Quantifying your Self? Need a human-centered data structure?
April 5, 2011 § 2 Comments
Quantified Self is not a distant dream. Miniature cameras, practically unlimited memory capacity, an increasing number of sensors selective to motion, vibration, light, radiation, acoustic signals, chemicals, position, weather, bio substances, pressure and other physical phenomena, together with mobile physiological recording systems, and intelligent devices provide an unlimited architecture for quantifying ourselves. The question is, what to do with all this data and how to accomplish it so that it could provide unforeseen human benefits.
New form of psychology
My first encounter with the Quantified Self (QS) concept took place in a seminar arranged in Microsoft premises, Mountain View, in summer 2010. The audience consisted of a group of innovative people with a background in ict, psychology, anthropology, social sciences, medicine, and other disciplines. They all shared a serious interest in the knowledge revolution that is taking place in the recording and management of human, individual data. When they all, about 150 attendees were asked to describe their QS interest in three words it became clear that a new form of science and practice of multi-disciplinary psychology is being born.
Present recording-oriented human brain sciences and the superficial psychological methodologies will be complemented by the new science that is seriously interested in the real life and living context of the individual. This is something that many modern psychological disciplines have forgotten: authenticity is nearly an alien concept to these disciplines that use their brute force (e.g. simple minded brain recordings combined with simple psychology, exaggerating generalization of the gene impacts on our future, mechanical profiling by personality measures) applications to our lives.
If you don’t believe this somewhat gloomy claim, read any basic book on neuropsychology, cognitive psychology or unfortunately, a book of perceptual psychology. Try finding everyday life, our simple joys, sorrows and life styles in them. Luckily, you can still be delighted by some introductory psychology textbooks that approach the reader by “real-life” examples.
The proponents of QS are designing methods to collect and organize massive amounts of personally relevant data from single individuals – in real time or close to it – from a number of sources like various behavior measures, non-invasive probes, location indicators, body sensors, self-assessments, personal audio or video, health and other personal records, life-event logs, chemical recorders, or even gene data. But I left the meeting in Mountain View somewhat surprised, noticing that not much discussion concerned the strategies for collecting and especially organizing such huge data masses. This will be a real challenge in future.
Human-centered data structures needed
In the Quantified self -approach, the individual is always in the center of all data collection and retrieval, just like we all are, as subjects, in the center of our own information architectures, our own and to-us-relevant world. We manage our perceptual, experiential, motor and thought process data from the first person, me-perspective. Our memory for all this information and knowledge is built on the subject-oriented architecture where information from this perspective can be organized in an intelligent and meaningful way. It is not an accident that our memory is based on the first-person perspective. Its main aim is not the accuracy of retrieval but the subjective relevance of all memory data. This is where computer systems often fail – by being accurate but irrelevant.
To the best of my knowledge, there is no known artificial data structure and architecture model that would be designed for this purpose, i.e. to serve the need for organizing first person oriented data collection, storage, management, and retrieval. I’m convinced that we need to build data architectures that are solely meant to serve this purpose, to manage the first person –oriented data masses. It is not only a question of linking a person code with the data concerning him or her: even if we can relate all the data that we have recorded from an individual or from the sources related to him, we still need a meaningful integration of the data. The principles of this integration are an interesting and ambitious puzzle.
As a thought experiment, try remembering an accident that you witnessed or a visit to a memorable place, long time ago. Typically you first try to memorize the key events, times, places, or people present in them. Essentially you view your own history from the first person perspective by constructing an intimate mental scene and a play that only you can follow. This amazing process is an effective demonstration of the human-centric, first person -oriented data structures and architectures that underlie our memory performance.
Our memory for personal events is not perfect, far from it, but it is always relevant unlike the artificial memory systems that we use. In fact, the worst mistakes made by artificial memory systems are those that have lost the knowledge of the personal relevance of the data. For a human, even in the case of memory slips the subjective material recovered has a personal relevance. We do not make any memory slips or any mistakes: they are nearly always somehow meaningful to us, right or wrong.
Human centric data architecture
The power of the human-centric data (HCD) architecture is that it allows and preserves the personal perspective, whatever data or information is processed, even when the memories are fabricated, incomplete, under the influence of suggestions, or simply false. Personal relevance is an extremely valuable property of this memory as it helps individuals to survive in complex and even dangerous environments. There is no formal data structure that would describe exactly what personal relevance means.
