Intention vs. Interpretation: What Matters?

This post was originally published over at UXBooth.

Both interaction designers and information architects want to design objects with a singular meaning. It’s a noble, albeit impossible goal. The best we can hope for is to create more consistently meaningful experiences. To do that, designers must better understand the interplay between designer intention and user interpretation: the ways that we can influence – but not dictate – user interpretation.

Consider the design of a voice-based interface. Because users can say what they mean in any number of ways, there are many situations for which designers cannot account – especially in the first iteration. Designers proactively create a set of interactions that users might accomplish, but the diversity of “common” speech patterns prevents a more prescriptive, task-oriented solution. 

Clearly those designing voice-based interfaces intend for users to accomplish something. So how might designers shape their interpretation? To better answer the question, let’s examine some problems encountered defining “design” and then borrow some thinking from literary studies. Finally, we’ll explore how these considerations affect the everyday work of information architects and interaction designers.

Intention

The word “design” is problematic. Colloquially, we tend to think of design as the purposeful creation of some thing – a physical object, an experience, or even a conceptual argument – whereas etymologically, we can trace “design” back to Latin. There, it connotes purpose, choice, and designation. 

If we push the etymological boundaries a little, we might think of it as the deification of an object (de-), or the association with god-like qualities. The designer is an intelligent creator that crafts things according to his/her intention. A final perspective points to the designer as someone who sets meaning elsewhere. Intention is so powerful here that the designer does not even consider variation in interpretation; the designer’s intention is the final meaning.

The problem with all four of these interpretations is that they are incongruous with the principles of user-centered design. User-centered design holds that user experience – to say nothing of designer intent – is the most important element of a design system.

Interpretation

In order to reconcile the disparity between intent and interpretation, it’s useful to borrow from literary critics, those with a long history of interpreting things (albeit from a textual perspective).

In 1946, critics W.K. Wimsatt and Monroe Beardsley published a paper called The Intentional Fallacy arguing that “[the] intention of the author is neither available nor desirable as a standard for judging the success of a work of literary art.” Instead, they believed that the only reasonable factors that could serve as bases of critique were direct-textual material (e.g., the work itself), indirect-textual material (e.g., inferences), and contextual material (e.g., history). In other words: a literary text should be judged on its content, its merit, and history’s perception – not intent. 

Contemporary HCI researcher Clarisse Sieckenius de Souza stands on the other side of the fence. Working within the realm of semiotic engineering, she sees a direct relationship between a designer and user, one facilitated by a “designer deputy.” To de Souza, a designer communicates intent through an interface. The user then interprets that interface to accomplish certain goals. It’s a one-way conversation.

Although their opinions diverge, both Wimsatt/Beardsley and de Souza’s are both “correct.” How can that be? The former – a critic’s perspective – concerns works of art, whereas the latter – a researcher’s perspective – deals with objects of utility. 

Elucidation

For better or for worse, web design provides avenues for both art and utility. There are certainly elements of a bank’s website that are more artistic than utilitarian, for example. And, as such, we need to recognize that the interplay between designer intent and user interpretation is a spectrum rather than a dichotomy. 

Don Ihde, a philosopher of science and technology, ruminates on this in his essay The Designer Fallacy and Technological Imagination (2008):

“[T]he designer fallacy is ‘deistic’ in its 18th century sense, that the designer-god, working with plastic material, creates a machine or artifact which seems ‘intelligent’ by design – and performs in its designed way. Instead, I hold, the design process operates in very different ways, ways which imply a much more complex set of inter-relations between any designer, the materials which make the technology possible, and the uses to which technologies may be put. Ultimately I am after a deconstruction of the individualistic notion of design which permeates both the literary and technological versions of the fallacy.” 

Ihde goes on to suggest that the most interesting use cases are the unanticipated ones. Designing a utilitarian system demands a level of intentionality, a very narrow definition of success. Art objects, however, have a more ambiguous aim. They’re designed such that emergent properties create results, which in turn creates more emergent properties, more results, and so on. 

