Facial and Emotion Recognition: Where Innovation Meets Regulation

Unsplash / Collage by Robert Handrow.

1. Human face is the new oil?

The market was always about the creation of value. About the production of goods and offering of services in exchange for money. Now, the word ‘good’ usually serves to describe things, whether exhaustible or inexhaustible, tangible or intangible, private or public. But the one thing they have (or maybe had) in common is that they were separate from humans. However, this may no longer be the case. The first step towards broadening the meaning of this term happened as soon as we started treating data as a good that could be commercialized and used to build a business model. (Thanks Google.) And now we are rapidly entering yet another era, where our faces are also up for sale. This happens through the commercialization of two key features of human faces. One which allows it to uniquely identify a person, and the other which indicates what a person is feeling and experiencing at the moment.

1. Reading emotions as a business model

Emotion recognition predominantly relies on micro-expressions, or “short, involuntary facial expression[s]”,¹ supposedly giving away our true feelings even when we are actively trying to hide them.² And while the idea that one can deduce inner emotional states by simply looking at a person may not be as surprising (we might have to thank Tim Roth for popularizing the concept in the series ‘Lie to Me’), the practice of automatically ‘reading’ people’s faces is rapidly entering our everyday lives at scale. This scale can come as a bit of a surprise.

For instance, let’s consider Affectiva, a company established in 2009 and recently acquired by the Swedish tech magnate “to create a transatlantic AI juggernaut”.³ The (now expanded) Smart Eye Group does everything from designing Driver Monitoring Systems, over Interior Sensing for improving the driving experience, up to Media Analytics, which is where we will linger on for a second. Affectiva’s Media Analytics service is advertised as a way of “[optimizing] content and media spend by measuring consumer emotional responses to videos, ads, movies and TV shows – unobtrusively and at scale.”⁴ Now, as much as any single person may want to see an ultimately enjoyable and ‘perfect’ TV show, there are certain lines, such as a company selling analyses of your facial expressions collected e.g. while streaming content online,⁵ which probably should not be crossed. Even beyond legal implications, one should seriously consider the wider implications of today’s culture of ever-increasing personalization of content. Such technologies, leading towards maximization of engagement and effectively gluing people to their screens, should be scrutinized and examined for efficient safeguards, before becoming a regular part of any average Joe’s life.

And there is more. For instance, tools like Cigna StressWave Test,⁶ allow both individuals and companies to measure people’s stress levels by analyzing 1-minute voice recording samples. Based on the samples, specific recommendations as well as stress care plans are offered to the test subjects. Now besides the fact that the technology behind the test results is shaky and unreliable at best,⁷ the fact remains that they are analyzing data to infer the emotional states of people and use them to gain profit.

And, sadly enough, it’s not just the commercial entities that are trying to use our emotions and reaction for personal profit. For instance, in July 2023 the Italian city of Naples hired CONFORM to create a short, animated film implementing MorphCast Emotion AI.⁸ The system allows the recording and measuring of emotional reactions of viewers in order to provide personalized recommendations of other multimedia content, like 360° virtual tours, short films, documentaries, feature films etc., based on the “detected emotional state”. Although in this case at least the city isn’t extracting any direct profit from the use of this technology, what and how MorphCast processes the relevant data (other than claiming they do it responsibly)⁹ can be anyone’s guess.

2. When your online and offline lives merge

On the other side of the creepiness coin, lays yet another major issue. Namely, it is not just our emotions which are ‘up for grabs’. It is also our identities. This again may be nothing new. After all, marketing cookies deployed to create our online identities and sell our personal preferences to the highest bidder are a thing coming out of fashion, rather than the new hype. And while it remains to be seen how online advertisers will get out of that particular ‘pickle’, it seems that the offline world is also under attack. Or maybe, at the very least, under analysis.

To name one instance where on- and off-line personas are already being merged, we are going to jump over the pond and onto US grounds. There, a decade-old California-based fashion brand ‘Ruti’ very early on realized how unfair life can be out there in the harsh real world, where you have absolutely no idea what a person entering the store might like. And this particular brand decided to take destiny into its own hands and do something about it. The brand owner and designer proclaimed in an interview all the way back in 2019 that “retail with no technology is not going to work.”¹⁰ In the same article, the brand introduced its new on-site technology, helping them remain competitive.¹¹ Namely, they started recording and scanning the face of every customer entering the store. Following the customer’s approval, the photos are then uploaded to Ruti’s CRM system, which uses facial recognition technology to identify the customer and access their online profile. Then their online personality (likes, dislikes, size and preferences) is merged with actual in-store and online purchase history, as well as items previously tried on, to offer the ultimate personalized shopping experience.¹² Good luck saying ‘no’ to your dream sweater, in your favourite colour, and ideal size, which you probably do not need.

