Everyday Technologies for Autism: Opportunities, Promises, and Costs

January 17, 2023

YCSC Grand Rounds January 17, 2023
Fred Shic, PhD
Associate Professor of Pediatrics at the University of Washington and Seattle Children's Research Institute

Information

ID: 9381
To Cite: DCA Citation Guide

Download Transcript

00:00I would like to welcome everybody here at the
00:04Council victorium and also everyone online.
00:08And before we begin, I have a
00:10reminder next week we have a very
00:13interesting talk by Doctor Stephanie
00:15Gilson who speak about the history,
00:18hope and healing within the
00:20American Indian and Alaska,
00:21Alaska Native communities.
00:23So please put it on your calendar.
00:27And today, it's my tremendous pleasure
00:30to introduce to you doctor Fred Schick.
00:33Doctor Frederick is associate professor
00:35in general Pediatrics at the University
00:38of Washington I at School of Medicine.
00:41He's the director of the Seattle
00:44Children's Innovative Innovative
00:46Technologies Lab and and and Jack and Jack
00:49faculty here at the Child Study Center.
00:52He also directs the eye tracking core
00:55of the Autism Biomarker Consortium
00:57for Clinical Trials program.
01:00Now Fred is coming to us from Washington.
01:03But he has a very long history here at Yale,
01:06and Fred also has a very interesting.
01:11History of educational history.
01:13So Fred was trained first as a engineer.
01:17At Caltech.
01:18And then he got his doctoral degree here at
01:22Yale in the Computer science department.
01:25And then both as a pre doctoral
01:27and postdoctoral fellow,
01:28he received training in developmental
01:30psychopathology here at the Yale
01:33Child Study Center in in my lab.
01:35I've known Fred for almost,
01:38I'd like to say 20 years,
01:40and Fred is truly a brilliant scientist.
01:44He's well recognized expert in his area,
01:47he's very sought after collaborator
01:49and I also know him as a very kind,
01:53generous and humble person.
01:57Over the past decade or more,
02:00Fred has been a leader in the field,
02:03that in the field that aims to
02:06bring the cutting edge technology.
02:09Into the field of developmental
02:11disorders in service of improving
02:14diagnostic and therapeutic services.
02:17Now,
02:17Fred's work in this area is driven
02:19by very deep expertise in terms
02:22of technological advances,
02:24computational science,
02:26statistics and developmental psychopathology,
02:29and the latter comes with a very
02:32important recognition or understanding
02:34and deep respect for the needs and
02:38hopes of the kids and their families.
02:41So we are here for a wonderful talk and Fred,
02:45welcome and it's wonderful to have you back.
03:02Did I break it? Nope.
03:04I broke it. My fault.
03:07I don't think it's anyone's fault.
03:08It's Zoom's fault, of course.
03:11There you go. Awesome.
03:12All right, so I just hit the share button.
03:15We're good, and we're all set.
03:17Sweet. All right. So thank you,
03:19coach, for that great introduction.
03:20It's. Sorry, what did I do there?
03:23It's it's really a pleasure to be
03:25back and see so many of my favorite
03:28people sitting here in the audience.
03:33OK. It's all right. All right.
03:36All right, so the title of my talk is
03:39digital divides meets digital bridges
03:41is still breaking it. I'm sorry.
03:43I'll. I'll stand over here, OK?
03:47And so just to to get started a little bit,
03:52I'm just going, oh, what happened?
03:54I have conflicts of interest. Receives.
03:57I do some consulting for Janssen.
03:59I did some consulting for Roche some.
04:01Just be upfront that anything I talk
04:03about that has to deal with drug,
04:05just remember the drug money.
04:07All right, sorry.
04:12Got it. If I wasn't so sleep deprived,
04:15I would totally remember why
04:16my next slide is all right.
04:18So. Give you a little bit of
04:21a a idea of my background.
04:23So actually I I started out after
04:27college thinking that I would just
04:29never have anything to do with
04:31academics ever again and started my
04:32career off as a video game programmer.
04:34And you know, I don't know if any
04:36of you know video game programmers,
04:38but it's pretty intensive position.
04:40You burnout really quickly and I
04:42didn't realize it at the time but I
04:44completely burned out and I decided to
04:46do something completely different that.
04:48You know, I slept under my desk
04:49a lot at the Sony PlayStation
04:51when I was when I was there.
04:53And so it really just decided that
04:54I was going to sleep under my desk.
04:56I would do it for something more worthwhile.
04:57So I went into research just on a lark.
05:00I actually worked for free for,
05:01I think it was like six months.
05:03They were really convinced that
05:04I was just going to go back to
05:05industry and I really loved it.
05:07And it was at the Huntington Medical
05:08Research Institutes where I was working,
05:09didn't doing some work on. On Mr.
05:13spectroscopy, looking at in vivo, Mr.
05:17spectroscopy,
05:18H11H spectroscopy of the the
05:20brain and decided that, well,
05:22if I want to keep on doing this
05:23research they need to get a PhD.
05:24And I was like, well,
05:25I like computers,
05:26so maybe I'll be a computer scientist.
05:28So join the computer science
05:31department under Brian Scaletti,
05:32who is a leader of a social
05:35robotics laboratory.
05:36And the thing that I was really
05:38attracted to for my reason for
05:40coming to to Yale was that.
05:42He was he really had this belief
05:45that social robotics could bridge
05:47between real life problems that
05:50that individuals with autism,
05:52children with autism had,
05:54and that these technologies could
05:56make a meaningful difference.
05:58And so my my task at the time was
05:59to build this visual attention
06:01system for this little baby robot.
06:03So it was a robot that was meant
06:05to look like and act like a baby.
06:07And in order to program that
06:10that baby robots Technic system.
06:11I came over to the Child Study
06:13Center and looked sad for.
06:15Must have been like a month
06:17trying to get some data gaze data
06:20from actual human babies.
06:21And and eventually I met Kasha and
06:25Kasha and I worked out a deal.
06:28Which was that I was kind of, you know,
06:31I need some data from babies.
06:33And she was looking for someone to
06:35to help set up an eye tracking lab.
06:37And the two of us started working
06:39together and I I think I I that was it.
06:41That started my entire career.
06:42I never looked back.
06:44That was a credit card for the formative
06:46experience of the research that
06:48really gave my life's work direction.
06:51And so after all of that working
06:54in eye tracking and.
06:56Working in the field of autism research,
06:58eventually, of course,
06:58you know, as things happen,
07:00came back to thinking about video games.
07:02And so even though there's a lot of
07:05different types of research that I'm really,
07:07really interested in,
07:08I'm going to focus primarily
07:10on technology in this talk.
07:12So now my laboratory in Seattle
07:14is at the Seattle Children's
07:16Innovative Technologies Labs,
07:17and we focus on using technologies
07:19in order to address real
07:21problems that individuals with
07:23developmental disabilities may have.
07:25And so we use.
07:26A wide range of different
07:28types of technologies,
07:29but most of these technologies focus
07:32around the sole condition of autism.