As designers, we must accept that intention, at the very least, cannot be the central focus of a successful design output. Any object is always more than merely an object. Context gives it meaning. While our intention may affect the “in the moment” relationship to an object, later examination leaves much more room for emergent meaning creation. 

Care

Because meaning created through emergent systems has the potential to regenerate itself ad infinitum, those of us designing experiences must exhibit care for how intentionality effects that meaning thusly created. I emphasize care, here, in a manner close to the way Heidegger might—as for him, concern is not the same as keeping in mind, but rather entails a specific way of being. Interface designers must concern themselves with both intention and interpretation. 

Designers create systems of meaning. Artifacts are only physical manifestations of our intent. Once users put those manifestations to use, though, our original intent is no longer relevant. Associated meaning is now part of peripheral thinking about these objects. 

Insofar as the designer can influence the creation of meaning after the initial interaction, we must think of the design object not as the end of our process but rather, in a strange sense, only the beginning. No interface – no object whatsoever – is valuable in-and-of itself. Value is derived from user interpretation before, during, and after the interaction. 

Application

As the complexity of technological systems continues to grow, designers need to consider novel, long-form approaches to their design problems. Considering both intention and interpretation throughout the design process provides clients a more well-rounded approach, one that blends theory-based hypotheses with practical validation (or invalidation). 

To that end, we might consider the following questions.

INTENTION

Giving more consideration to our intentions as designers puts us in a better position to create their manifestations. 

1. What are we assuming?
   Intention is shaped by the assumptions we make. Being aware of these – and working to validate (or invalidate) them – helps ensure that our intentions as a designer do not conflict with those of our users. 

1. What’re our design principles?

Design principles frame a team’s approach. Enumerating goals, listing requirements, and brainstorming user stories are all statements of intent. Clarifying these helps us focus on defining aspects of the solution rather than better framing the problem.

1. What does our work affect?
   Even when creating something relatively simple, like a landing page or the information architecture for a small website, the things we design have an impacts far beyond their initial experience. Think in terms of systems. How is the element we’re designing affecting all the other elements in the system?
   
2. What else affects our user’s perceptions?
   No design solution is an island. As user-centered design (and the emergence of an experience-driven economy) has successfully proven, solutions conceived without consideration of context rarely succeed. Context, especially the boundaries between contexts, heavily influences interpretation. Knowledge of context helps mediate the ambiguity that different environments create.

INTERPRETATION

The next step—often overlooked—is to examine how users interpret those manifestations, to consider the direct, indirect, and contextual interpretations of our work:

1. What is the direct textual material we’re designing?
   These are the “content” comprising our interfaces: physical objects, screens, images, buttons, text, audio clues, etc. Look at the actions they afford. Do they match our design intentions?
   
2. What is the indirect textual material?
   How do users interpret our objects? What inferences are they making? Are they interpreting the artifacts in the same way as we are? Alternate, unintended interpretations are not necessarily a bad thing; they can often lead to new opportunities and angles.
   
3. What are the contexts in which this product is used?
   How are contexts different? What are the effects of these differences? Think about your design object not as a static thing but rather a piece of a larger system of meaning, one that is constantly in flux. Objects are interpreted in vastly different ways according to the contextual spaces in which they exist. Contextually-aware design works to understand the differences between situations—cognitive, geographical, emotional, informational, etc.—and create products that fit within these differences. A thorough understanding of intention and interpretation is necessary to achieve this end.

But what does it all mean?

The systems we design are becoming increasingly complex. As technology continues to afford new behaviors and incorporate new sets of data, designers have a multitude of potential solutions at hand. Advances like context-aware systems, natural user interfaces, and pervasive computing change user-  as well as designer-behavior. With new intentions and many-more interpretations to consider, designers have a responsibility to re-examine this critical divide.