And it gets a lot creepier than that. While Ruti, at least at some point (albeit slightly too late to be GDPR compliant for instance), asks for the customer’s consent to upload and process the recordings, other technologies are designed in a way that basically makes collecting consent impossible. Let’s take Vuzix facial recognition glasses as an example.¹³ The glasses support a facial recognition app, which when turned on, automatically detects and recognizes all faces in the camera frame, providing a name and the system’s estimated confidence level in the accuracy of the recognition.¹⁴ According to an interview with the CEO of Clearview, Ton That, the glasses, when connected with Clearview’s database, could even provide the profession of the recognized person, as well as references and links to photos, posts and social media accounts of the identified person. Leaving aside law enforcement and the prospects of how it may feel to live in China or any other surveillance state,¹⁵ the glasses have their commercial use as well and that nowhere other than in Europe. Namely, Vuzix already signed a distribution deal with an Austria-based company called Barotec in December 2021.¹⁶ The glasses are still available on the company’s webstore and advertised as a tool for healthcare and remote assistance,¹⁷ but from there it takes only a bit of creativity and downloading the facial recognition app and one enters a completely different ball game. Actually, you might not even need to be that creative. You can just take one look at the Dubai-based company called RECFACES,¹⁸ and you will simply be overflowing with ideas as to how to commercialize all these recognized faces. From improved communication, over personalized media content or services, up to security checks, nothing is more than one app and a few scans away.

2. Legal backdrop

The global commercial, over-production, and over-consumption culture, especially when paired with ever more personalized goods and services, is definitely not taking a break any time soon. And while some may be very happy with their lives being ever more personalized,¹⁹ others may be less ecstatic about the current state of affairs and future prospects. Leaving personal preferences aside, there are still some legal implications the use of such technologies necessarily brings to the fore, that ought to be considered.

What certainly cannot be overlooked in this regard is the GDPR, which (at least for the time being) should protect European customers from overly invasive technologies. However, when it comes to distributing glasses supporting facial recognition apps, car systems automatically adjusting the temperature or creating the next perfectly addictive TV series, this apparently does not apply. For instance, the mentioned Affectiva interior sensors are already used en mass in cars produced and sold in Europe, such as BMW.²⁰ The implemented sensors are portrayed as primarily lifesaving tools, helping the driver refocus on the road or even initiate preventive action when the driver doesn’t. Thus avoiding at least some fatal accidents.²¹ Aside from saving lives, however, the cars can apparently do a lot more. For instance, BMW, “with user opt-in and consent, of course,” uses the collected data to “provide recommendations, or help improve your condition.”²² The senior machine learning engineer of the BMW Group, Sean Batir, also provided some examples of how their cars may do that. Among other things, he stated they may “[understand] if someone is going through depression, especially if [they] have historical data on [the person].”²³ Such (presumably extensive) driver data would allow your favourite car producer to notice the deviation from your “normal big broad postures to more reticent displays”, which “could be an early warning indicator that [it] could connect to clinical data sources”.²⁴ Disregarding any personal opinions about using a car as your personal therapist, this constellation can hardly be reconciled with the GDPR.

Starting from the fact that we are talking about the processing of sensitive data for which explicit consent is needed. Over the sheer amount of information and the clarity with which these would have to be communicated to the data subject to make the consent effective and lawful. Coupled with the rather unfortunate circumstance that not every single person entering the car will have given their consent in advance. And the fact that (even with valid explicit consent) the person is still being profiled, which implies at the very least the consideration of Article 22, forbidding automated decisions leading to legal or similarly important consequences. Because although the pure deduction that the driver might be depressed as such would not yet trigger the application of the article, if the car would also do something about it to ‘help’ the poor potentially depressed driver, the answer might be very different. In any event, even if all discussed systems are still on the ‘right side of the law’ and only used ‘for good’, the fact remains that they leave a door open for ever broader use purposes of such systems, even if only “with user opt-in and consent, of course.”²⁵