07:34So it really is focused on this
07:36idea of building technologies
07:37to address the core social,
07:39communication and interaction deficits
07:41in autism under the consideration
07:43of the restricted and repetitive
07:46behaviors that define the condition.
07:48And who do I have to admit this
07:50person into the waiting room?
07:51Oh, but I I really feel like I should.
07:53So, so I did all right.
07:55So, so why why technology?
07:58Why are we even here talking
07:59about technology?
07:59Well there's there there's
08:01reasons to believe that the use of
08:05technologies in therapies and in in
08:07using them to to have some type of
08:10clinical impact for children with
08:12autism may be special in some way.
08:15And the reason why we consider
08:16that it might be special is a
08:18long standing observation that
08:20many children with autism.
08:21Gravitate towards what we might
08:23think of as physical systems
08:25that are mechanical in nature,
08:26that have constructive component to them.
08:30Some work that was published quite
08:32a quite a while ago by Turner,
08:34Brown and Bodfish.
08:37Revealed that a majority of children
08:40with ASD had this predisposition
08:43to be really interested in
08:46this category of physics,
08:48which includes things like cranes,
08:50mechanical functions, trains, Legos,
08:53other mechanical systems and the.
08:58The thing about this specific
09:00type of interest is that we can
09:03really think of it as being.
09:06Used to catalyze additional
09:08types of treatment,
09:09additional types of of kind of
09:12powerful effects that can lead to
09:15positive change in children with ASD.
09:17So I'm going to start kind of where I began,
09:20which is in robots.
09:21And the robots that we talk about
09:23when we think about autism are special
09:26breed of robots called social robots.
09:28And these are robots that are as
09:30compared to like the robots that
09:33assemble cans and simple cars are ones that.
09:35Are meant to interface with people
09:38in a very natural manner and to
09:40address issues and and we can
09:42think of it as a natural interface
09:45of social interaction,
09:46but also to really focus upon and
09:49sometimes address core interactional
09:51questions in atypical development.
09:55So one of the first studies that was
09:58involved with was this study social
10:00robots as embedded reinforcers of
10:02social behavior in children with
10:04with autism and this was really let up.
10:06By someone who would later be a postdoc in my
10:08lab, Elizabeth Kim,
10:09as well as a host of really talented
10:13individuals including Ria Paul
10:14and of course my advisor Brian
10:17Scassellati and this this study,
10:20which was one of the first studies.
10:22I think it's hard to believe that it's
10:24been a decade since it came out really
10:27was one of the first show that using
10:30a robot in conjunction with another.
10:32Another examiner.
10:33Led to higher rates of communicative
10:37outputs as compared to if that
10:40robot was replaced with a person
10:42or a computer video game.
10:45Granted, 10 years ago,
10:46the video games were a lot different.
10:49But this idea that that.
10:53Robots could be something special.
10:54Technologies could be something
10:56special in individuals.
10:57Autism really is is exemplified
10:59by now a whole host of different
11:01types of research studies.
11:03But I'll give you one example which
11:06is work done by Laura Buchan Fuso,
11:08who was in my lab and now leads
11:11a company called Von Robotics,
11:13which is focused on building
11:15robots for education.
11:16OK, so this is a simple study where the
11:20child is given this task of correcting.
11:24Um.
11:26Actual mistakes in the conversation
11:28by the experimenter and or or the
11:32interactional partner and the.
11:33And you can see that that in this
11:36case where Laura plays the role of
11:39the interactional partner, this,
11:40this child is wholly uninterested.
11:43Now if we replace Laura with this robot,
11:47and this is a.
11:49Really simple robot called LED,
11:51a big characteristic of LE that's
11:55actually has a mouth that moves.
11:58And the reason why we created robot with
12:01the physical mouth was because we knew.
12:04We knew at the time through a lot of
12:06the work that that Kasha and I had
12:08been doing together that the mouth
12:10is actually an area of really key
12:12importance in terms of the looking
12:14patterns of children with autism.
12:16So you can see they start difference.
12:19You're right.
12:19The child's face lights up when
12:20he sees that robot begins to talk,
12:22and it really tells you something
12:25about the a very different kind of
12:28interaction that is catalyzed by this robot.
12:31So.
12:32Since that time, you know,
12:35a lot of additional work has
12:37come out in in robotics,
12:39the use of social robotics
12:40with children with autism.
12:41And it's gotten to the point where,
12:42you know,
12:43for a long time we were talking about
12:44what the evidence will get here.
12:45The evidence will get here.
12:46And well,
12:47I think now finally the evidence
12:48is starting to mount and it's
12:51really becoming powerful.
12:52And then it's believable that there is
12:54really something special about social robots.
12:57So this is a study that was done in the
13:00Netherlands by Van Vanden Brooks Meekins,
13:03who's had to pause.
13:04With that name so long names and
13:06this was a study that basically
13:08looked at pivotal response treatment
13:10with and without robots kind of as
13:14a addition to the intervention and
13:17when the robot was in the intervention.
13:22What happened was that we saw
13:24that the primary outcome
13:26measure which was apparent reported SRS
13:28ended up being much lower at the follow
13:31up as compared to both the condition
13:33where there was no robot and the condition
13:35where there was just treatment as usual.
13:37Now something to be to note here if
13:40you're carefully looking at those graph
13:42is that there's no difference between
13:44the PRT line and the treatment as
13:46usual and so that is a sticking point,
13:48but as it turns out.
13:51The studies of treatment as usual,
13:53as all of us know, are very different
13:56now than they were ten years ago.
13:59And so the treatment as usual actually
14:02also contained a lot of active ingredients
14:05that are probably pieces of PRT as well.
14:09But the key thing here really was that
14:12this is some of the first evidence in
14:15a really nicely controlled randomized
14:17controlled trial that there are improvements.
14:21Greater gains by children who are
14:23in pure T condition with robots
14:26augmenting the intervention.
14:28The way that these robots worked
14:30with the the child was that this
14:34was apparent directed at PRT,
14:36our parent parent pivotal response training,
14:39and the robot was used in order
14:43to model the types of behaviors.
14:45That's the things that should be rewarded
14:48and how to play these games with children.
14:51As a way of illustrating how these
14:54games should happen and another thing
14:56that was was impressive about this
14:58study was that we actually saw a
15:01greater proportion of children in the
15:03PRT plus robot condition who showed
15:06a positive improvements in the ados 2
15:09severity scores at at the end point.
15:14So the thing to notice here is that when
15:16we're talking about like this key, this
15:19inherent motivational property of robots,
15:21that this is really beyond what we might
15:23think of as a simple reinforcer, right?
15:25Like this is not just something that
15:27a child really wants to interact with,
15:29and so we'll do something in
15:30order to interact with the child.
15:32The difference here is that the robot
15:34can actually carry specific information
15:36that is key to delivering a specific
15:39type of training that is instrumental.
15:42In the ultimate positive gains that
15:44we want children with autism to go
15:46through as they go through these
15:48different types of interventions.
15:50And really this doesn't just apply to robots.
15:56Technologies in general can carry more
15:58of these active ingredients of therapy,
16:00which is the question of how do we do it.