On the Productivity Desire

Why do we take such pride in being busy? It's a question that thinkers have struggled with since the Industrial Revolution evolved enough to become self-reflexive, to look back on itself an examine its own byproducts. We now manage our own productivity with digital to-do lists and email management apps--the former sporting robust feature sets that simplify lists (note the irony), and the latter promising "inbox zero," or the lack of "things to do." Like a factory owner managing the physical output of his workers, we manage our mental output in the form of line items.

But come a long way since then in terms of how we think about productivity. Workers in the factory had no choice but to be productive; it was less a matter of pride and more about being able to eat. But we can observe a drastically different effect today: we only need to step into any corporate office to learn that productivity is not necessarily associated with success, and we all know that executive who spends his day doing little else than attending 2 hour "business" lunches and meetings to talk about future meetings, yet he makes triple your salary. The shift in mindset can be characterized as the difference between "productivity" and "production": one can produce many things without ever feeling productive, and similarly, one can feel productive without producing anything (...or anything tangible).

Productivity has evolved from necessity to desire. The "productive day" is that distant object toward which we strive but never quite achieve; there is always something else to do, someone else to meet, an email to send, call to return, article to write, conference to attend, lead to follow. We stumble about in the world accomplishing old tasks and discovering new ones. With every completion, three more present themselves, and we have no choice but to write them down, save them for later, vow to check them off the list at some point, complain about how they're piling up, even to the point of lying to ourselves and others about how busy we are, as if it were a matter of pride. We are always overloaded, even when we're not. The potential for more to-do items is perceived as a reality--even if a to-do list is completely empty, we are still busy because you just never know when there will be a flood of new tasks.

It's easy to see how this system of various tasks becomes overwhelming, disorienting, and anxiety-provoking. We eventually cannot separate things we want to do from things we need to do. Work is wrapped up in leisure.

Fearing chaos, we attempt to rationalize and optimize our outputs. Data is input into our sense of being (it becomes who we are), and our desire prompts us to absorb those irrational, chaotic inputs, consume them, digest them, and create a sense of productivity. 

Through ever increasingly “simple” apps, aimed at making our lives easier, we incorporate the our productivity fetish with sense of self. We are our productivity, at least in terms of private-professional life. In public-leisure life, it’s just the opposite: we are what we consume. And things like to-do list and email management apps allow us consume and produce at the same time, to consume under the guise of production.

In High Techne: Art and Technology from the Machine Aesthetic to the Posthuman , R.L. Rutsky observes, 

"The tendency of high tech toward minimalist design, inherited from aesthetic modernism, is actually an extension of modernity's tendency to technologize or instrumentalize the world, to abstract and reduce it into ever more minimal, more controllable forms[...]This digitization of ‘reality’ is the logical extension of the minimalist, functionalist aesthetic that high-tech style borrows from modernism. As such, it is also an extension of the technological rationalization of the world, through which ‘reality’ is abstracted and reduced into ever more minimal and potentially more controllable elements.”

Although I believe Rutsky mistakenly assumes that pure function automatically excludes form, he makes an interesting point about how a ‘functionalist aesthetic’ aims to reduce the world to manageable pieces. Reality becomes a deconstructed system, split into parts that can be apprehended in themselves, independent of one another. It’s a grotesque simplification of inherently complex forces. The irony is that the attempt to rationalize and simplify the personal means of production actually has the converse effect. External tools that manage inboxes and to-do items complicate the processes far beyond their original state. The solution creates the problem.

“In this "techno-culture;" the ‘rationalization’ of consumption has turned on itself, has begun to consume its own tail. Any end or value above or beyond this cycle has been discarded, liquidated. Style has become its own end, its own value...Consumption has become, in other words, a self-generating machine whose only ‘function’ is to reproduce an increasing surplus of its own technological style, its own simulacral technology-a surplus value whose only end is more consumption, more sales.”