Finally, the analysis does not stop with the GDPR. What should also be considered in this context is the (soon?) incoming AI Act.²⁶ Because, even if the systems are reconciled with the obligations for sensitive data processing contained in the GDPR, the AI Act (once in force) brings with it a whole different and rather extensive set of obligations. This is predominantly the case because, according to Annex III of the current draft (which is also likely to stay the same at least in this respect), any biometric and emotion recognition system is a high-risk system. And developers of high-risk systems have basically two whole Chapters (Chapters 2 and 3) of the Act devoted to their obligations. Starting from disclosure and transparency obligations going beyond those of disclosing the processing to the data subject as demanded by the GDPR.²⁷ Up to implementing effective human oversight²⁸ and running efficient risk and quality management.²⁹ On the other hand, according to the current draft, systems enabling real-time biometric identification, as well as emotion recognition systems used in the area of law enforcement, border management, workplace and education institutions might be banned altogether.³⁰

Of course, the AI Act is still lying somewhere in the unforeseeable future and facial and emotion recognition remain one of its most contentious discussion points. However, taking all of the above into consideration, lawfully implementing these systems in the EU ought to be quite a challenge anyway. Unless we all tacitly agree that, since car safety is very important and the stores can already track us online, it would only be fair to let cars monitor our every blink and offline stores identify every potential customer to ‘stay in the game’. Whether this is a world any single person would want to live in, then remains open for (mostly philosophical) discussion.

3. Final thoughts

Like with many other advanced technologies, there are enthusiasts and there are sceptics. There are devotees, who don’t mind giving up their personal data if it means their car is going to ‘hit’ them with the perfect song every single time and if their personal shopping assistant would know them better than they know themselves. However, maybe certain things should not be left to the market, maybe certain things should be clearly guard-railed before they escalate, and when it comes to biometrics, things are escalating fast. So the facial biometrics market is expected to grow at an annual rate of 17.2%, reaching an approximate worth of 8.5 billion USD by 2025 from the 3.8 billion USD worth it had in 2020. And finally, the fact remains that the European Union (for better or for worse) already has some rather strict privacy regulations that are virtually unreconcilable with constant recording of people’s faces, even if it serves the purpose of making them feel better or more focused on the road. As is often the case with technology, however, it will probably take a while before the problems come under scrutiny by the general public. And an even longer time before we find a way to reconcile them with existing (as well as upcoming) regulations.

¹Definition of a ‚micro expression‘ from the macmillandictionary open dictionary. https://www.macmillandictionary.com/dictionary/british/micro-expression [accessed on the 26 of May 2023].

²Thuong-Khanh Tran et al., Micro-expression spotting: A new benchmark, Neurocomputing Vol. 443, (2021) p 356 https://www.sciencedirect.com/science/article/pii/S092523122100268X [accessed on the 26 of May 2023].

³Our Origins – About Affectiva, Affectiva, https://www.affectiva.com/about-affectiva/ [accessed on the 26 of May 2023].

⁴Affectiva Media Analytics, Affectiva, https://go.affectiva.com/affdex-for-market-research [accessed on the 26 of May 2023].

⁵Gabi Zijderveld, ‘The World’s Largest Emotion Database: 5.3 Million Faces and Counting’, Affectiva, https://blog.affectiva.com/the-worlds-largest-emotion-database-5.3-million-faces-and-counting [accessed on the 2 of June 2023].

⁶https://www.cignaglobal.com/stress-care/individuals/voice-tool [accessed on the 28 of November 2023].

⁷B. A.Yawer, J. Liss  and V. Berisha, Reliability and validity of a widely-available AI tool for assessment of stress based on speech, Scientific Reports 13:20224 (2023) https://doi.org/10.1038/s41598-023-47153-1 [accessed on the 28 of November 2023].

⁸ANEMOTION | Animated short film by CONFORM, MorphCast Client Side Emotion AI Software, https://www.morphcast.com/showcase/anemotion-animated-short-film-by-conform/ [accessed on the 28 of November 2023].

⁹Guidelines and Policies for responsible use of Emotion AI, https://www.morphcast.com/responsible-guidelines/ [accessed on the 28 of November 2023].

¹⁰Katie, Richards, ‘Retail without technology is not going to work’: Ruti is using facial recognition technology in stores to boost sales’, 23 of September 2019, https://www.glossy.co/fashion/retail-without-technology-is-not-going-to-work-ruti-is-using-facial-recognition-technology-in-stores-to-boost-sales/ [accessed on the 26 of May 2023].

¹¹Katie, Richards, ‘Retail without technology is not going to work’: Ruti is using facial recognition technology in stores to boost sales’, 23 of September 2019, https://www.glossy.co/fashion/retail-without-technology-is-not-going-to-work-ruti-is-using-facial-recognition-technology-in-stores-to-boost-sales/ [accessed on the 26 of May 2023].

