy ecosystem. Wikipedia is, perhaps, the only notable exception from the world of commons-based peer production that has both steady revenues from a large donor base and a sizable community of editors to provide stewardship of its content. Therefore, paradoxically, we need firms that are able to build business models around these small data CRM solutions to emerge and provide competition to those invested in the traditional big data-based advertising paradigm.
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.