Human-Centric AI: The Missing Piece of the Debate on AI Networks

A few weeks ago, Seth Rosenberg, a Partner at Greylock, one of the most prominent VCs in Silicon Valley, wrote a fascinating and insightful article about AI-driven networks and marketplaces. The article provides numerous insights that instill much optimism about the opportunities with generative AI. Yet, it misses one key point: humans, not AI, should be at the center of this once-in-a-lifetime technological transformation. Let me explain.

Towards AI-Powered Networks and Marketplaces

In the first part of this article, Mr. Seth Rosenberg provided an overview of the evolution of networks and marketplaces. He noted that currently, we are at the cusp of a new arms race to build the next AI-first network. More specifically, during the past decades, we moved from networks that connect people to algorithms that connect people to content. Now, we’re moving to algorithms that replace people.

Here’s the overview of these network models that many of us have witnessed:

1. Pre-AI network → people connected with people and businesses
2. AI-powered network → people posting and consuming content for/by the algorithm
3. AI-only network → AI creating personalized content for each person

While this representation seems compelling, in my humble opinion, it misses one crucial phenomenon: the emergence of a new, human-centric AI network.

My proposal to include a “human-centric” approach in the discussion about the impact of AI on networks.

AI-centric Networks: for Enterprises

Let’s pause for a moment and dive deeper into the concepts of “AI-powered” and “AI-only” networks.

Mr. S. Rosenberg rightly observes that AI is affecting all platforms, all services, and industry verticals. Each and every app we use today is likely to have a generative AI component, which means that you can talk to your service provider and interact in new liberating ways. Generative AI technologies open new ways to deliver products and services and spur innovation even in the oldest sectors that were really hard to optimize.

Furthermore, we can see how AI will unlock new opportunities in marketplaces that connect two different groups of users (think of job search sites such as Deel, matching platforms such as AirBnb or Uber, online shopping or gaming). Incumbent technology giants are racing to figure out how to implement ML and AI technologies by incorporating AI in every functionality that their platforms currently operate.

Probably not everything should be “AI-ified”, and quite likely there are myriads of areas where incumbents are unable to build new services. That’s where the greatest opportunities for new players exist. Let’s hope that the period of trial and error is relatively short.

The two AI-centric network models outlined above essentially correspond to the product-centric approach of enterprises. Here’s what is true for product-centric AI networks today:

• Individual users are the nodes in the network.
• Each individual user has an account through which they authenticate their identities, subscription levels, and categorize them into different segments. Let’s take Airbnb as an example: a high-paying customer who can afford a luxury room, a median customer whose payment history shows the willingness to pay $200–250 per night, a budget traveler, etc.
• User experience is designed to capture “median/average” users (there is one app for all users).
• Users give away their personal details to each product/service provider for an illusionary expectation to get more personalized offerings. We pay with our own data. Ultimately, we are the product for service providers.

Human-centric AI: for Individuals

As we enter the new AI-driven era, one of the major transformations occurs in the shift from a product-centric to a human-centric approach to data and AI.

In this new human-centric model, each individual is at the center of the hub of interactions, and AI-powered applications come to the user, and run in the user’s own data environment (e.g., on top of users’ consolidated data, or in the user’s phone).

There are myriads of large companies and startups around the world that are building personal AI assistants (e.g., ChatGPT, AI doctors, AI coaches, AI financial advisors…).

What technological, UX and business infrastructure is needed to make those AI assistants truly personal?

You need to bring such and “AI-only” or “AI-centric” product to the individual user. We need to turn the tables and put the individual human user in the center of interactions with such AIs — otherwise, they will all be “generalists”.

Let’s take “personal AI travel assistant” as an example. How can we make “AI travel planning assistants” truly personal? What data infrastructure is needed to build a truly personal AI assistant?