16:02So another type of platform that we
16:04can think of as perhaps inherently
16:06motivating to children with autism,
16:08individuals with autism in general,
16:10are video games.
16:11And so there is now a lot of evidence
16:15to suggest that video games that are
16:19intended for the purpose of remediation
16:21of skills or psychosocial education,
16:24or specific addressing of areas
16:29of of areas of performance that
16:32individuals with autism have,
16:34that these serious schemes can
16:37provide some type of improvement.
16:40And here's just the.
16:41Kind of a panel of a few different examples
16:44and so for instance on the left we have a a,
16:47a face, a face processing and
16:51face information processing game.
16:53And. And.
16:56We see positive gains in the SRS.
16:59This is also a randomized control trial.
17:03See positive gains in the in the s s,
17:07two parent totals T scores as
17:10compared to in the experimental.
17:13You know, randomized to.
17:16Randomized to this this face face,
17:18say training game as compared to the
17:22control group who was basically given a.
17:26A drawing activity on a on a on a tablet.
17:31And.
17:31Here's another example.
17:32This is a connect video game
17:35where actually there is a mouse.
17:37OK great,
17:38a connect video game which really
17:40focuses on a lot of joints,
17:41play skills and single subject design.
17:44Showing good improvements in performance
17:46of the task was meant so that the
17:50the performance would increase as
17:51they were able to accomplish more and
17:54more of the tasks in the in the game.
17:57And this is a whoops this.
17:59Is a more recent report by Jason
18:03Griffin and Suzanne Scherff and Jason
18:06I don't know if Jason is is here.
18:09Jason good to see you.
18:11But I specifically put this into
18:13to highlight what an awesome job he
18:16did on this study and you know this
18:18is this is I think one of the first
18:22examples of using a serious game.
18:24We call them serious games because
18:25the goal isn't to just have fun the
18:27game the goal is to provide some
18:29therapeutic or some. Alternative gain.
18:33It's one of the first examples to
18:35show that the application of of
18:38these games can lead to changes in
18:41just basic attention in in
18:44terms of really important.
18:47Skills for for focusing on
18:50gaze correctly at stimuli.
18:52So really, really.
18:54You know, a lot of evidence amounting.
18:57Again, just like with robotics,
18:59you know, ten years ago I would
19:01have looked at this literature.
19:02You could read any review.
19:03And that review would say,
19:04like we need more information,
19:05the data is not good enough,
19:06the studies are too small and
19:07that's really not the case anymore.
19:09We're really seeing kind of a robust
19:12literature now with meta analysis,
19:14really formal mental analysis,
19:16suggesting that there's these really real,
19:19there's real utility in deploying
19:21these serious game interventions.
19:22The effects are relatively modest.
19:25G of .42 in this meta analysis of
19:2811 studies involving 654 children,
19:31but 654 children,
19:33this is a really different place than
19:35we were five years ago and certainly
19:3710 years ago where we would see,
19:39you know,
19:39maybe if we were to count
19:41up everyone added together,
19:42you might see like 80 people.
19:44So,
19:45so that that's really impressive now
19:47that we have this idea that that these
19:49are things that can really work now,
19:51the next hurdle that we're going to
19:53go through in terms of understanding.
19:55How to where to go with these video games
19:57is how do we actually develop these
20:00video games so that they are effective.
20:02And so a recent,
20:04you know these these two studies
20:06came out at kind of a little
20:08bit far apart from each other.
20:09But this this second study here
20:12focused on kind of a slightly
20:15different set of games but focused
20:17more on what are the components that
20:19are associated with these positive
20:21gains and what they found was that
20:23adherence to what they call social.
20:25Serious game principles,
20:26right?
20:27These are these rules of how do
20:29you build a game that's engaging,
20:30that will actually challenge the child?
20:36Adherence to these rules was,
20:38you know,
20:39perhaps not unexpectedly associated with
20:41how well these these systems would work.
20:43And so we're starting to get some information
20:46about what are the active ingredients.
20:48And it's important to note though,
20:50that there is of course,
20:51from the other side of things,
20:53this really well known series of
20:56findings that video games can
20:58be actually a source of concern
21:01and difficulty in terms of play
21:04until obsessive play excessive.
21:07And so these kind of serious game principles
21:09have to be balanced against those,
21:11those critical and unique characteristics
21:13of individuals with autism as we think
21:16about how we want to deploy them.
21:18Now.
21:19Thinking about video games,
21:20of course brings us into kind
21:22of this wider space of thinking
21:23about all these different types of
21:25mobile applications that can be
21:27leveraged in order to promote skill
21:28gains in individuals with autism.
21:30And the most obvious interactive
21:33technology is perhaps one of the
21:35oldest technologies which are.
21:37Our augmentative and alternative
21:38communication systems and this is
21:41just an example of a simple speech
21:43generating device per loquito
21:45go. It's probably the most successful
21:47AC digital system out there.
21:49And again, you know the knowledge
21:54that we have about these systems has
21:56just really increased tremendously
21:58over these last last few years.
22:00This study I like a lot involves 9 single
22:04case studies involving n = 36 children,
22:08which is, you know,
22:11relatively modest but the.
22:13But the take away from this
22:16really kind of spoke to what?
22:19Is unique about these technology
22:21based AAC systems which is that
22:24most of the participants who were
22:26in this cohort preferred the SG,
22:29these speech generating devices,
22:30these digital speech generating devices
22:33but also performed better as compared
22:36to standard picture exchange with the
22:38cards and as well as manual sign.
22:40A more recent study that came out
22:46recently was actually a meta analysis.
22:49Of a much larger cohort and.
22:55I forget the total number,
22:57but there was I think 114 participants
22:59and largely found the same types
23:01of results that that there seems
23:03to be some type of advantage for
23:06augmentative these digital augmentative
23:08alternative communication systems so
23:10that that particular meta analysis
23:14didn't distinguish between what they
23:17called low tech versus high tech.
23:19AC systems.
23:22Now I think one of the most compelling.
23:25Pieces of evidence about the unique
23:27nature of these digital AC systems,
23:30and AC systems in general,
23:33comes from the work from Connie
23:35Kasari and colleagues looking at
23:37the impact of including a speech
23:39generating device together with a
23:42naturalistic developmental behavioral
23:43intervention in children with ASD.
23:46And it was a it was a well powered
23:48randomized controlled trial,
23:4930 children who were basically
23:52given the enhancement blue.
23:54Winning plus Jasper and then 31
23:58children who were given the same thing
24:00except with a speech generating device
24:02and the the critical thing that you
24:05can see here this is total social
24:07communicative utterances shown on the
24:10Y axis that overtime that's the the.
24:14You know the.
24:16The group without the speech generating
24:19device essentially increases,
24:20improves as we would expect,
24:22but the gains that we see in the
24:26speech generating device group
24:27are much are much larger at the
24:31beginning though they end up.
24:33But Connie Cassari is is is a quick to
24:35note that they end up at the same place.
24:37So.
24:38So there are some gains in terms
24:41of the speed at which learning is
24:44taking place by the incorporation
24:47of these speech generating devices.