With a slight exaggeration one can claim that artificial memory systems are designed to store and recover even the slightest details correctly while the recollection perspective is secondary. For the human centric memory, the opposite is true: recollection perspective is primary and the detail accuracy of the retrieved information is only secondary.
In recollecting the distant memories we identify the key elements, enrich them and link them with other relevant memorized events. After a while we have created a coherent memory structure that we can internally view like a movie. It may not be completely true as Elisabeth Loftus and other researchers have shown, but it is amazing how, piece by piece, we weave such a functional lace of memory of experiences. Endel Tulving, the famous cognitive psychologist has described this specific human memory as episodic memory.
Computers and their memory systems don’t come even close to the human memory performance when it comes to remembering events – why?
First of all – the computer has no idea who is memorizing and what. It has not been possible to program a person into a computer. This is the problem of any memory system that is based on the idea of storing objective data without its context and history. The human memory system is a contextual memory that stores both the objects of interest (data) and the context (associated data). Both of them can act as search keys later when needed. That is why smells are associated with the memory of significant events or why certain familiar places feel so wonderful to us.
A problem in dealing with masses of personal (subject-oriented) data is that our present data structures have not been designed to represent, store, organize and recover human-centered information or knowledge. So the question is: what would be the best candidate for a human-centric data base model that would support human/person -centric data processing? There is an increasing need for this type of data structures and data base theory.
I believe that for QS developers as well as for anybody dealing with masses of personal information (which is commonplace today), a subject-oriented data structure can provide a means to organize data in subjectively meaningful and effective manner.
So, as a candidate for the human-centric memory I suggest here the episodic data structure concept (superficially) that I have discussed with a colleague of mine (Jari Lehto) already for years but never had a chance/time to present or develop it into a product. In a pilot study some time ago with Sauli Laitinen we could demonstrate the power (and weaknesses) of the human episode structures for collecting and retrieving our personal experience data.
The basic idea for the human-centered data structure is simple: an episode is a structure that can be taken as the basic entity in a subject-oriented data structure and the memory architecture can be built on that. There is a lot of literature of episodic memory (cf. Endel Tulving, especially). This is how I define the HCD in general terms:
1. An episode is a structural element of our memory of experiences and actions. When we remember something from our history (like painting a garage, that we used as an example with my colleague) we recollect it as a structure of episodes (going to a shop, buying the paint, preparing the work, doing it, cleaning the place, admiring what we did …). Then, after 2 years we try to recollect from our memory, where did we put the paint cans if we want to know what colors were used? The episodic chain structure of subjective historical data comes to help.
2. An episode includes an actor (me), task/action/goals, relevant content (data, knowledge, associations, perceptions, attentive focus, etc), a temporal window and timeline, and a context (social or other).
3. Episodes are strongly linked with each other so that it is possible to recover detailed data through a linked set of personal episodes. The memory information in the episodes is not exact, but it is always relevant, subjective, and first-person oriented.
4. For an unknown reason, the episodic data structure in the human memory is psychologically very coherent, detailed, and well integrated, almost like our perceptions. This makes it an extremely efficient and robust structure to deposit, store, and recover the essential elements of its data, even after several years.
5. The access to the episodes activates automatically many of its constituents and it provides a very rich collection of subjective data, including the emotional, motivational and social aspects that are strongly interlinked.
6. It is possible to build an episodic data structure that allows matching the data structures of different individuals and even masses of people. This has a huge application potential, some of which I have sketched for specific purposes (not here).
7. Combining numerous QS data sources with and episodic representation and a related data base allows efficient search of subjectively relevant data and experiences.
8. The episodic data base system need not perfect: it can include fuzzy elements, but it must to preserve the personal relevance of all of its data and processes. Of course, it is possible to support a high-performance memory by innovative technologies, but the first-person, subjective data can never be perfect and complete. This is not a problem for our normal human memory.
This kind of an episodic memory system is possible to build and test: I’m curious to follow this development which seems unavoidable in front of the increasingly rich personal data sources and the pressing need to manage the data masses. Episodic architecture and data structures allow the building a relevant context to all subjective data, including and especially in the QS applications.