This paper proposes the creation of what I call the “Legacy System,” a system whose design begins, in phase one, with a person’s systematic capture of their own personal data. It is a system for ensuring that data generated by a person remains of that person and for that person.[i] The system includes (1) an organization (conceived as non-profit or not-solely-for-profit), that issues (2) an iron-clad, user-protecting contract for (3) a device and operating system running (4) an application (“legacy software” / “legacy app”) that backs up (5) personal data to (6) a private, secure, user-controlled virtual machine. In phase two, the “big data” on that personal machine is subjected to (7) artificial intelligence algorithms (machine learning code) whose goal is to maximize (8) personal happiness (conceived as an ongoing exercise of virtue, with respect to both success and fulfillment).
We begin with human existence and meaningful human action as our primary value. Humans are a technological species. We use tools. History demonstrates that our species began differentiating itself from others with the invention of the handaxe. Following philosopher Andy Clark, the handaxe can be seen as an extension of the human body, of the human mind. Perhaps even more importantly, language is a human invention, a human technology. Language helps us form thoughts and communicate them to others. Language is the original telepathy. Fast forward to the digital age, and humans are still humans, but we are using digital technologies and, because of that, we are leaving digital traces or “data”. Following Matt Ridley, there is a reason why the handaxe and the smartphone are roughly the same size and shape. The human hand holds a smartphone as it would a handaxe. Both are extensions of the human body, the human mind. These observations make clear why it is crucial that in the Legacy system the human interface begins at the device and operating system level. In the generation of “data”, there is no break in the chain of “input” from human mind to human hand to smartphone to operating system to application. This point is just as important for user experience as it is for the legal protection of any data generated by such means.
Humans, using technology, are the wellspring of data. The fundamental idea behind the Legacy system is to establish a private pool at the wellspring of data—before it escapes into the wider world. To use another metaphor, the Legacy system keeps the original “wet-ink” data, and releases a “copy” into the world (through an app, etc.). If users track themselves and retain a copy of their actions at the device and operating system levels, there can be no legal argument against the claim that the user owns the original.
Think about the sensors (and actuators) in your smartphone device. To name a few: camera(s), microphone, radio, Bluetooth, Wi-Fi, GPS, gyroscope, accelerometer, magnetometer, proximity sensor, thermometer, hygrometer, barometer, and ambient light sensor. Channeled through an operating system, these sensors and actuators provide the hardware infrastructure for the primary software functionalities that comprise the reasons we carry our smartphones: phone calls, SMS, email, internet, social media, navigation, and myriads of applications.[ii] Each time we use any of those higher level software functionalities, someone else is capturing our data inputs (e.g. Google search, Facebook like, etc.) Originally, however, that “search” or “like” originated with our all-too-human life and its perceived needs. Why put our lives in the hands of someone else who manifestly does not have our best interests at heart?
Personal data privacy has been in the headlines since Snowden. Recently, the Facebook and Cambridge Analytica scandals have highlighted the issue once again and have prompted some to call for a “User Data Bill of Rights.” Holding businesses accountable for their collection of user data (which is sometimes massive—looking at you, Google and Facebook) is certainly a good start, but doesn’t strike at the root of the issue. The Legacy system does.
The core vision of the Legacy system (and its early concept) is something I call “VirtuAlly,” inspired by a seminal article by Danny Hillis, the Quantified Self movement, and of course, Aristotle’s discussion of “Friends of Virtue” in his Nicomachean Ethics. The idea is to turn the self into Big Data and run Machine Learning over that data. IOW, the goal is to build a system to collect as many of my own digital traces as possible into a database. The machine learning that runs over that data would have the explicit goal of making my life better (and not, for instance, serving me ads or trying to sell me shit I don’t need). A truly personal AI. The following mind-map provides a glimpse into the data-capture side:
One shortcoming of this early concept is that data capture operates downstream from the application layer. As such, it is beholden to any number of “contracts of cohesion” which may cede the data as belonging to the platform.
Moving from concept to prototype, I have developed the following personal journaling system using IFTTT and Evernote, a project I call “LifeLine” (think “Life Timeline”). IFTTT.com (If This, Then That) is a service that allows users to create “recipes” (basically little logic modules) to connect up various popular online applications using front-door APIs. The “If” side specifies the inputs (or “triggers”) and the “Then” side specifies outputs (or “actions”). So IFTTT provides the logic, and Evernote serves as the data repository / database. Here is a sampling of the types of logic rules I have setup to generate input:
And a sampling of the output:
Steve Jobs promoted the principle that technology should be either beautiful or invisible. A benefit of the above system is that it operates invisibly. I simply go about my daily life, and the logic rules work behind the scenes to capture the data I’ve told them to capture and to archive it in Evernote. In my Quantified Self practice, I primarily use such data for health purposes (the system provides excellent data benchmarks for diet and exercise),[iii] time-management, and a sort of externalized (and infallible) memory. As more features are added to the system, it becomes clear that the database itself could function as a sort of “digital legacy” to be handed down to heirs—along with, or instead of, a shoebox full of photos. It is a step in the direction of digital immortality.
A problem remains, however. As mentioned before, the current early prototype falls prey to the same issues as the early concept—in the current architecture, data capture operates downstream from the applications themselves (e.g. Facebook, Gmail).
Prototyping Proposal (Data Capture)
Fast-track for prototyping the Legacy system. The hardware (device), operating system, and VM components could be off-the-shelf solutions. Device (smartphone) is conceived as GSM phone. Operating system is conceived as a kernel-hardened, open-source version of Android. Virtual machines would use something like Amazon Web Services (likely running Linux). Smartphone data plan could be negotiated via strategic partnership with a company like FreedomPop (uses Sprint & AT&T networks; currently offering “Privacy Phone” / “Snowden Phone”). If we are able to use off-the-shelf infrastructure, the main work would be building the “Legacy software” app, which is basically a massively powerful key-logger (actually, an all-activity-logger) that uploads daily to a Virtual Machine proprietary to the user.
