When we reach the end of our days, our life experience will equal what we have paid attention to, whether by choice or default. - Tim Wu, The Attention Merchants
One of the early insights that emerged inside Google during their early days was that the data they were collecting about their users might actually be more valuable than their search technology itself. Steven Levy, in his book In the Plex, describes how Google realized that the contents of their server logs might be the key to their future. The eclectic technological pursuits of Google engineering during the 21st century become more explicable if they are understood through the lens of data collection. Chrome browsers, Android devices, Google glasses, and even self-driving vehicles may best be understood in the context of facilitating the growth of Google's massive stash of data about you and me.
In 2009 Peter Norvig and two other Google employees wrote a seminal paper on the qualitative impact of massive data. It was titled The Unreasonable Effectiveness of Data and in it they made the observation (I paraphrase) that sufficient amounts of data alter, not only the quantitative approaches to computation, but the qualitative opportunities as well. Put another way, if you have sufficient amounts of data it fundamentally alters, not only how you go about using it, but the very essence of what is possible at all.
This was an important insight and turned out to be a sort of Rosetta stone for explaining where it is we find ourselves today, especially in terms of artificial intelligence (AI) and machine learning (ML). The cost efficiencies achieved in storage technology over the last two decades, combined with a massive extension to our reach and scale for data collection (e.g. smartphones and other "smart" devices, ubiquitous wireless data communication at high bandwidths, etc.) has created the perfect storm of opportunity for exploiting data in ways never before possible.
It should come as no surprise, then, that Google has continued to be the source of multiple breakthroughs in the application of their early insights about data, specifically in the area of machine learning. There have been major technological breakthroughs in the last few years in the use of machine learning for natural language processing. Actually, there have been many other breakthroughs as well. But for purposes of this post, I want to highlight a key breakthrough described in Google's 2017 paper Attention is All You Need. There is a deliciously rich, dripping irony in the paper's title, considering the fact that Google itself has an economic incentive to control, some might even say deprive, the attention and focus of its users.
Without going into the technical details, the insight of the paper revolves around the fact that implementing the ability for a machine learning model to adapt the interpretational emphasis or weight placed on various words produces more human-like understanding of the text. This kind of attention, based on, essentially, the context (e.g. word position and placement relative to other words in a sentence/paragraph/document) is sufficient by itself to do the heavy lifting of natural language processing. It is computationally expensive but applicationally transformative.
Attention, in this context, is a term of art but it intentionally borrows from the more mainstream definition to highlight that what is being described is a form of adaptive focus and prioritization. The paper describes how to do this computationally in areas like processing text, but the concept highlights the general criticality of how even human beings prioritize what gets our attention.
Like any emerging technology, AI presents both opportunities and risks. At the end of the day, AI is mostly just statistics, albeit highly advanced statistics informed by mind-bending volumes of data. But it's still just statistics. It is a powerful tool, and like any powerful tool, fallen beings will bend it to pursue fallen ends. But we will also use it for doing good, so as in every other context, we are a morally mixed bag.
Potential for Deception
My own primary concerns revolve around the use of this technology to deceive and to manipulate. It may be the most powerful technology ever developed for mass deception. There are myriad good uses for AI ranging from medicine to science to agriculture and beyond. But we should remember that the leading inventors of AI are engaged in a business model that depends upon manipulating the attention and behavior of their users.
The image you see below is from a web site called thispersondoesnotexist.com. Pay attention to the extreme detail and consider the fact that this is not a photograph of an actual person but a fictional and contrived picture generated by AI technology known as a generative adversarial network (GAN).
AI is able to generate videos that are visually indistinguishable from actual video recordings. The GPT-3 language model developed by openai.com is able to generate entire essays in response to questions. The essays it computationally produces can be indistinguishable from essays written by actual human beings. We should expect these kinds of capabilities to be put to use to manipulate elections, destroy reputations, or divert the attention of cognitively vulnerable people away from more vital activities.
Morally Compromised Business Models
Any business that makes a living by selling ads must attract the public's attention in order to get paid. The power of internet advertising has been its ability to track and quantify the effectiveness of advertising as a means of manipulating the behavior of users. The economic incentives of online advertising are so lucrative that all of the "free" services offered by online companies to attract their users are affordable merely as a pretext for hoovering up user data to facilitate more "effective" ads. By "effective" I mean ads that more successfully alter a user's behavior compared to what that user would have done had he been left alone.
It is this perverse incentive to manipulate, when combined with the power of AI, that I find to be the biggest concern. In a different world, we might have knowledgeable and sophisticated legislators who were crafting informed legislation to prevent bad behavior on the part of online businesses. But that is not the world we live in. Congress has never been comprised of geniuses and angels.
Even now, the level of public deception and manipulation may be higher than we can quantify. After all, by definition people don't know when they're being deceived. Many have commented on the elevated strife that attends social media use. Social media companies have long understood that that exposure to negative content motivates behavior and retains attention more than exposure to positive content. This knowledge has given rise to outrage-inducing clickbait which foments a constant state of agitation on the part of many social media users. And this outrage, alas, is not limited to the young and inexperienced. Perhaps you have experienced the phenomenon of people in your acquaintance who are dominated by their concern over issues and events over which they have no control, and which are unlikely to have any significant impact on their lives.
Let the reader understand.
Controlling Our Attention
The irony of recent discoveries about the power of attention in artificial intelligence is that those very discoveries were made by organizations whose entire business model revolves around distracting their users. That's worth remembering. They know that the nagging, attention-getting distractions their applications utilize are intentional efforts to reduce our ability to focus on other things. But that's how they get paid. Remember that: they profit from our distractibility.
The only way to avoid the deception and manipulation is to become far more intentional about where we focus our own attention. It is wise to eliminate the ability of social media to call attention to itself on your phone. Deleting social media apps, and only ever using social media through a web browser, is one way to contain social media's ability to introduce distractions.
Developing a healthy skepticism about all media content is another useful response to our current moment. You cannot assume that any video or photo or audio recording you see or hear has not been doctored or otherwise manufactured.
Finally, pay attention to higher things. Apply sustained focus to things that are healthy and good. There is wisdom to be found in the ancients. The Christian apostle Paul encouraged the early Christians in Philippi to apply their sustained focus to things that, notably, didn't involve outrage and strife.
Here was his list:
True
Honorable
Right
Pure
Lovely
Admirable
Excellent
Worthy of praise
To be clear, the grammar of the original language indicates that the apostle was suggesting that our sustained attention should be on things that meet the lofty standards of this list. He even assumes that the recipients of his letter possessed the discernment necessary to differentiate those things which make the cut.
You may want to take charge of your own attention. Your freedom probably depends upon it.