petr slovak at aha

(first speaker i listened to.. intro’d me to mit aha speakers on ai and advancing humans with ai (aha) and aha symposium 2025..)

via tweet @medialab [https://x.com/medialab/status/1958219736719765719]:

At 4:30pm ET on Thursday, August 21, join us for the next installment of the seminar series hosted by @AHA_MediaLab! Our guest speaker will be Petr Slovak, Reader (Associate Professor) in Human-Computer Interaction at @KingsCollegeLon.

from petra’s page [https://www.media.mit.edu/events/aha-speaker-series-petr-slovak/]:

Abstract

How might we design AI systems that support rather than replace human cognition, particularly in sensitive domains like mental health? This talk will explore our human-centered design approach to develop interventions that help users develop key mental health skills in everyday contexts. I will present insights from two recent award-winning papers (CHI’24, CHI’25) on gen-AI driven mental health intervention design, alongside 7+ years of research on Purrble—an interactive ‘situated’ emotion regulation intervention that has evolved from research prototype to commercially available socially assistive robot with multiple RCTs, over 100,000 units sold, and recognition as one of TIME Magazine’s “Best Inventions of 2021.”

Together, these projects show how new technical capabilities can let us rethink where, when, and how interventions are delivered: innovating on intervention ‘form’ while preserving (and sometimes extending) their psychological ‘function’. I’ll close by inviting discussion on both the opportunities and the risks that gen-AI introduces for human-AI collaboration in mental-health contexts, and how explicit alignment with psychological theory can guide responsible AI design in these contexts. 

Speaker Bio

Petr Slovak, PhD, is an Associate Professor (‘Reader’) in Human-Computer Interaction at King’s College London, with visiting appointments at the University of Oxford. His research blends user-centered design with clinical science to create digital mental-health interventions that fit seamlessly into users’ daily life. Petr has secured more than $9.5M in funding as PI/Co-I, including a prestigious 7-year $2.5M UKRI Future Leaders Fellowship. His research appears regularly in top-tier computer science and HCI venues and has earned 11 Best Paper/Honourable Mention awards in CHI/CSCW since 2017. Across projects, he focuses on designing Generative AI and embodied technologies that build users’ emotion-regulation and other mental health competencies—advancing human agency rather than replacing it.

@ozzulak

notes/quotes from talk/zoom:

pattie maes: intros petr

petr: talking about a long term agenda at work.. using simple techs to support mental health then going into more complex.. i’ve studied psych/socio and computer sci.. how can we build digital interventions to enable learning of mental health competiencies for the users.. esp as go about daily lives.. esp designing for ‘planned obsolescence’.. teaches you the skill then it goes away.. tension.. so focus for 7 yrs on research .. fosuc of interventions.. so combine psych/socio/tech.. systems that can combine these

petr: emerging tech esp gen ai enable new ways esp for mental health.. how to innovate w/o losing psych function.. psych thinks in theories of change.. but can change that into experiences.. ie: by pressing this in brain.. this will feel better.. psych focuses on intervention and intervention implementation

petr: underlying models we find helpful: capabilities ie: feature to express your emotion.. allows us to api out the psych problems we’re trying to solve.. each scope brings diff evaluation criteria..

petra: 2 case studies 1\ emotion regulation for children (stuffed animal bot) called purrble.. weird idea turn into stuff out in world.. what does it do? has bunch of sensors in it.. if hold it.. has a heart beat.. if fast.. need to take care of it.. so reduces anxiety/depression.. simple tech gone to clinical efficacy.. using it all over where patients have anxiety.. just counts if you moved it too fast and if you’ve touched its back.. why does it work.. taken a specific trajectory.. that people go thru automatically.. so tap into those psych process.. forces you to pay attention to it.. so it redirects your attention.. because you’re supposed to take care of it.. no longer thinking about self.. allows diff model for emotion reg.. w/them 24/7.. they decide to go back to it or not so learn it’s something they can do.. designed for 6 yr old but works for all ages.. so don’t have to go to sessions for weeks.. intervention is just along side you

petr: 2nd case study – helping people in articulating and reflect on emotional stories.. ‘micro narratives’.. asked most traumatic moment on social media.. don’t have ability to collect narrative data at scale.. then analysis

so bag the anal.. and use it for nonjudgmental expo labeling for imagine if we ness et al.. with no ‘driving question’ just listening..

