The smiles may be fake but synthetic media's potential is real

From providing customer intelligence while abiding with privacy regulation to engaging customers with ads filmed without actual actors or locations, deepfakes could prove a valuable and cost-effective tool for accelerating your business

Royi Benyossef 10:3610.08.20
The term synthetic media, more commonly known as deepfakes, refers to content that is generated or modified through algorithmic means. Examples can vary in complexity from the manipulation of a person’s image or voice to personalized digital experiences generated from scratch.


As the U.S. general election in November draws near, concerns that campaigns or outside actors may seek to weaponize deep fakes of the candidates are growing.


In the hands of non-malicious actors, however, synthetic media can help us to create new kinds of content and services much more quickly and on a far bigger scale, using up fewer resources than ever before.


Overcoming the content creation barriers


The primary challenge in creating content is a limited supply of valuable resources such as time, money, and talent.


Actors, crew, and a location are all resource-heavy investments traditionally required for creating a video commercial, for example. But, what if we didn’t need actual humans or a physical location?


With the right data, technology, and computing power, synthetic media means we can create, with very little effort compared to traditional means, new content that can even be taken a step further and personalized for the individual viewer.


Tel Aviv-based Hour One AI Ltd. is one example of a company that is already working with clients to generate synthetic characters for commercials and ads that are based on real people, all with the goal of increasing sales and engagement with a lower price tag.


Customer service and experience


Interacting with bots has become a common, though, admittedly still clunky, experience in modern customer service.


Synthetic media could vastly improve this experience by layering a realistic avatar that mimics human interaction over the bot, making the service feel a bit more personable.


Paris-based startup Alta Voce, for example, shapes speech and voice in real-time, to improve communication between customers and virtual support agents.


Personalization and privacy


Providing a personalized experience depends on your algorithm’s ability to collect and process information. However, we are starting to see a push for stronger regulations against surveillance, with the European Union’s General Data Protection Regulation (GDPR) and the California Consumer Privacy Act of 2018 (CCPA) the most prominent examples.


These laws make personalization far more challenging. One example that recently made headlines is the collection of in-store video footage by retailers, who used it not only for security purposes but for observing customers and obtaining intelligence on their behavior to help drive better decisions and sales.


Under the GDPR, this type of data collection can quickly land businesses in hot water. By utilizing synthetic media, however, companies can anonymize the identity of filmed customers to ensure privacy while holding onto valuable data.


One startup already active in this field is Tel Aviv-based D-ID, incorporated as De-Identification Ltd. D-ID’s technology processes video content, creating a new version that cannot be used for facial identification. This pseudonymization of the videos allows retailers to still reap most of the business benefits of surveillance, short of identifying a specific client.


Technological advances are pushing progress in synthetic media


Several technological advancements in recent years have made synthetic media financially viable at scale. The first two areas of significant improvement have been in generative adversarial networks (GAN) and style transfer. These techniques allow creators to produce synthetic media that automatically responds to the user, provided they have enough examples and computing power. Improved cloud availability is already a factor in helping to improve content creators’ ability to utilize GAN and style transfer.


There are, however, still challenges that will need to be addressed through innovation when it comes to increasing computing power, decreasing the datasets needed for personalization, and optimizing costs. Like the need for advances in anonymization, bringing solutions to these challenges will spell opportunity for companies that are willing to take up the challenge.


Royi Benyossef is an investor at Samsung NEXT TLV, a venture investment arm of Samsung.



Royi Benyossef. Photo: Areiel Efron Royi Benyossef. Photo: Areiel Efron