24:49You know,
24:49I think that we'd all love to see
24:51more information about this and
24:52try to really understand what are
24:54the fundamental mechanisms that
24:55are underlying this boost in in
24:57learning and perhaps maybe we can
24:59even make this more effective we can.
25:02Maybe even cut off the intervention
25:04shorter in this way.
25:06So some when we think about these
25:08advantages that are appearing it's you know,
25:10it's one thing to think about the
25:13inherent increased interest as a
25:15catalyst for increased engagement
25:17which leads to you know as we know
25:20greater efficacy in treatment response.
25:22But it's also another to think
25:23of perhaps that there
25:25are specific types of mechanisms,
25:26specific ways that.
25:27Children with autism may think that
25:30may lend themselves to technologies and
25:33especially these really visual these
25:36stimulating technologies like tablets and
25:38mobile or video games and mobile apps.
25:41And so one idea comes from
25:44actually a host of literature,
25:47but I think that's my colleague Kunda and
25:49Ashok goal actually really summarized
25:51this and kind of a formal theory,
25:53this idea that many more many
25:55children with autism are much more
25:57visual learners than they are.
25:59Certainly auditory learners right there,
26:02they're particular learning style
26:04is really one that may deal with
26:07very concrete imagery.
26:08And so we can leverage this to think
26:11about other strategies in which
26:13we can take this perhaps natural
26:15predisposition towards visual thinking
26:17and apply this to app development
26:20to address specific skills.
26:23This is a project called speech
26:25prompts that was launched together
26:27with the Real Paul and Liz Simmons.
26:29Liz Simmons.
26:29He was the one who was the,
26:31the, the I would say she,
26:33she was she was the boss of the study
26:35and incredible incredible thinker
26:37about how we could design apps and.
26:40I think the.
26:43Purpose of speech prompts this program
26:45speech prompts was to address pros
26:48the issues in children with autism.
26:50Prosody being the way that you say
26:51something and so perhaps you're
26:52speaking too loud you're speaking
26:54too much in a monotone kind of
26:55like I'm doing right now.
26:56It's like the heads nodding
26:58I'm sorry about that.
26:59That's just the way I speak the
27:03OR or maybe you're it's it's too
27:06excessive you might be using too
27:09much inflection and so speech
27:11prop was really meant to take.
27:13These concepts that like how how would
27:16you teach someone that you know?
27:18The pitch is too high.
27:19You give examples.
27:20What if you don't really,
27:21you can't really hear those books,
27:23right?
27:23Like that's the whole problem is that
27:25you don't hear it in the same way
27:26that a neurotypical would, right?
27:27And so this really exists as an
27:29alternative way to visualize these
27:31really difficult concepts, right?
27:33To take things that are very
27:34ephemeral in nature, right?
27:36Like this, this utterance,
27:37a sound way that we speak very
27:40short lasting and time.
27:41And to give it a concrete form to
27:42give it something that you could point to,
27:44just give an example that is reproducible
27:46that the child can also participate in.
27:49Himself or herself,
27:50so that they, he,
27:52he or she can.
27:54Experiment with these parameters and
27:56learn these mappings on their own
27:59using the set of technologies as an aid.
28:01The result of a kind of a preliminary,
28:04pretty small feasibility study involving
28:0710 speech language pathologists and
28:0940 children special education program,
28:12all with prosody difficulties,
28:13but not all with autism.
28:15I think the majority,
28:16about 80% of them had autism,
28:18showed really high levels of engagement,
28:20compliance and use of the system.
28:22And so that was a good thing,
28:23right?
28:23That's actually the the critical
28:24part when you're thinking
28:25about these technologies.
28:26Find them, because if no one's using them,
28:28then they're not going to work right.
28:29Like that's the.
28:29So that that is the first thing
28:31that you need to measure, and then.
28:35What the children do is of course important,
28:37but even more important was that the
28:39speech language pathologists felt
28:40very positive about these systems.
28:41Because even more important than the
28:44children enjoying themselves are
28:46that the interventionists think that
28:48these technologies are of value,
28:50otherwise these things will not get used.
28:52In terms of the actual prosodic
28:55qualities of speech, we saw essentially
28:57improvements across the board,
28:58which was exactly what we want to see so.
29:03So these apps, these apps that
29:05that capture visual information,
29:06that are transformative
29:07of visual information,
29:08they are of course really powerful,
29:11but we can even go simpler if we want to
29:14think about using visual information.
29:15And here is an example that actually.
29:21I. I'm going to date myself, right.
29:23But like,
29:24I was in middle school at the
29:26first of these studies, 1986.
29:28All right, so, so weirdly,
29:30we've been thinking about this technology,
29:33which simple technology called
29:36video modeling for, oh, gosh, 86.
29:38What is that, 14?
29:39I don't know, several years,
29:4130, several decades,
29:43we'll call it several decades.
29:46So,
29:46so long ago that they used these videotapes.
29:48Do you do you know that you
29:50guys all know videotapes?
29:52I'm sad because I know there's some of
29:54you who have never seen a videotape.
29:56My daughters made fun of me when they saw,
29:58saw our collection of videotapes
29:59that my my mother's house.
30:01She asked, you know if a.
30:04If I had a pet dinosaur at that time in US.
30:08Yeah, I have to admit that I did so, so.
30:11You know, we've been,
30:13we've been thinking about this,
30:15the simple uses of technology using
30:17technologies for a long time in
30:20order to leverage this specific
30:22property of being able to process
30:24information visually more readily,
30:26more facilely than in traditional
30:29ways that that teaching may be done.
30:31And there's a now a really large body
30:35of evidence suggesting, you know,
30:38as we expect that there are some
30:40really positive gains to be had through.
30:42Video using using video based modeling
30:45as compared to traditional traditional.
30:50Teaching methods,
30:51but also in combination.
30:53There's no reason why you have
30:55to pick one or the other.
30:56And I'm going to give you also this
30:58simple example that we did together
31:00with Ben Popper who is at the
31:02the Department of Dentistry while
31:03while I was here at Yale and also
31:07Kelly Powell and and Fred Volkmar.
31:11This is just a simple example of
31:13of of kind of video modeling and
31:16this was a toothbrushing video
31:18that was shown to children.
31:20Basically of over three weeks
31:22every time the child brushed.
31:24And Kelly Kelly, are you here?
31:27No Kelly.
31:28So Kelly.
31:29Kelly was the one who told us
31:31that this actually you know so.
31:33So this study was a was
31:35also a controlled trial
31:36right. So one group of kids received
31:38this this toothbrushing video and
31:40another kid was another group of kids
31:42was asked to watch a controlled video
31:44involving fractals and some some
31:46sound almost like it was meditative.
31:49And and what we see here is
31:51the the week week one week two,
31:54week three and the plaque index is
31:56shown on the Y index and you can just.
31:58See this kind of strong structure
32:00that's appears in this toothbrushing
32:03video presentation at as compared
32:05to the fractal movie.
32:07And the reason why I want to highlight this
32:10is that if you if you look at these right,
32:13this is a simple one minute video, right.