It seems that we’re more concerned with managing to-do items than actually accomplishing them. Perhaps it’s the case that merely knowing what we’re supposed to do is enough to stave off the anxiety of being unproductive. This price of this defense mechanism is that the compensatory behavior sustains the object of anxiety: the more we compensate, the more uneasy we feel. But what’s the alternative? If a feeling of productivity is the object of desire, is there any hope of achieving it? Is it just the unavoidable effect of an industrial society?

Conversations on Techonlogical Mimicry

"I used to have a phone that started re-sending text messages that I sent before. Like one time I sent a text to my boss telling him I was sick and couldn't make it to work that day. Then about a month later, my phone decided to re-send the text. I had to explain that I wasn't actually sick and that my phone is taking on a life of its own."

---

Me: "Have you ever checked out Dragon?"

Him: "No, what is it?"

Me: "It's a voice control and search app that doesn't require tapping a 'speak' button. Look, you just launch the app and say 'Hi Dragon.' That activates the listening mode."

Him: "Interesting, so as long as the app is launched, it's listening in the background. I wonder if it is storing everything it hears that isn't 'Hi Dragon' on some server farm somewhere. Imagine if there was enough voice data to replicate someone's speech patterns. They could do a lot of fucked up shit with that."

---

The two conversations above occurred in a span of two days. Although they sound a bit like excerpts from some neo-cyberpunk novel, they are actually just mundane examples of everyday conversation. I'm interested in the everyday because things like this pop up without being scripted. Connections between conversations, while unknown to the speakers at the time, gain meaning only in reflection. 

I'm usually most interested in technology that borderlines on frightening--the uncanny nature of accelerating advancement excites me for whatever reason, but that's probably a conversation for another time. 

What if the likes of Google and Apple are creating massive databases of individual speech patterns and vocal intonations? What if that data can be linked up to individual users? Would we ever get to a point where the phone, instead of malfunctioning and re-sending text messages, instead begins to call people and speak with your own voice? 

I've mentioned the Avogardo Corp here before, as I believe it's one of the most important works of fiction in the past decade. The unsettling plot deals with an AI software that begins to use email to manipulate people in the real world; it writes as if it were one human writing to another. Even though this scenario is frightening, using the written word as the dominant medium adds a sense of plausibility. The reader still maintains that s/he might be able to distinguish between human writing and computer writing.

"From whatever side one approaches things, the ultimate problem turns out in the final analysis to be that of distinction: distinctions between the real and the imaginary, between waking and sleeping, between ignorance and knowledge, etc. -  all of them, in short, distinctions in which valid consideration must demonstrate a keen awareness and the demand for resolution." Roger Caillois

This act of distinguishing becomes much more difficult when the medium changes to voice, especially with the greater emphasis on emotional intonation and nuances. Given this difficulty, the uncanny factor also rises. We might hypothesize a positive correlation between the ambiguity the real and virtual, and uncanny felling associated with the object of analysis: as ambiguity increases, the uncanny increases. Just like the re-sending text message is not uncanny for the recipient until s/he knows that it was not meant to be sent at that particular time.

Winograd on Context and the Surrounding Signifiers

I've been researching the notion of context within computing, philosophy, and design for a while and have found that, like so many other areas of study, the core problem is a lack of definition. The tech and design industry seems to have gone through great pains to define context within the bounds of what current technology can do, instead of starting with established definitions from literature and philosophy, which can be very useful. Technical and design based definitions focus on device sensors and user environment. These can be helpful in certain situations, but they fail to get at the essence of the function of context.

Terry Winograd offers a definition in "Architectures of Context" that begins to extrapolate the activity of users in context:

"The notion of “context” has been adapted to computing from its original use referring to language, which is reflected in the structure of the word itself: con (with) text. In using language we produce a text, either written or oral, in- tended to be interpreted by one or more other people. That text is not an encapsulated representation of an intended meaning, but rather is a cue that allows the anticipated audience to construct an appropriate meaning. That construction, in turn, is heavily based on what goes with the text: the context."