petr: collect sm stories of 900 young people.. was helpful for them.. though they were hired just for their data.. which piqued our interest..

yeah.. what i just said.. let go of the anal.. what we need is ariana’s nonjudge ness..

petr: relying on vignettes as narrative structure to capture stories.. to do deep story telling in limited amt of space.. we turned this around.. they created first.. which pieces do you need human to do and which are you delegating.. who’s good/bad guy .. how should it end.. et al.. leave it to researchers to provide that template.. easy questions people could answer.. 3-4 questions then have skeleton of a story

ooof

petra: the thing works because we’ve understood this underlying cognitive trajectory.. easy for people.. automate aspects of narrative.. while giving individuals agency/control/revision.. not just getting them to say what they already had in their mind

petra: what else could we do.. beyond reflection et al.. that we could scaffold for people.. and what other types of intervention could we design.. ie: know process that is important/helpful.. transformation grounded in psych theory so could mimic.. so started thinking about starting with basic and level up like purrble.. for ie: delusion manipulation et al.. and develop new interventions

to me.. this is just perpetuating not-us (myth of normal) et al.. mostly because ‘thinking/sharing is prompted’.. rather than itch-in-the-soul ness et al.. but also because monitoring ness et al..

pattie: i like the minimal approach were person is to do the (thing)

meera in chat: have you looked into how the language of the bot affects how the participant responds to he question? ie: neutral or being expressive or like appearing to show a personality

to me pattie.. this is where we need the non judge ing part.. ariana talked about at symposium.. need to not lead the person.. if want it to be legit/lasting change

petra: they are very much context dependent

pattie: what if leads to expression in negative way..

petra: we’ve hired someone to help us think about this: physical health, mental health, perception of ai.. then link to types of stories/questions.. with safe guards et al

pattie: sometimes tricky.. requires a lot of context looking at whole history of convo

but shouldn’t be.. ie: graeber unpredictability/surprise law et al

petra: we don’t believe in free lunches.. allows you to go really deep but only in specificity.. we haven’t deployed interventions yet and will be very cautious when we do

petra: if question is about teaching people to articulate stories rather than collecting them.. deciding when to reduce scaffolding according to what they learn..et al.. gets so deep into process of learning.. once we have systems like this we can build machines.. by understanding details context..

to me.. you are teaching them if you have prompts

pat pataranutaporn in chat: design better systems.. how do you know which theory of change is good.. psych sometimes faces replication issues

petr: you are co producing story with person

to me.. that trumps authenticity

pat in chat: cog trajectories, decision making and nudging.. which is most applicable to hci (human computer interaction)

behnaz farahi in chat: are we really in charge? are we manipulated or regulated? or perhaps both?

all of the above.. if any form of any form of m\a\p.. decision making is unmooring us law et al

petra: do you mean w/in these systems or generally.. for me these questions are often content that is technically edible/helpful.. so more about what you do w/it afterwards.. create insights for you.. questions are pre selected so are manipulated.. don’t already have that template in your head.. so depends on what manipulated means.. if manipulated all the time.. you’re just choosing which manipulation

yeah that.. all over the place.. and we just keep talking over that issue.. need to try the thing we’ve not yet tried the unconditional part of left to own devices ness.. sans manip et al

petra: mostly doing this in clinical spaces.. so if want to build systems that work for people go to clinicians first.. trying to look at societal level..

petra: why i like thinking about scaffolding processes in very crowdsourced way that break it into steps that follow the flow rather than machine flow.. seems simple enough to be trustworthy enough.. ie: ask 3 questions and gather responses.. what is level of support now.. and how to get it further away from zero

ooof

rosalind has a purble: says thank you petr for showing how something very simple works

_______

______

______

______

_____

_____

_____