In very practical terms, to personalize AI, we need to tap into the combined data of the user. For Personal travel AI assistant, we most likely need:

(i) Individual’s previous bookings on Airbnb platform,

(ii) individual’s starred locations on Google Maps (places I liked and places I marked as to be visited in the future);

(iii) Access to individual’s personal and work calendars, and

(iv) Individual’s payment history.

In the human-centric data and AI infrastructure, we put the individual in the center of the network, and applications come to the user.

New Data Market for Personal AIs

There are three major driving forces that have contributed to the emergence of this new market for personal AIs: (i) technological progress; (ii) increasing demand among businesses and consumers; and (iii) new regulations that push forward the opening of new human-centric approaches to data and AI.

Tech. In the past decade, we have witnessed the rapid advancement of machine learning, data processing, cloud computing, edge computing, and AI technologies. Naturally, as many more sensors and data are generated in the real world, these technological developments aim to unlock the value of data.

Market demand. in the current highly centralized verticals, much of the data that is generated by users — individual consumers as well as business users — remains locked away in silos and remains unused. There is a huge demand to unlock this data and use to create new services. Consumers too, are eager to get more value from their own consolidated data.

Regulatory. Since the adoption of the GDPR, we have witnessed a huge influx of new rules and regulatory frameworks that aim to make data available to individuals and businesses who generate data through the use of various services and products. The EU Data Act is noteworthy in this regard because it will force all IoT device makers to make all data available to users.

Human-centric Approach in the EU AI Act

The EU AI Act proposal is a particularly interesting document because it introduces the human-centric approach as the foundational cornerstone of the way the EU regulates AI systems. Recital 4 of the EU AI Act provides that:

“As a pre-requisite, artificial intelligence should be a human-centric technology. It should not substitute human autonomy or assume the loss of individual freedom and should primarily serve the needs of the society and the common good.”

According to the EU legislator, such a human-centric approach to AI is rooted in the values of the Charter of Fundamental Rights of the EU and the values on which the Union is founded, including the protection of fundamental rights, human agency and oversight, technical robustness and safety, privacy and data governance, transparency, non-discrimination and fairness and societal and environmental wellbeing.

Article 4 of the EU AI Act mandates all operators to make their best efforts to develop and use AI systems that promote a coherent human-centric European approach to ethical and trustworthy AI.

The EU regulator is technology agnostic: it does not provide any clear technological solutions for how this vision of human-centric AI systems should look like, I believe that in practical terms it means a new data ecosystem where individuals have their own personal data environments where (a) they can collect and consolidate data from different platforms and services; and (b) where AI-powered apps and services come to the user, and run on top of user’s own data, privately.

Paths Forward

Returning to the vision of how the networks are progressing over time, we can see that we are at the cusp of major transformation: new data and AI models are forcing us to rethink the existing technological frameworks and open new avenues for innovation.

Does that mean a human-centric approach to AI competes with enterprise/product/AI-centric approaches?

Definitely not: there are myriads instances where the currently prevailing product/AI-centric models work and will remain in place. At the same time, we can see how a new human-centric approach to data and AI becomes crucial in building truly personal services.

Human-centric approach also offers some hope, that it will deliver more privacy towards how individuals’ data is being processed. As alluded to before, in the human-centric model, individual users’ data is private-by-default. In this new data infrastructure, developers and service providers do not have access to their users’ data; but if they want to gain access to certain segments of users’ data, they need to justify such a request.

In a world that’s rapidly becoming more algorithmic and data-driven, the human-centric AI paradigm brings back to light individual human beings. It recognizes that every individual, with their unique experiences, aspirations, and needs, is irreplaceable.

It forces us to place individuals at the forefront of the AI revolution: at the end of the day, AI is the tool that is supposed to augment human abilities and help us live happier and healthier lives. Let’s ensure that technology, no matter how advanced, always resonates with the heartbeat of humanity, and that AI serves our personal needs and aspirations.

Thanks for reading! I hope you find this post insightful. Feel free to reach out to me if you’d like to discuss this further!