¹²Glenn Taylor, ‘Ruti Overcomes In-Store Personalization Challenge With Opt-In Facial Recognition’, 3 of January 2020, retail TouchPoints, https://www.retailtouchpoints.com/topics/customer-experience/ruti-overcomes-in-store-personalization-challenge-with-opt-in-facial-recognition [accessed on the 26 of May 2023].

¹³https://www.vuzix.com/blogs/vuzix-blog/ai-powered-face-recognition-on-the-vuzix-blade [accessed on the 28 of November 2023].

¹⁴You can see a demo on YouTube: https://www.youtube.com/watch?v=sAFWt-oqDD0 [accessed on the 28 of November 2023].

¹⁵J. Vincent, Chinese police are using facial recognition sunglasses to track citizens, The Verge, 8 of February 2018, https://www.theverge.com/2018/2/8/16990030/china-facial-recognition-sunglasses-surveillance [accessed on the 28 of November 2023].

¹⁶F. Hersey, New worldwide deals, facial recognition integration for Vuzix smart glasses, BIOMETRICUPDATE.COM, 4 of January 2022, https://www.biometricupdate.com/202201/new-worldwide-deals-facial-recognition-integration-for-vuzix-smart-glasses [accessed on the 28 of November 2023].

¹⁷ https://www.barcotec.at/en/mobile-devices/industrial-wearables/vuzix-m400 [accessed on the 28 of November 2023].

¹⁸https://recfaces.com/products [accessed on the 28 of November 2023].

¹⁹See, for example, ‘Gen Z and Millennial Consumers are Changing Loyalty—Open to Behavior Tracking, Paid Loyalty, and Deeper Brand Engagement Through Technology’, bond, https://info.bondbrandloyalty.com/press-release-loyaltyreport2018 [accessed on the 26 of May 2023].

²⁰See, for example, Ashley McManus, ‘BMW: How In-Cabin Sensing Helps Build the Ultimate In-Vehicle Experience’, 24 of July 2020, Affectiva, https://blog.affectiva.com/bmw-how-in-cabin-sensing-helps-build-the-ultimate-in-vehicle-experience (accessed on the 31 of May 2023).

²¹Polestar 3 with Smart Eye’s driver monitoring system on show at CES.

²²Ashley McManus, ‘BMW: How In-Cabin Sensing Helps Build the Ultimate In-Vehicle Experience’.

²³Id.

²⁴Id.

²⁵Id.

²⁶This Article was written based on the ‘Proposal for a Regulation of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts - General approach’, Brussels, 25 November 2022 (OR. en) 14954/22, https://artificialintelligenceact.eu/wp-content/uploads/2022/12/AIA-%E2%80%93-CZ-%E2%80%93-General-Approach-25-Nov-22.pdf (accessed on the 31 of May 2023) and the ‘DRAFT Compromise Amendments on the ‘Draft Report Proposal for a regulation of the European Parliament and of the Council on harmonised rules on Artificial Intelligence (Artificial Intelligence Act) and amending certain Union Legislative Acts’ (COM(2021)0206 – C9 0146/2021 – 2021/0106(COD)), 16 May 2023, https://www.europarl.europa.eu/resources/library/media/20230516RES90302/20230516RES90302.pdf [accessed on the 31 of May 2023].

²⁷See Article 16 (c) and (j) of the ‘Proposal for a Regulation of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts - General approach’ and Article 16 (ac), (c) and (j) of the Draft Report Proposal for a regulation of the European Parliament and of the Council on harmonised rules on Artificial Intelligence (Artificial Intelligence Act) and amending certain Union Legislative Acts’.

²⁸See Article 14 of the ‘Proposal for a Regulation of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts - General approach’.

²⁹See Articles 9, 16 (b) and (e), and 17 of the ‘Proposal for a Regulation of the European Parliament and of the Council laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts - General approach’ and of the Draft Report Proposal for a regulation of the European Parliament and of the Council on harmonised rules on Artificial Intelligence (Artificial Intelligence Act) and amending certain Union Legislative Acts’ respectively.

³⁰See, Amendments 215 to 231 adopted by the European Parliament on 14 June 2023 on the proposal for a regulation of the European Parliament and of the Council on laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) and amending certain Union legislative acts (COM(2021)0206 – C9-0146/2021 – 2021/0106(COD)), available at: https://www.europarl.europa.eu/doceo/document/TA-9-2023-0236_EN.html.