Ideally, all data would be stored on the blockchain for security. Of current solutions, Ethereum seems adequate for the task.
Prototyping Proposal (General AI)
A discussion of the General AI involved merits its own conversation, and a separate paper, “Developing Conscious Agents”, is forthcoming (in collaboration with a developmental psychologist). For present purposes, initial prototypes for the Legacy system would begin with off-the-shelf machine learning techniques. This means the “Big Data” of the self would be collected privately and analyzed privately by a personal AI. Private, personal data collection lays the real and legal foundation for a culture of consent with respect to data. Opportunities would exist within this culture for sharing specific amounts and degrees of personal information, anonymized appropriately, with a communal AI whose goal would still be to help the community and its individuals maximize their personal and communal virtue. To be clear, there are two levels here: the personal AI, and an opt-in communal AI.
A highly abbreviated summary of “Developing Conscious Agents” is worth sharing, as its core ideas will scaffold the AI in all later generation VirtuAlly instances. The word “developing” in the title is critical. Much ink has been spilled of late wondering if AI is best approached using the model of child development. Let’s take this strategy to its logical conclusion. The idea is to clone human intelligence as it develops in real time. In short, we propose developing a virtual agent modeled after a live newborn subject. In each instance of the experiment, the experimental design would include two developing agents: (1) an infant with real senses (and also equipped with virtualizing sensors, including camera, microphone, environmental sensors, et al), and (2) a virtual infant with virtual senses living in a virtual environment. The virtual environment, and all virtual bodies within it, are a physically realistic construct of the real world, driven by a highly accurate and granular physics engine (including, but not limited to an optics engine).[iv] Sensory data, collected from the real infant’s experience, streams to the virtual infant’s database where nested modules of machine learning algorithms constantly run over the collected data. The physical infant’s sleep periods provide extra windows for processing and engineering assessment. The virtual infant has the opportunity to learn EXACTLY what (and how) the real infant learns. Because the virtual agent’s conscious experience is simply actual experience copied into a virtual environment, the virtual agent “develops” exactly as the infant does, with dynamics such as joint attention, visual cliff, mirror phase, and theory of mind emerging for both agents simultaneously in real time.
The benefits of this experimental design are too many to elaborate here. To highlight one, per Saussure’s linguistics, the virtual agent will inhabit a rich world of signifiers but also a rich world of signifieds. Like the real infant, the virtual infant learns through interaction with its caregivers and adapts to a rich physical environment and a warm social environment infused with a wealth of linguistic content. The mapping of physical experience to linguistic meaning allows for the formation of concepts and practical reason. The first 200 words a baby learns are not necessarily the “top 200 words” output by frequency analysis algorithms—although significant overlap is likely to occur. More importantly, the way in which an infant learns language (through oral repetition and the labor of learning to vocalize phonemes in the context of joint attention) will allow its virtual agent to follow the same path. Many impatient types in Silicon Valley will despise this experimental design because the experiment will take at least 18 years to complete. However, it solves the AI alignment problem.
It is this AI, properly aligned with human values, that will eventually serve individuals and communities as their VirtuAlly, their Friend of Virtue.
We misunderstand Danny Hillis’ dream of Aristotle (as an artificially intelligent personal tutor) if we assume it to be equivalent to what some today call “AI personal assistants”, e.g. Siri or Alexa. If we care about augmenting our own virtue, using everything from today’s computerized technologies to ancient techniques, we must set our sights higher.
In discussing existing prototypes for the Legacy system project above, I outlined my “LifeLine” project. Actually, before that, for years, I kept a journal. And even before that, I engaged in a pursuit of virtue as a social animal. That’s the true underlying technology here. That’s what’s foundational. If language is a technology, how much more so is how you speak (your idiolect, as well as exactly what you choose to say and when). If philosophy is a technology, how much more so is your personal philosophy a technology? And personal virtue is a technology. Once we understand personal virtue as a technology, we can hack it, tweak it, make it better. Like Susan Sontag, “I’m only interested in people engaged in a project of self-transformation.” If these kinds of people come together, the novelty of the technology we use for communal and personal transformation is immaterial. Our resources are both of the moment and of the millennia.
Aristotle, Nicomachean Ethics (350 BCE)
Andy Clark, Natural Born Cyborgs (2004)
Joel Doerfel, Slices and Traces (2012)
Daniel Hillis, Aristotle (The Knowledge Web) (2004)
Jaron Lanier, Who Owns the Future? (2014)
Cathy O’Neil, Congress is Missing the Point on Facebook: Americans Need a Data Bill of Rights (2018)
Tim Palmieri, What Sensors are in a Smartphone? (2018)
Matt Ridley, The Rational Optimist (2011)
Ferdinand de Saussure, Course in General Linguistics (1913)
Doc Searls, The Intention Economy (2012)
Gary Wolf, What is the Quantified Self? (2012)
[i] Following thinkers like Jaron Lanier (2014), “data” is defined here primarily as any information or digital trace generated in digital space by the actions of a human person, and secondarily as private information deriving from an outside source that is the rightful sole property of that person.
[ii] In Andy Clark’s sense, these become human functionalities, extensions of our human functioning (e.g. when is the last time you navigated without GPS?).
[iii] A high-ranking, explicit motivation in capturing data about myself is to track my physical and mental health. As such, all data captured should be subject to HIPAA protection.
[iv] IOW, the virtual environment is basically the Matrix. A side benefit of the experiment is that afterwards, you also have the Matrix (and can use it for things like discoveries in physics; like Feynman says, “there’s plenty of room at the bottom”).