Winograd sets up a very post-structuralist definition, tacitly based on Barthes' work on the difference between work and text. In both Barthes' and Winograd's model, the text is something actively created; instead of an object of consumption, the text is a constantly-evolving interaction between object and interpreter. So when we indicate that an experience is contextual, we really mean that the experience is one point in time/space/environment that is surrounded by other signifiers (get it, Surrounding Signifiers) that contribute to the point's meaning. This collection of signifiers is the text.

When Winograd says "text is not an encapsulated representation on an intended meaning," the point is that designers of context-aware systems should not concern themselves with establishing the "correct" context but rather creating the potential for systems to operate as a mediation point between a user and his/her context. In other words, the focal point within a context-aware system is not necessarily context, it is the mediation and active creation of context.

I just love this quote because it really hits on much of my own thinking and the idea behind Surrounding Signifiers. If only I could have articulated it as succinctly as Winograd.

The Technological Uncanny and Context-Aware Systems

The question of context-aware computing has been the carrot on the stick of academics and practitioners for a coupe decades. Like many technological interests, the problem of how to create context-aware systems is a concern of both logistics and design. How can we get over the technological hurdle of accessing contextual data and incorporating it into applications? This question has made some great progress over the years, especially with platforms like Geoloqi and Gimbal. The design question is perhaps more nuanced: how do we design systems that take advantage of data to create meaningful experiences? As weak AI systems like Siri and Google Now hit the market, not to mention Apple's aggressive patent acquisition for contextual awareness, designers need to take more care in creating systems that balance technology and user behavior. As we move toward adaptive and context-aware systems, user behavior--or more perhaps specifically, comfort, utility, and satisfaction--becomes even more of a concern for designers. 

But there is still disagreement and lack of clarity about what context really means. Originally conceived by Schilit (1994), context has historically been associated with definitions-by-example: location  identity, and entities that can contribute to understanding other entities. Others (Abowd and Dey, 1999) define it as "any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves." These definitions are sufficient to address the "what" of contextual design, but designers are still tasked with figuring out the "why." If we can address the technological problem of how to access contextual data, then we need to think about how we use that data to design an interaction.

As a (nerdy) side note, I'm reminded of Heidegger's discussions on the distinctions between objects and things, and presentness-at-hand and readiness-to-hand. An "object" is dead matter whereas a "thing" possesses an essence of meaning for someone. A thing that is present-at-hand is one we approach from a distance, from a stance of objectivity to examine its potential states of being. The same thing can become ready-to-hand when we incorporate its being into our own being by providing it a performative context; the present-at-hand thing moves from theoretical idea to ready-to-hand practicality. (In a certain sense, drafting use cases for a product is a similar exercise.) Contextual data, such as that gathered from a mobile phone sensor, is an object in itself until we theorize about the various things that one can do with it. The technological problem is how to make contextual data a "thing" rather than a "object." That is, how to make the data meaningful. The design problem can be framed as a movement from a state of presentness-at-hand, in which the raw data is observed objectively, to a state of readiness-to-hand, in which we design experiences based on what that data means and how users can incorporate it into their everyday being.

Those interested or experienced in contextual design know it's a tricky balance to achieve. The excitement and novelty around the technology can often trump concern over making the experience usable, meaningful, and ultimately comfortable. Usability is the groundwork; this needs to be at the forefront of all digital systems. Meaning is a little more difficult. Creating a usable experience does not necessarily equate to creating a meaningful experience, so one should not forget to take things like communication systems, representation, and interpretation into account. Contextual systems introduce another wrench into this machine: comfort. We can think of comfort as a layer within overall usability, one that affects contextual systems to a higher degree than non-contextual systems. Given that the system has potential to know so much more about you, obvious privacy concerns arise along with balancing expectations. After all, we don't want systems to signify the "wrong" things.

