Eric Xing and the Age of AI Empowerment

Subscribe | Advertise

Good Morning. Welcome to this Saturday longread.

Our podcast — The Deep View: Conversations — is in its second episode now, which features Dr. Eric Xing, the president of the world’s first AI-only research institution.

I met up with Xing in Abu Dhabi, and we talked for nearly 90 minutes about the realities, pitfalls and promises of artificial intelligence.

You can watch the whole thing below:

The Age of Enlightenment

The invention of the printing press in the mid-1400s sparked a gradual revolution that shaped the construction of the modern world.

Before the printing press, books were hand-written by scribes, where they were read by a very slim slice of the population. The printing press allowed for the cheap distribution of knowledge; it began with religious texts, but quickly spread to philosophical and scientific works. Thoughts and ideas began to spread across Europe as literacy rates increased and more and more authors published their work. 

The path to global revolution was paved by the likes of John Locke and Francis Bacon, those pre-Enlightenment philosophers whose work informed what has since become known as the Age of Enlightenment or the Age of Reason. 

The ideas that informed the American and French revolutions were shaped and spread — by the printing press — during this era. The science that spread during this period is also credited with sparking industrial revolutions around the world. 

That tenet of widespread, intellectual communication enabled by the printing press is the pillar at the center of modern society; the internet and all the applications that are rooted in it are based on the simple idea of enabling faster, more widespread communication. 

And while some have likened artificial intelligence — a term many researchers dislike for being scientifically inaccurate — to the internet in terms of scope and scale of impact, others think it’ll be bigger. Dr. Eric Xing, the president of the Mohamed bin Zayed University of Artificial Intelligence, believes that AI will be as impactful as the printing press. But where the printing press sparked that all-important Age of Enlightenment, Xing believes AI will spark an Age of Empowerment. 

The challenge

The immediate challenge to achieving an age of individual empowerment through AI involves the ethical roots of the technology.

First, the formation of the large language models (LLMs) that have defined the past few years of generative AI, beyond copyright concerns, has given rise to a host of data privacy concerns. A Stanford investigation conducted last year, for example, found hundreds of instances of child sexual abuse material in LAION-5B, a massively popular training set that has been used to train Stability AI and other models. 

There’s also the human element of AI training, in which low-paid workers have to sift through and match text and image pairs, a vital element in the construction of these models. There have been mounting concerns that in many cases, the practice is an abusive one. 

Then, there’s the water and energy requirements behind the training and operation of generative AI models; the GPU-lined data centers that power generative AI have enormous electricity requirements, such that Big Tech companies are moving further from their sustainability goals even as more coal plants are coming back online to feed AI’s voracious energy appetite. There are risks here of grid destabilization on top of active issues of increasing carbon emissions and increasing energy demand at a time when we need to be reducing energy demand and decreasing both emissions and water consumption. 

And that’s just in the construction of these models.

There are massive concerns about the present use of AI in automated decision-making, considering its propensity to produce biased and hallucinated outputs. Researchers have been tracking instances of harm, resulting from algorithmic discrimination, for more than a decade. 

In 2016, for example, ProPublica found that an algorithm used by the U.S. judicial system to predict the likelihood of criminal reoffending was racially biased; it predicted that black defendants had a higher likelihood of reoffending than white defendants. 

The potential offshoots from the simple, unsolved problem of biases and hallucinations are enormous, potentially impacting health, insurance, finance, fraud, education, military and judicial applications. 

It also, as Signal President Meredith Whitaker has said, cannot be separated from the simple fact that the AI we see and know today is a result of an ever-encroaching surveillance business model that has become entrenched as a grudgingly accepted norm over the past 20 years. “AI is basically a way of deriving more power, more revenue, more market reach,” she said. 

In its current form, the broad umbrella of AI is both problematic and ethically fraught. 

The promise of AI rests on a summit that can be reached only by addressing those issues. 

An Age of Empowerment

The link between AI and the printing press, to Xing, is knowledge. 

The printing press, he said, brought knowledge from the few to the many, increasing literacy and allowing the masses to first understand, and then interpret. 

“If you look at (LLMs), for example, it is trained not on one book, it is not trained on one sector of knowledge. It is literally trained on everything ever written into language,” Xing said. “In a sense, you can imagine it is about turning all the libraries in the world into one device and then make it available next to your fingertip.”

This rapid accessibility of knowledge, according to Xing, combined with the autonomous functionality of generative systems, means that not only can people access new levels of information, they can access individualized problem-solving. 

“It is not about just giving you the space to think. It is now giving you the tools to make you a more powerful enabler or actor in whatever occupation that you do,” Xing said.

Playing this out, the idea of individual empowerment isn’t hard to imagine. If generative AI systems could be guaranteed to be reliable and unbiased, and could function in a manner that isn’t highly environmentally impactful, they could enable mass individual access to preventive healthcare across everything from oral healthcare to cancer screenings; they could enable targeted, personalized education; they could enable automated operational efficiencies that reduce peoples’ carbon footprints. 

But this is, of course, complicated on several fronts. One of these, again, in line with the business model that these systems are being born into, is a mounting concern of economic inequity in which AI is leveraged to replace workers, with a result of disenfranchisement, rather than empowerment. 

But while Xing acknowledged that there is risk inherent to a widespread integration of AI, he said the key is in regulation for risk mitigation. And when it comes to AI, he said that we’re not dealing with a unique threat: “In the past, people figured out that there is a clear distinction between the science, the technology and the product.” 

The key thing with AI, he said, is that its capabilities are heavily overhyped.  

“AI has a lot of limitations. It is not a living creature that has self-identity, self-drive, agency, free will and so forth,” he said.

“In fact, it is unclear to me at least where all those will come from based on the current architecture.”

Check out the full episode here.