Human-Centered AI: From Interactive Tools to Agentive Partners in the World
Artificial Intelligence (AI) has demonstrated enormous potential for creating value and a better future for humanity. AI techniques are being used in a wide range of systems and services, such as in healthcare, where it can assist in surgery, drug creation, X-ray analysis, and patient tracking. In automotive, AI is the foundation for self-driving cars, but also in optimizing logistics, transport, and mobility. In the personal space, Alexa, Siri and other natural language interaction services are examples of ubiquitous AI-powered technology. And the list goes on: AI is found in Media, Finance, Insurance, Security, Retail, and Customer Service just to name a few sectors. But as with all technology, it has also shown a darker side – that of unintentional negative impact on individuals, organizations, and society itself. Algorithms that make decisions based on biased data can put entire populations at a disadvantage, and there are many other ethical aspects of using AI in our societies that need thorough investigation.
The success of AI-enabled services and products depends on our ability to design for positive human impact. As designers, we have always committed to designing human-centered products and services – using the best tools and materials for the job. And, as we shall see, these tools and materials change continuously, and now we’re entering a new digital design era. But in order to understand our future, we need to go back in history.
From Clocks and Steam Engines to Agents
During the scientific revolution in the 17th century, the hallmark of design (and the metaphor for the universe, and even animals and plants) was the clock. An intricate piece of design with cogs and wheels that, once started, moved like…well, clockwork. The key issue here is that apart from the winding of the clock, the ideal machine was a perpetual machine that predictably went about doing exactly what it was set to do. This is the domain of the classical Engineer. Carefully and meticulously designing a wonder of machinery that runs robustly on its own. This view continued during the industrial revolution, even though the focus shifted from clocks to steam engines and the harnessing of large quantities of energy to be used in factory production.
But in the 20th century, interactive technology emerged. This technology didn’t just operate blindly, with no contextual reference to the surrounding world. The new interactive technology — spearheaded by the programmable computer — was more responsive and required more and continuous input to guide its future actions. This gave rise to fields like Interaction Design, where the interaction loop between man and machine was in focus. Instead of an automaton, we started to focus on moment-to-moment interaction patterns in products and services that required human input. Filling out fields, tapping buttons, connecting on social media; the more we interact and “engage”, the better these systems perform. And now, with the advent of robust machine learning platforms, large data sets, viable sensor technology, and data distribution platforms and cloud infrastructures, ”AI” is rapidly becoming a powerful addition to the design repertoire. In one sense, we’re returning to the automaton model. In another sense, we’re at the same time advancing the interactive system model to include an even greater multitude of rich input, but where the moment-to-moment human input becomes less central. The new type of agentive services respond to our interests over time, and act on our behalf — without moment-to-moment human supervision. This leads to two obvious things:
- First, it allows us to off-load our cognitive abilities and focus on things that are more important to us than baby-sitting digital services.
- Second, by outsourcing decision-power and initiative to machines, we have less power and control over the unintended bad effects that an AI system can have.
Our goal as human-centered designers is to design and ship AI solutions that are fair, reliable, inclusive, transparent, and accountable. And this means that it has to start with the human front and center. AI technology should be used to enhance and augment human potential, and provide positive impact for indivudals, organizations, and society itself.
The success of AI-enabled services and products depends on our ability to design for positive human impact.
A New Design Material
In one sense, this is not new at all. Human-centered design has always been about that. And the methodologies and approaches we use to design have gracefully transitioned from text-based interfaces to graphical user interfaces, from stationary computers to mobile phones, from stand-alone machines to internet-based clients, etc. But it has so far always been about assistive technology and moment-to-moment tools. As mentioned above, the potential of AI technologies make possible new kinds of services — systems that act on our behalf, and take their own initiative. In short, we are transitioning from designing tools to designing partners. And suddenly, a whole new set of challenges surface; challenges that our traditional design approaches are not optimized for in a number of ways. Here are a few of them:
1. New Interaction Models
The fundamental input-process-output cycle of interactive tools has been modified, and AI-enabled agents take more or less their own initiative and operate more like thoughtful butlers or partners, so that humans can focus on other things. Designing agents with its own agency and initiative is different than designing moment-to-moment tools. This sort of behavior requires a whole new level of trust-building transparency and accountability, for example. Configuring and setting up the system requires advanced user modeling in real-time, which is different from designing a fixed set of functionality at “design-time” which is then consumed at “run-time”. The input model is also extended to go far beyond human-machine interfaces. The majority of input comes via other kinds of sensors and data streams, which need to be accommodated in the design.
2. Ubiquitous Augmentation
In addition, AI-enabled services may become almost entirely ubiquitous and invisible – merely augmenting human activity behind the scenes. No visible interface, no screens, no direct manipulation. Instead, our profoundly human everyday activities and cognitive abilities are seamlessly and automatically boosted. Even though we have a classical saying within Interaction Design that a good interface is an invisible interface, this takes that saying to a whole new level. For example: what is best-practice when it comes to good brain interfaces, and what are the design principles behind AI-powered augmented reality? How can generative AI be used to strengthen human capacity, and not replace or diminish it?
3. Adaptable and Inclusive
Furthermore, personalization and adaptation will play an even greater role in the design and implementation of human-centered AI. A learning agent can, if well-designed, adapt to the different needs, skills, and abilities of individual users. This implies a longer relationship between humans and AI agents than traditional tools. Maybe even life-long relationships will form? Longer customer life-cycles need to be considered, and in a much wider sense. It also puts the finger on the potential of an even higher degree of inclusive design for all.
4. Physicality and Society
And finally, in the wake of more advanced and accessible sensors, AI-enabled services will spread to – and affect –the core infrastructure of our world. IoT services, energy grids, transportation and mobility will all be affected by learning and agentive systems that act on their own to a larger degree. And even deeper into the fabric of society than that: digitalized democratic elections will be affected. We have already seen this in recent elections across the world: manipulating individual voters by analyzing data about them and exploiting their prejudices with targeted messages is an easy AI task. And in the even longer perspective, what if an AI assistant “knows better” than individual humans about what effects different political decisions will have in the long run? Would that make humans redundant in the society-building process…?
AI is currently surrounded with a lot of hype, and when the dust has settled we will probably face some disappointment in the coming months and even years. But in the long run, AI will profoundly affect infrastructure, jobs, and the fabric of society itself.
The Road Ahead
All this requires us to think differently and carefully about the design process, methods, interaction patterns, and the various impact these new types of AI-enabled services can provide. We know from history that we have a tendency to overestimate the effects of a new technology in the short run, and underestimate them in the long run. AI is currently surrounded with a lot of hype, and when the dust has settled we will probably face some disappointment in the coming months and even years. But in the long run, AI will profoundly affect infrastructure, jobs, and the fabric of society itself. Exactly how, no one can tell. But I am committed to carefully study, explore, and design human-centered AI-enabled services together with industry, government, and academia in order to provide thoughtful positive societal impact.