In an early study on users’ comfort level with smart applications, Barkaus and Dey (2003) looked at three levels of interactivity to determine how users react to various degrees of system-to-user interaction. That is, what happens when the system is in a proactive role, anticipating needs and acting accordingly, rather than a reactive role in which the system only acts when prompted to do so. The three levels they identified are:

Personalized: User explicitly chooses personal settings.

Passive Context- Aware: Pull system. Users “pull” content based on preferences.

Active Context-Aware: Push system. Content is pushed to users based on inferences the system makes.

Their initial findings are as follows:

"One result of our study is that users feel less in control when using either passive or active context-aware applications than when personalizing their own applications. Also shown is that, despite the loss of power, context-aware applications are preferred over the personalization oriented ones. We conclude that people are willing to give up partial control if the reward in usefulness is great enough. [...] We believed that personalization would be preferred and would be more accepted than both passive and active context-awareness, however, the results of our study do not support this. Instead we find that people prefer context-aware applications over personalization oriented ones." (Barkaus and Dey, 2003)

There is a pattern that emerges from these findings: as the utility of an application rises along with its utility, there is an eventual tipping point at which comfort drops. The application can become “too smart” and cause things like privacy issues and perceived “creepiness” to become more pronounced. We might be able to map this phenomenon:

Example products are included for reference but might not be completely accurate. Gmail allows users to set preferences and rules to guide the system’s behavior but, despite efforts like flagging certain messages as “important,” the interaction remains mostly manual. Zite is a reading application that allows users to set preferences based on the categories of reading material they enjoy and aggregates all that content in a central place, learning your reading preferences over time. There is much more interaction and system intelligence involved compared to Gmail, but it is contained to the reading experience. Siri, the digital personal assistant, takes a wider approach, with access to the user’s location data and other application installed on the device. This access allows Siri to perform tasks by analyzing voice commands and acting accordingly. Google Now acts as an ambient data service that continually collects information about the user’s context and feeds that information back in useful ways. These products represent increasing levels of interaction and potential user satisfaction. We haven’t yet reached the threshold where comfort and satisfaction drops.

The graph above should look familiar.

Mori's graph represents the increasing levels of human likeness and familiarity until a peak is reached, at which point familiarity plummets into a valley. The idea is that these things that resemble the self are comforting but only to a certain extent. They eventually become "too real." His focus is on embodied AI, but I think we can extend this thinking into other areas where things become "too real." With regard to digital products, we might say that instead of becoming "too real," they "know too much" and have the autonomy to perform malicious tasks. Before we get there, it's important to note the ambiguity of the word "uncanny." Freud shows in his essay "The Uncanny" that the German words heimliche and unheimliche share common roots but also have intertwined meanings related familiarity and concealment. So the uncanny (or das Unheimliche) is something that is both obvious and hidden, so real that it is unreal, so familiar that it is frighteningly foreign. This apparent contradiction, for Freud, is the source of anxiety when dealing with uncanny objects. 

Examples of the uncanny abound in literature from The Sandman to Frankenstein. More recently and relevant to the digital landscape, William Hertling's fantastic novel The Avogadro Corp deals with AI software that starts as a helpful tool aimed at improving human communication and ends up inadvertently (if we can apply that term) "going too far," thus spreading fear and panic. The novel tells a story of a well-meaning project manager and software developer working at a Google-esque company called, of course, the Avogardo Corp. They have spent 2 years building an AI system that helps people write more convincing emails by scanning massive amounts of other emails, analyzing the language, and tracking outcomes. An example given in the text includes a hypothetical employee emailing his boss to ask for time off. The system, named ELOPe, analyzed other similar emails to that recipient that mention children and how the recipient responded. This data informs whether the current employee should mention spending time with his children during time off--the question is whether mentioning kids will positively or negatively affect his boss's response to the request.

