Call for participation: 10th Summer School & Symposium on Computational Interaction (S³CIX), University of Glasgow, Scotland. June 16-20, 2026.
Summer School expanded to include Symposium.
Call for Workshops (14th Feb) & Student Attendence (14th March). s3cix.computationalinteraction.org
The breakthrough insight behind ChatGPT was not just a better algorithm, it was a systems-level realization: the objective of the algorithm was reformulated from predicting tokens to predicting conversationally relevant tokens.
Further, we argue that systems modeling can not only produce better systems but drive innovation.
4 - Prototype futures worth studying
5 - Involve stakeholders as collaborators, not just evaluators
The paper revisits five principles for systems-oriented generative AI research:
1 - Draw boundaries that include both algorithm and people
2 - Map decision spaces across AI capabilities and human needs
3 - Measure how algorithmic performance affects user outcomes
...
Such causal relationships can only become visible when we draw system boundaries that encompass BOTH the algorithm's inner workings and the human context.
For example, how does a 1.5% improvement in model accuracy impact a designer's workflow? And why does it affect it, what is the causal chain that connects the two? We argue that systems approaches need to be revisited to help us answer questions like this.
The HCI community often treats AI as an opaque box, while ML researchers model systems without humans as active actors. Neither view is sufficient.
Instead, we need to trace the causal connections end to end.
Treating AI as a black box may be holding back progress.
Our fresh piece in IEEE Computer with @bjo3rn.bsky.social argues that to advance generative AI, we need to stop treating algorithms and users as separate concerns.
Feedback in RLHF is collected from human raters who are prompted by the system to compare pairs and express preferences. What advantages do we get if humans can exhibit higher agency? doi.org/10.1111/cgf....
Give us a model of the healthcare system and we'll do that.
Teaching HCI in Higher Education event starting:
lecture-series.hcistudio.org
Happy to announce that our CHI 2026 Meet-up proposal 'The Role of AI in HCI Education' has been accepted!
The goal is to serve as a catalyst for a community-wide discussion on how we ought to consider AI in HCI education going forward.
Please forward and join if you can!
I saw an exciting (or, rather, dystopian) workshop talk by Mark McGill. He gave a much more pessimistic view. In short, companies will control what you see, and it will mean more, not less, advertisements. And ones that are impossible to avoid. arxiv.org/pdf/2504.04440?
Launch of the new ELLIS Institute Finland. Excited to be part it!
One thing I’m especially proud of in our HCI intro book is that it includes an entire section dedicated to Engineering in HCI. As interactive technologies grow more complex, these methods are more important than ever.
academic.oup.com/book/60808
Aleksi’s thesis here:
aaltodoc.aalto.fi/items/cb4994...
User-in-the-box here:
github.com/User-in-the-...
We started collaborating with Aleksi some time after the onset of COVID19.
After two years of hard work, we got the first paper, “User in the Box” (UIST’22) accepted.
It still serves as the model for our models. It shows how to learn embodied control policies for HCI tasks using RL
Aleksi Ikkala on stage defending his thesis “Learning to Interact: Simulating Users with RL in HCI”, with Per Ola Kristensson as the opponent and Perttu Hämäläinen as the supervisor.
It's great to have Per Ola in Helsinki. We go way back. I met him in 2005 at the CHI doctoral colloquium. Since then we have worked on numerous projects on text entry and computational design. And, with Kasper Hornbaek, we wrote the HCI intro book, which took 7 years end-to-end. Time sure flies!
Prof Per Ola Kristensson / Cambridge giving a talk on how to design for the emergent properties of human-AI systems at Aalto. Per Ola is the opponent of Aleksi Ikkala who is defending his doctoral thesis on Friday, with Perttu Hämäläinen as the supervisor.
Yue's path takes her to University of Utah, where she is starting as an Assistant Professor. Congrats Yue!
One of the highlights of her work is Graph4GUI, a paper involving also Yue's second supervisor Vikas Garg, where we studied GNNs (Graph Neural Networks) as a way to learn layout representations in a controllable way.
So proud of our PhD student Yue Jiang, who defended her thesis "Computational Representations for User Interfaces" last Friday, with Brian A. Smith as the opponent.
Brian A. Smith giving his talk "Computers as Partners in Human Ability" today at Aalto. Impressive work. I especially appreciate his work on creating auditory interfaces that allow blind users play real-time video games like driving. Brian is Yue Jiang's opponent and visiting us today and tomorrow.
Just presented "WigglyEyes: Inferring Eye Movements from Keypress Data" at Ubicomp/ISWC'25.
It's a method that, given keypress data only, infers how users move their eyes when they type. It uses a user simulator for data augmentation.
dl.acm.org/doi/pdf/10.1...
github.com/quintus0505/...
Ubicomp'25 starting at Dipoli at Aalto. 793 attendees this year
This is the 2025 cohort of the Ubicomp Doctoral Colloquium. Some of the best PhD students in this area.
We're standing here in front of Dipoli, the beautiful main venue of the conference.
I look forward to hearing about these students as they move on after their PhDs.