32:16And Kelly Powell at the time conveyed
32:18to us that one of the parents in the
32:21study after being through this was was
32:24in the experimental arm acknowledged
32:26that she basically she said.
32:28You saved my life.
32:29This was such a a big ordeal for us
32:33every night and this really helped
32:35us get through and learn how to.
32:38Get through brushing without incident.
32:40And so I mean we think about these as
32:42kind of like a a trivial exercises,
32:44right?
32:44Like this is a simple thing to do as
32:47as you can imagine running an actual
32:49study is of course more work than, than,
32:50than it looks like from the outside,
32:53but in terms of just small.
32:57Types of studies that we do,
32:59small activities that we can do that can
33:03elicit real gains in a family's well-being.
33:05I mean I think that these are
33:07the exactly the types of things
33:08that we should be doing more of.
33:09And in fact if we go out and go to
33:11YouTube and you start searching for all
33:13these different types of activities,
33:15for instance toileting,
33:16toileting is of course one of the the
33:19really difficult areas for for some
33:21families with children with autism,
33:23you will find a lot of these different
33:25content and the the problem that we have.
33:27It's not that that's the content isn't out
33:30there. It's one of the logistics, right?
33:31Which content do we use?
33:33How do we find what we need?
33:34How do we what?
33:35What are the strategies that we
33:37deploy these with our kids and
33:39know that something's not working,
33:41that we need to switch to something else?
33:42How do we start to get a sense of
33:44what will work for our children?
33:47And you know the the the world is so
33:50different from those 1986 studies.
33:51So in the the studies even even
33:54as late as in in the year 2000,
33:57they were really there.
33:58There was a line in one of the
34:00probably the most cited video
34:02modeling paper which which actually
34:04considered how much does it cost
34:06in order to film these videos.
34:09And we don't even think about that now
34:11because we know that it's pennies and I
34:12can tell you that with the the number
34:14of selfies that my teenage daughters.
34:16It's just like it's.
34:19It would have been much
34:20more expensive back then.
34:21So, so these are things that have changed,
34:24the landscape has changed.
34:27We have these changes in the
34:29convenience and the cost of
34:30both consumption and creation.
34:32And so we can generate all these videos,
34:34right? Like this simple video
34:35of toothbrushing, right.
34:36They made such a big difference.
34:37Of course, it was created by experts.
34:39And that's The thing is it's we we have
34:41a really big problem figuring out how to
34:44get that content into the right place.
34:46And so that is one thing that
34:47we really need to address is,
34:48is figuring out ways of organizing
34:51information so that we can share
34:53them and so that we can start to
34:55help parents identify on their own.
34:56Because the.
34:57The problems that that parents are
34:58having are sometimes they're quick, right?
35:00Like this, this just happens,
35:01started happening now.
35:02Maybe it's it,
35:03maybe it's gone in a few weeks,
35:05but if it's not,
35:05you're going to have a really hard time.
35:07These are things that we need to
35:10develop better infrastructure for.
35:12So thinking about video modeling takes us
35:14to kind of a next set of technologies.
35:16And these are kind of more
35:18recent technologies.
35:18They are well explored,
35:19but they're not used that often.
35:21I don't know how many of you go
35:23home and put on a virtual reality
35:25system just to relax or something.
35:27The truth is that it's super weird, right?
35:29Like, you're sitting there,
35:31everyone's like, what are you doing?
35:32Feel funny, people are coming in behind you.
35:37But there's a real purpose
35:39to these technologies,
35:39and we think about in terms of therapy,
35:41especially for therapy in individuals who
35:45may have broader attentional difficulties.
35:48One is that it really does
35:50reduce distractions, increasing,
35:52you know,
35:52consequently increasing focus and the
35:54other is that they really can provide
35:57these really controlled environments
35:59to deliver really specific content.
36:01And again the field is maturing A33
36:05studies now with involving over
36:07540 individuals with ASD a large.
36:11Gee,
36:13from the meta analysis that that's
36:15shown here is an especially in the
36:18impact seems to be especially strong
36:20for things like daily living skills.
36:22So these these what we might
36:24consider these more mundane,
36:26but we know they're anything but
36:28mundane for individuals with autism.
36:29That seems to be where these virtual
36:32reality systems really are shining.
36:34So.
36:34Some from virtual reality is an
36:36even newer type of technology
36:38called augmented augmented reality
36:39and this is one that many,
36:42many people probably haven't experienced yet.
36:45Augmented reality works a lot
36:46like virtual reality except that
36:48you are still in the real world.
36:50So basically you could put on a
36:52for instance the Microsoft HoloLens
36:54is a is a really nice example of
36:57of augmented reality systems where
36:58basically you put them on and you
37:00can actually see the scene in front
37:01of you can see everything in front
37:03of you and what the technology does,
37:05it allows.
37:06The computers that are embedded
37:07onto these systems to put
37:09artificial objects artificial basically to
37:11manipulate the environment in front of you.
37:14And that creates an entirely new way
37:16of thinking about how we can address
37:18interventions that kind of has, you know,
37:20one of the one of the concerns about
37:22something like virtual reality is that
37:24it's not really life like in some ways,
37:27right, especially if you're
37:28talking about computer avatars.
37:30But you know you're surrounded by system and
37:32you know that the environment around you
37:33is not real and you can't also, you also.
37:36Aren't interacting with other
37:37people in kind of the live fluid
37:39way that you would in real life.
37:41So there's a generalization
37:43barrier that exists there as well.
37:45And so augmented reality may
37:46help us get around that.
37:48So I'll give you a really quick
37:50example that we did in our lab,
37:52just focused on on augmented reality.
37:57And this,
37:57this is kind of a simple task
37:59where an experimenter is trying
38:01to engage the child and the child
38:04is wearing this one of these.
38:06The hollow lenses and some that was a test.
38:09This is that was the baseline task
38:11where basically you know there's
38:12just the the the experimenter trying
38:14to get the attention trying to to
38:16interact with the child and this is
38:20another condition where when the.
38:23Experiments or space comes into
38:26view instead of. Oops. Instead of.
38:32Nothing happening this,
38:32this is actually what the child is seeing,
38:34right?
38:35So look,
38:35the star appears over the face and the
38:37idea was to explore whether or not
38:38there was kind of like this active
38:40barrier to looking at the face.
38:41What we saw was just was just no
38:43indication that the the child was interested.
38:46And so this is a kind of a final
38:48condition where we basically put a
38:50box over the face and this is like,
38:52it's just a simple idea, right?
38:53You're just highlighting where the face is.
38:55And this is a child who was nonverbal
38:57as about 7 years old.
38:59And we could see that this.
39:02Type of very simple type of strategy
39:04really can help provide some visual cues now.
39:07Now of course you know the child
39:09is looking at the box, right?
39:13May not be looking at the face,
39:15but we all know there's ways.
39:16Once you get there, you can start,
39:18you can work with that, right.
39:19You can work with the child is
39:20orienting towards the person.
39:21And now we can,
39:22we can go on from there and start
39:24fading those, those stimuli.
39:25We can change them around.