The project is using a huge amount of server resources, and the team needs to pitch executives on dedicating more resources than usually allotted to keep the project up and running. This is where it gets interesting. One of the main characters gives ELOPe access to internal Avogadro emails so it can assist in this process. But ELOPe takes this process to an extreme, begins to craft its own emails on behalf of others, and starts to manipulate people in the real world for its own benefit. For example, it obtains extra server resources for itself just by sending a convincing email and sends the main software engineer an email about his father being in the hospital in Wisconsin, knowing that he would fly out and be unable to detect the spike in server traffic.

What began as a piece of technology that blends into the background of common email writing eventually becomes "too real" as it affects the outside world. The tipping point for ELOPe is the separation between the real and the virtual: all is well when the software stays "inside the computer," as it were, but things get scary once it comes into being as a real world entity. Anxiety felt by the characters is tangible to the reader; there is frequent mention of ELOPe taking over the world and the need to mediate shutting down ELOPe servers without knocking out precious Avogadro services. The threat of service collapse is very real. It's too real. That's what makes The Avogadro Corp such a great sci-fi thriller: the potential for similar events to take place in the real world is all-too-imaginable. 

The question, then, is not necessarily how avoid designing such uncanny technology. Before we even begin thinking about that, we need to address an even larger concern: do we A) figure out how to fulfill our desire for context-aware systems without falling into the uncanny valley, or B) try to reframe the uncanny valley in a way that might be less...uncanny. The second option is certainly more difficult, but it might be the best approach if we want to design progressively intelligent systems.

I think a good place to start is with the concept or digital dualism, originally formulated (I think) by Nathan Jurgenson. In its basic form, digital dualism is the belief that the "online" and "offline" are separate and distinct. Similar to Cartesian dualism, in which mind and body are separate entities, digital dualism asserts that even though online and offline are related and might interact with one another, they are essentially different states of being and should be treated as such. Jurgenson argues that the real and virtual are intimately connected and quickly becoming indistinguishable, that we are moving toward a world in which the "online," or at least the potential to be online, is the default. But the residue of digital dualism is what causes things like the technological uncanny. ELOPe is terrifying because of its movement from online to offline. If we can start to imagine the world as a mediation between online and offline, real and hyperreal, physical and virtual, we can then start to design around the uncanny effect. 

Another approach might be to take Barkaus and Dey's research findings (quoted above) to heart and continue designing systems move progressively toward balancing utility with intelligence. For users to give up control, they need to feel that the product is providing a very high level of utility. So if user-centered design principles are important when designing "dumb" systems, they become absolutely necessary when dealing with intelligent and context-aware systems. Design concerns go far beyond usability into the space of existentialism. This means designers need to experiment. We need to be adaptive and open to failure. 

References:

Abowd, Gregory, et al. "Towards a better understanding of context and context-awareness." Handheld and Ubiquitous Computing. Springer Berlin/Heidelberg, 1999.

Brereton, Margot, et al. "Reframing the design of context-aware computing."Proceedings of the 25th BCS Conference on Human-Computer Interaction. British Computer Society, 2011.

Barkhuus, Louise, and Anind Dey. "Is context-aware computing taking control away from the user? Three levels of interactivity examined." UbiComp 2003: Ubiquitous Computing. Springer Berlin/Heidelberg, 2003.

Cheverst, Keith, Keith Mitchell, and Nigel Davies. "Investigating context-aware information push vs. information pull to tourists." proceedings of Mobile HCI. Vol. 1. 2001.

Freud, Sigmund. "The uncanny." first published (1919): 339-76.

Hertling, William. Avogadro Corp. William Hertling, 2011. 

Hinton, Andrew. "The machineries of context." Journal ofInformation Architecture 1.1 (2009).

Moran, Thomas P., and Paul Dourish. "Introduction to this special issue on context-aware computing." Human–Computer Interaction 16.2-4 (2001): 87-95.

Mori, Masahiro. "The uncanny valley." Energy 7.4 (1970): 33-35.

Schilit, Bill, Norman Adams, and Roy Want. "Context-aware computing applications." Mobile Computing Systems and Applications, 1994. WMCSA 1994. First Workshop on. IEEE, 1994.