Dear privacy pros,

I hope those of you with school-going children had a wonderful time with them during the recent term break.

A number of articles that caught my attention the past couple of days relate to the artificial intelligence chatbot that Meta recently launched. Meta proudly proclaimed that the BlenderBot 3.0 chatbot can harness information from the internet to build a long-term memory, and thus acquire knowledge over time based on prior information or past ideas.

In one instance when asked for its opinion on Meta co-founder and CEO Mark Zuckerberg, the chatbot unabashedly described him as “a bad person” and “too creepy and manipulative.” Its responses also allegedly stated that Meta’s business practices are “not always ethical” and claimed that Meta “exploits people for money."

It is hardly surprising that groundbreaking technology like BlenderBot 3.0 would have some kinks that need to be ironed out at launch. Given time, the chatbot could indeed live up to the high expectations laid out for it. However, I suspect that in order for machine-learning technology to achieve a semblance of empathy and emulate human qualities in conversation, we will need a fundamentally different modus operandi than that employed by the Big Tech companies thus far.

Machine-learning algorithms train complicated statistical models to predict outcomes of interest in a way similar to how the human brain works: by combining inputs in complex ways to derive “features” of the phenomenon being studied, which in turn determine other features until the output (in the form of the prediction) is eventually determined. Big Tech companies like Meta and Google have managed to overcome the sample complexity required to resolve most simple questions thrown at an AI algorithm without overfitting the model by leveraging the huge amount data they hold.

Unfortunately, there is often very little curation and labeling available for such data, given that these are the flotsams of our collective digital existence, mindlessly generated and freely handed over to these companies as we post updates or pictures on our social media pages, search for the best route to get somewhere, or buy something online. Yet, the supply of labels or contextual understanding to data by humans is one of the ways to improve how an AI algorithm can learn more effectively.  This is how the facial recognition software in the photo gallery of my mobile device can be trained to recognize a person so well over time (with each disambiguation that I perform) that it can recognize that person even when they are wearing a face mask. It is also why we are still asked to perform the seemingly straightforward task of identifying an object in isolated fragments of pictures to prove we are not bots.

A company like Meta could gain some insight into the data they collect by encouraging users to perform simple, organic actions like tagging or reacting with an emoticon, but anything beyond that would likely introduce too much friction or a “creep” factor, potentially hampering usage (and the collection of more data). Some platforms like Amazon’s Mechanical Turk could be used to crowdsource higher-quality, labeled data to train machine-learning algorithms, but a major change in how data is collected may be needed for a steep improvement in AI algorithm output quality. This could include the need to pay users for helping label or provide more information contextualizing the data.

It is highly unlikely that such a momentous shift would be practical under the current business models of Big Tech incumbents like Meta or Google. However, it is entirely possible that one of the many blossoming Web3 startups would be able to build and scale up such a product. If so, we could potentially be working in the metaverse and earning an income just for posting information on social media in the near future, a prospect that my children would no doubt find most enticing. For a peek at what such a world could look like, you can check out the “Data as Labour” chapter in "Radical Markets" by Eric Posner and Glen Weyl.

Before you do that, however, please make sure you check out the rest of the APAC digest below. Happy reading!