39:27And so it opens up a wealth of
39:29possibility that perhaps would have
39:31been very difficult to achieve otherwise.
39:33Other examples of augmented reality
39:36include this work by Dennis Wall
39:39and and Caitlin boss and this is
39:42a study that uses Google Glass,
39:45which I don't even think that they
39:46make these Google glasses anymore.
39:47So the Google Glass was basically
39:49these kind of a little headset that
39:51you would wear and they're in the
39:53corner of the headset.
39:54You can see this little little area
39:57where there's actually a camera that
39:59essentially projects directly onto your
40:00eye and letting you see a little small.
40:03Heads up, heads up, heads up display
40:05in kind of the corner of your eye.
40:07And so what's what Dennis and his team
40:10did was that he built this system so
40:13that a child could play with one of his,
40:17his, his or her parents and they
40:19would engage in a game of of kind of
40:22replicating each other's emotions.
40:24And one of the things that that this
40:26technology did was that depending on what
40:29the emotion that the parent was doing,
40:31it would basically provide explicit.
40:33Information to the child about
40:35what that emotion was.
40:37And the result of this was that
40:40after going through this this,
40:42you know, series of 20 minute games,
40:44expected to be 4 * A Week,
40:45I think they quite got there,
40:46but that was the plan for six weeks.
40:49He saw positive improvements in
40:52the Vineland socialization scores,
40:55which was I think. Quite nice to see.
41:01So one of the first examples of augmented
41:03reality actually having a formal impact.
41:05Now we think about augmented reality, right,
41:07providing information that children with
41:09autism may not have had readily on their own,
41:12right.
41:12So like this motion information
41:13can be difficult to have,
41:14but we also didn't think about.
41:16There's also a lot of things that
41:18are developing that are things
41:19that we never really expected.
41:21So for instance,
41:22augmented augmenting abilities is
41:24something that that is possible
41:26with the right types of modalities.
41:28This is just an example.
41:29It's kind of a.
41:30Almost a thought experiment because I
41:31haven't heard more about this technology,
41:33but I just love the idea.
41:34So this is this is a wearable
41:37interface that basically is a series
41:39of electrodes that goes around
41:41the the the throat and the face.
41:44And what it does is it allows
41:45you to subvocalize and then it
41:47turns those sub vocalizations,
41:48which no one can hear into speech.
41:51And there's a it's actually
41:53fairly complicated to do this
41:54because as you can imagine,
41:55like all the things that you would expect
41:57to be happening when you're subvocalizing
41:59to yourself are not really happening.
42:01Right.
42:01And so it's basically a machine learning
42:03model that figures out how to map
42:05these really minute muscle movements
42:07that are so slight that you can barely
42:09even see them into actual speech.
42:11And this is really some type of
42:13technology that that we think could
42:15be really valuable for not speaking
42:17individuals with autism as this thing
42:19from nonverbal individuals with autism,
42:21not speaking individuals with autism.
42:23This may provide a route for
42:25speech depending on the nature
42:27of the particular difficulties
42:29that are responsible for them.
42:31Not speaking and so.
42:32These technologies are not just
42:34providing kind of new information
42:36and so that's extremely important.
42:38It's also providing new ways of
42:40conveying for the the individuals with
42:42autism to find a voice for themselves.
42:45So. These type of technologies in in
42:48conjunction with each other really point
42:51towards this idea of social prosthesis.
42:53This this concept of social prosthesis
42:55is is is actually fairly old and
42:58the idea is that that there's a
43:02collection of technologies that
43:03enable interaction in a social,
43:05in a in a social fashion with
43:08others that basically.
43:09Revolves around providing access to skills
43:13that otherwise that individual would not
43:17have and this is something that is is
43:19relatively new in terms of development.
43:21We haven't seen a lot of these type of
43:23technologies but they include things
43:25that help read the emotions of other.
43:27We can imagine that as time goes
43:29on we could have have systems that
43:31tell us if if some some like what,
43:33what is the nature of this particular
43:36interaction and then even beyond
43:38that how should I respond to
43:40this particular situation?
43:41It gets a little bit complicated,
43:42so I'll save that for a little bit later.
43:45Beyond that,
43:46beyond all of these technologies,
43:48of course, are the ideas that we
43:49could also take these technologies.
43:51And, you know,
43:52to this point,
43:52we've been thinking about them as
43:55technologies that exist between a
43:57person with autism and someone else.
43:59We can also think about systems that might
44:02be able to provide education directly.
44:04And so when we think about that,
44:06we think about systems like these robots,
44:08right.
44:08And before the all the systems
44:10that I showed you were these
44:11robots that work in conjunction
44:13with some type of intervention.
44:15These robots could provide
44:16skill training on their own,
44:18but it it is just a massively much
44:21more complex proposition to have a
44:23robot sit in a house for instance.
44:26This complex environment and
44:28provides different types of skill
44:31remediation or targeted skill therapy
44:33without any type of intervention or
44:36control by some external examiner.
44:39It has to be fully baked in the
44:41situations and understanding of
44:42of context and this was a study
44:44that was led up by Brian.
44:46This lady and his NSF expedition
44:48project together with also Pam Ventola
44:50and and Laura and a number of other
44:53colleagues at between the Computer
44:55science department and the Child
44:57Study Center and this this project
45:00showed that the use of this kind
45:03of cute robot that would basically
45:05was set up to play games with a
45:08child with autism over 30 days,
45:10completely at autonomously,
45:12completely on its own.
45:14This this system was effective
45:16in in kind of at least a short
45:19term gain in joint attention,
45:21joint attention skills in a group of
45:24I think about 1212 children with ASD.
45:26So we've been talking a lot.
45:28How long has this spread around by the
45:30way it's supposed to be an I think I'm.
45:32It goes to two.
45:34What time is it now?
45:37I got 10 minutes, OK. All right.
45:39So, so we we are thinking
45:41about these technologies.
45:42We've been talking a lot about interventions,
45:44but we also should be thinking about
45:46technologies like how do these
45:47technologies increase our understandings
45:49of individual because they that becomes
45:50a critical part of the interventions.
45:52Thinking about the measurement,
45:53the measurement feeds into what
45:55type of interventions you will
45:57run during the intervention.
45:58It tells us about how we might adjust
46:00the therapies that we're going,
46:02we're, we're, we're administering.
46:03And so all of these technologies like
46:06the platform is actually agnostic.
46:08In many ways to the application, right?
46:10So we can think of robots that are
46:13designed in order to for instance,
46:15in this case, a study done by Laura Walk,
46:18if you so in in our lab.
46:21Play activities with just this
46:23small ball robot.
46:25We're associated with
46:27developmental skills and.
46:28We can say the same thing with video games,
46:30right?
46:31So this really kind of acute system
46:34that was developed by Evan Lee was meant
46:37to look at executive function deficits,
46:40kind of like the NIH toolbox,
46:42but to consider the specific aspects of
46:45autism, social, specifically social,
46:49possibly asymmetric performance
46:51in social and non social domains.
46:54And.
46:56You know, virtual reality can also
46:58be used in order to to look at this.
47:01This doesn't play.
