There is a limit to what we can accomplish with one-size-fits-all design. For one thing, by trying to be everything to everyone, many products become bloated and ultimately lose a lot of their usefulness. The Wenger 16999 Swiss Army Knife (see image to the right) provides an amusing illustration of this point. But the problem is even deeper than feature creep and bloat. One-size-fits-all design inadvertently, but inevitably leads to discrimination: User interfaces designed with able-bodied users in mind, even if they are technically `accessible', are inefficient for many impaired people to use not because of any inherent barriers to efficient interaction, but because of the mismatch between the user's effective abilities and the designer's assumptions. Interfaces designed with Western aesthetic sensibilities in mind, will appear less trustworty, engaging or usable to people from other cultures potentially causing them not to take advantage of valuable digital resources. As our study of Doodle users shows, people in different cultures perform social decision making tasks differently.

I argue that personalized and adaptive user interfaces are one way to accommodate these individual differences and, thus, to provide more equitable access to digital resources.

The Internet has opened new ways for behavioral researchers to conduct human subjects experiments. Yet remote, Internet-based experimentation is not yet part of the standard research toolkit because of the concerns about data validity, feasibility of recruiting representative participant samples, and perceived barriers due to the inconsistencies between contemporary research practice and the assumptions underlying the design of existing online experimentation tools (Law, et al. 2017). We are developing and validating tools and methods to enable novel ways of conducting large-scale empirical work with human subjects. Our core goals are to enable a much faster theory-to-experiment cycle, to facilitate access to larger and more diverse participant populations, and to enable larger scale experimentation (in terms of numbers of conditions and experiments) than what is feasible with lab-based methods.

Our LabintheWild platform is one of our projects designed to overcome these barriers. Through LabintheWild, we recruit unpaid online volunteers from all over the world to participate in behavioral studies in exchange for personalized feedback. Over the past three years, LabintheWild has attracted over 3.5M distinct visitors from over 200 countries and resulted in over 1.5M completed experimental sessions. We operationalized core theories of curiosity (Law, et al. 2016) and social comparison (Huber, Reinecke and Gajos, 2017) to attract intrinsically motivated participants. Our validation studies demonstrated that results obtained on LabintheWild match those obtained in traditional laboratory settings (Reinecke and Gajos, 2015). LabintheWild has made it possible for us to conduct research that would not have been feasible with traditional methods (e.g., a cross-sectional study with 500,000 participants showing how several aspects of motor and cognitive performance change during the life span, a study with 40,000 participants that characterized the impact of demographic variables on the perception of aesthetic appeal (Reinecke and Gajos, 2015) and a study with 16,000 participants that characterized the influence of the personality traits on the use of adaptive interfaces (Gajos and Chauncey, 2017)).

In our other work, we demonstrated how to use paid crowdsourcing to perform valid performance evaluations of user interfaces (Komarov, Reinecke and Gajos, 2013) and how to perform accurate measurements of pointing performance from in situ observations (Gajos, Reinecke and Herrmann, 2012). Our DERBI system makes it easier to report back personalized exposure results to the participants of large-scale biomonitoring studies (Boronow, et al., 2017).

I am broadly interested in what we can accomplish when we combine crowd and machine intelligence to empower individuals.

Like many others, I have done a number of projects with crowds hired on the Amazon Mechanical Turk: our PlateMate system demonstrated that crowds of untrained amateurs can perform nutritional analyses of meals as well as experts and our Mobi system provided a solution to the thorny problem of crowdsourcing problems with many inter-dependent subcomponents.

But, I find it even more interesting to look for ways to convince crowds of volunteer contributors to perform meaningful human computation tasks for free as a byproduct of activities that they are already intrinsically motivated to perform. In such "organic" crowdsourcing, people contribute to a human computation algorithm as a byproduct of engaging in an activity that is valuable to them in and of itself. Organic crowdsourcing goes beyond mere entertainment as a way to recruit participants. Instead, it requires the interactions and the underlying algorithms to be co-designed such that users naturally perform activities that reveal useful information to the system. We seek to uncover reusable design patterns to inform the design of future organic crowdsourcing systems. This is the approach we are taking in Organic Peer Assessment and Learnersourcing projects.

I aspire to build systems that make it possible to deploy effective pedagogical interventions at scale (e.g., Learnersourcing project) or in contexts where such interventions would be difficult to apply without help from technology (e.g., PETALS project).

I am rather disappointed with the state of the public discourse on accessibility. We are still fighting to make it even possible for people with impairments to access digital resources, but we fail to notice that we have created new sources of discrimination: In a world where work, education, civic engagement, commerce and social relationships all require computer access, we spend many hours every day in front of computer screens. In such a world, it is not enough just to make access possible. Instead, access also has to be equitable. That is, we need to enable access that is efficient (so that, for example, an injured college student can complete his computer-based assignments in a reasonable amount of time). It has become perfectly clear in my work that universal design is an elusive and dangerous dream. There is no way you can design a single interface to be efficient for an able-bodied user, a user with a severe dexterity impairment and a user with muscular dystrophy. You can, however, design an efficient interface for the individual with a dexterity impairment and a separate efficient interface for the individual with muscular dystrophy. Inefficient access is not inherent to those medical conditions. Rather, inefficiency is a result of a mismatch between the assumptions the designers make about their users and the effective abilities of users with impairments. For all those reasons, in most of my work on accessibility, I seek scalable mechanisms for delivering personalized and adaptive solutions that make the optimal use of what each individual user can do. This brief article published in the interactions magazine captures my vision for the future of accessibiltiy.