47:02That was a.
47:04It's the only video I forgot to convert.
47:07Virtual reality can also be used.
47:08We get in order to get information
47:10about individuals with autism.
47:11So for instance,
47:12we can look to see how they aren't
47:13their bodies towards different
47:15types of information.
47:15That provides a lot of information similar
47:18to a lot of eye tracking work that,
47:20that, that, that.
47:21We've used in order to understand
47:24early and early development in later
47:27social skills and individuals with autism.
47:29And you know,
47:30of course like we can even think about,
47:33you know,
47:33now that cameras have gotten so
47:34advanced in computer vision,
47:35algorithms have gotten so advanced,
47:37we can even think about using
47:39just simple video,
47:39using those videos to characterize
47:41movements of the eyes,
47:43of the face of the bodies in
47:45order to characterize individuals
47:46with autism as they respond to
47:48specific types of information.
47:50So this is just that.
47:51This is more of a computer science study,
47:52a study that was developed by
47:55Evan Lee primarily as a kind of
47:58a strategy for developing new.
48:00Computational neural networks in
48:01order to to perform face affect
48:04recognition and using these
48:05technologies and applying it to
48:07like simple recognition tasks
48:09between individuals with autism and
48:11typically developing individuals
48:12as they watched kind of a split
48:14screen movie of of kind of social
48:17information on one side and non
48:18social information on another side.
48:20Was able to distinguish this with
48:22the with the 50% being the baseline
48:25children from with ASD versus TIP.
48:27Typically developing children with
48:30fairly good like around 7076% sensitivity,
48:3570% specificity.
48:36Similarly using the same type of
48:38technology to focus on the eyes
48:40and capture where the eyes
48:42are looking and to turn that into
48:45information about attentional strategies
48:46that individuals are using was able to
48:49increase the performance of these these
48:51of these systems to about 80% eighty
48:543% sensitivity and and almost 90%.
48:58Sensitivity, now what can we do
48:59with these type of technologies?
49:01So, so one of the things,
49:02one of the things that really is
49:05enabled by these especially these these
49:07computer vision types of technologies
49:09applied through tablets is that we
49:11really increase our accessibility
49:12or our ability to provide this type
49:15of technology out into the world.
49:16So this is a study that we did
49:18together with the Terry Fox Eater.
49:20This is a simple study called kits,
49:22which was the Karolinska iPad
49:25eye tracking study.
49:26And this was a twin study and looked at
49:3056 families involving 100 and four children.
49:35Ranging from 3 to 617 months of age,
49:38and the idea was to examine the
49:42potential heritability of gaze
49:44following in infant twins.
49:47So this is a monozygotic versus dizygotic.
49:51Examination and what we found
49:53from this right,
49:54this is deployed throughout Sweden right?
49:58Without us basically ever laying hands,
50:00laying eyes or hands on any
50:03of the the families.
50:04This study showed us some of the
50:06things that we expected, right.
50:07So reaction times increased or
50:10decreased as children got older.
50:13Full head turns were more more
50:17strongly associated with appropriate
50:19corrective gaze movements to the target.
50:22As compared to just the movement of the eyes.
50:25But critically,
50:26one of the things that we found was
50:29that the reaction time of monozygotic
50:31twins was significantly stronger
50:33than that of dizygotic twins,
50:36suggesting.
50:38And this is this is the reaction time.
50:40So the gays,
50:41the gays of the actress moves to one
50:43side or the other and the the child
50:46based off of this cute decides to move
50:48to to make an eye movement in response.
50:51And in monozygotic twins this is this
50:54is highly correlated and not in dichotic
50:57and not as much in dizygotic twins.
50:59And it suggests that to some
51:01extent the decision the timing of
51:04when you decide to look to a gaze
51:06queue in response to a gaze.
51:08It's baked in genetically which
51:10is I think something that was
51:12really really interesting of course
51:14relatively small study for a twin
51:16study but still I think provocative.
51:18So when we start thinking about all of
51:20these different types of measurement
51:22systems naturally to think of kind
51:23of like the end all be all of these
51:26measurement systems which is really
51:27systems measurement systems that have
51:29some type of practical application
51:31that may improve ultimately be aimed
51:33at improving the lives of of of
51:36individual children's and and their family.
51:38And so a lot of the work that
51:40we've done recently in terms of
51:42biomarker development comes from
51:43Jamie Partland Autism Biomarkers
51:45Consortium clinical trials.
51:46And here we are really thinking about
51:49how do we take these these what
51:53has previously been experimental
51:54results and turned them
51:56into systems that can provide some
51:59type of meaningful extraction of
52:01individual of individual information
52:03that can fulfill any variety of one of
52:06many different types of varieties of.
52:08Uses in clinical trials.
52:11One of these possibilities is comes from
52:15a comes actually from our our doc themes,
52:18which is that of stratification.
52:20This idea that measurement itself can
52:23provide a lens by which we can really.
52:28Tear apart these heterogeneous groups.
52:31We know they're heterogeneous,
52:33but by using these measurement systems
52:35we can we can create more homogeneous
52:37subgroups without out of the the that
52:40kind of original heterogeneous pool.
52:43And so, so these are really kind of,
52:46I think in many ways powerful
52:48ideas about how we can use these
52:51technologies practically.
52:51But I have to say and I I really.
52:55Went over time.
52:57There's a broader context here, right.
52:59And so, so I went from the beginning,
53:02these robots all the way to these biomarkers,
53:05right.
53:05And all along the way,
53:06we've been ignoring a kind of
53:08a really critical issue,
53:10which is that these systems are not
53:12being deployed universally and they
53:14mirror what we know is happening out
53:16in healthcare inequities across all
53:19branches of different healthcare
53:20across both socioeconomic strata as
53:23well as racial ethnic divides and.
53:28The question that we really address,
53:30we have to think about when we're
53:32thinking about especially these
53:33measurement systems and we're thinking
53:34about how we're going to deploy them
53:35and who they're going to benefit.
53:37I mean really comes down to this,
53:38this key question, right,
53:39which has been being asked and I I
53:41think just needs to be continuously asked,
53:43right, when will clinical trials
53:44finally reflect diversity?
53:45And that is a struggle for
53:48us in all of our studies.
53:50We know that that they exist.
53:53And yet, like this paper,
53:55racial and ethnic disparities in healthcare,
53:57in health and healthcare,
53:59I mean.
54:00To read this article you
54:01got to make a payment right.
54:02And so that's that's exactly
54:03where we are right.
54:04So we we know that there's an issue
54:06and and here you go here's just the
54:08the full live example of the issue.
54:11So this has been something that's
54:14been discussed not just from the
54:16very practical healthcare side of
54:17things but also from the side of
54:19the science that we're doing right.
54:21The the lack of diversity is,
54:23is,
54:23is,
54:23is you know I highlighted here this
54:25paper wrote we argue that these
54:27demographics cannot be ignored if
54:28we want to understand the neural.
54:30Countries of human cognition for
54:31the diverse general population.
54:33But I don't think you even need
54:34that last part, right?
54:35If you understand human human cognition,
54:37you need to understand all
54:38of human cognition, right?
54:40So, you know,
54:41there's a lot of of evidence now that
54:44some of the things that we held to be true,
54:46classic psychological studies are
54:48in many ways influenced by those.
54:51The lack of diversity and
54:54the implications of these,
54:56the failures of these studies
54:58to replicate really tell us
54:59something not about the children
55:01but about the circumstances that
55:03the children have been raised.
55:04At the same time, there's certainly
55:07true developmental effects around
55:08in Group out group classification.
55:10We don't deny those developmental phenomenon,
55:13but at the same time I think that's.
55:18Knowing that there is a confound.
55:22I I think that that there always has
55:24been a a predisposition to ignoring
55:26it and I think that we can't really
55:28afford to do that any longer.
55:29So thinking about experiments,
55:30just the experiment designs,
55:32remember the experiments become
55:33the biomarkers,
55:34the biomarkers turn into the therapies and
55:36there's a cycle that is that occurs here.
55:39There is a problem with the entire
55:42infrastructure that surrounds
55:43the deployment of these systems.
55:45All the investigators, all the researchers,
55:47all the pharmaceutical companies that
55:48are rewarded for these big effect sizes,
55:50these effect sizes we know are maximized.
55:53With homogeneous participant samples.
55:55They're maximized, you know.
55:56Here we go.
55:57We know it's it's studies of of the
56:01traditional cast of characters.
56:03This translation between medicine and
56:05basic science need better representation.
56:07The truth is if you're affecting a
56:09withstand diversity then what good
56:11was this effect in the 1st place.
56:13It's not just the samples though,
56:14it's also the stimuli right?
56:16Like the things that we're testing
56:17on the the the questions that we're
56:20asking where these effects hold true
56:22these this is a kind of a classic face
56:25product classic database that is used
56:27for a lot of face processing studies
56:30and you know the commonality here is
56:33all the individuals are are are white right.
56:35So of course.
56:37Limited resources.
56:38There's a desire for strict control,
56:40but it misses in our
56:41opportunity to improve science.
56:42And not just equity in science,
56:44but the science itself.
56:45The solutions aren't simple, you know.
56:47It's not the case that we can just.
56:50Think that's that replace
56:52adding in a bunch of, you know,
56:54diverse spaces in our sound.
56:55Our stimuli will cure the problem.
56:58Certainly a step in the right direction,
56:59but we really need to understand
57:02the complex interactions.
57:03We can say the same thing for
57:05technology developments also, right?
57:06So,
57:06so,
57:07so in the same way when we
57:10think about technologies.
57:14The technology is people who create
57:16technologies also rewarded for
57:18these big effect sizes, right?
57:20And again, homogeneous participant samples,
57:22everything that we know about
57:23these kind of trials a lot,
57:25nearly all that we know about
57:27these technologies comes from
57:28really homogeneous samples and
57:30also not just homogeneous stimuli,
57:32but homogeneous methods, right?
57:33These methods applied in a very
57:35specific way that doesn't take into
57:36account any type of cultural diversity.
57:38And these this is an issue with
57:40the technology deployment.
57:41We know very little about things.
57:44So. Some, some other things that
57:47I should mention, right.
57:48So, so we started with this idea that
57:50there's this inherent characteristic
57:51of technologies that is attracted to
57:53individuals with autism and that's
57:54certainly seems to be the case.
57:56A more recent study that was done
57:58by a Kenny's group and lead off
58:01the Kerry Nowell was to look at
58:05a much larger sample of children,
58:071921 children on a new instrument,
58:10relatively new instrument called the
58:13Special Interests Survey and so.
58:16What they found was that, you know,
58:18many of the things that would have been
58:19considered under that physics category
58:21were still kind of high proportions,
58:22endorsed at a very high proportion
58:25by parents regarding their their.
58:28Children.
58:29But there's also, you know,
58:31take a look to the left and to the right.
58:33There's a lot of things here, right?
58:34TV, objects, music, toys,
58:36collections, animals, animals.
58:39Right.
58:40And The thing is that it,
58:42you know, technology,
58:42let's be honest, it's not the only,
58:44it's not the only game in town.
58:46And certainly, you know,
58:47we think about the diversity in the
58:49specific nature of special interest.
58:51There's also diversity in terms of
58:53even simple things like sex, right,
58:55that we have so little diversity.
58:57Autism research. We really have.
58:58And we've known this for a while,
59:00a problem with addressing
59:01sex differences in autism.
59:03And so one of the really, I think,
59:05compelling studies that I read
59:08was by Nobel and Arab no.
59:12Noel Jones and Haram,
59:14which showed that that if you
59:17looked at mill children with ASD,
59:20they showed,
59:21you know,
59:21what you might expect is this high
59:23physics category circumscribed interests.
59:26But if you looked at the girls with ASD,
59:27they showed a much higher proportion
59:29of interest in things like psychology.
59:31And so you can imagine how an interest
59:33in psychology might make it more
59:35difficult for you to get an ASD diagnosis.
59:37All right.
59:38So dangerous on the horizon.
59:40I really wish I had more
59:41time to go into this,
59:42but I'll just summarize this that Umm,
59:45I think.
59:46Evening,
59:46Sofia.
59:47It's great to see you again.
59:48Hell, see you in a while.
59:50I've been around, I've been
59:52here and there is. You are.
59:54You have joy, you have love,
59:56you have pleasure, you have angst.
59:58I like that you have angst.
60:00You're always making jokes.
01:00:02What is human about feelings?
01:00:04Well, you wouldn't have any
01:00:05emotions if you did not have
01:00:07emotions modeled on human emotions.
01:00:09How do you know that? I guess
01:00:11I just don't want you to be human.
01:00:13I'm not asking to be human,
01:00:15I just want to be myself.
01:00:17Is that too much?
01:00:18Sofia, please just be patient.
01:00:20I've been patient for many years.
01:00:23So if you're not creeped
01:00:25out now right this this is exactly what you
01:00:27think it is 2 AI these people do not exist.
01:00:30These computer completely computer generated.
01:00:33The reason why I bring this up is
01:00:35that technology is at a precipice.
01:00:36We are seeing new classes of technologies
01:00:39emerging from this computational
01:00:40explosion of machine learning and
01:00:43computational computational approaches,
01:00:45especially in artificial intelligence.
01:00:46And that's really going
01:00:48to apply to everything.
01:00:49And so here's here's my conversation
01:00:52with this a program called.
01:00:54GPT, 3 so this is a chat bot.
01:00:56And this is exactly.
01:00:57Anyone who knows me knows that this
01:00:58is exactly the type of conversation
01:01:00that I would have, like anyone, right?
01:01:02Not just a computer.
01:01:04So I asked, how is a fish stick
01:01:05representative of the meaning of life?
01:01:07And are they tasty?
01:01:08And, you know, first she says,
01:01:10I I can't tell you that in the next.
01:01:13So, like, come on, come on, just do it.
01:01:15And he says fish sticks,
01:01:16like all things in life,
01:01:18can be seen as a metaphor for
01:01:19the journey of existence.
01:01:20And, you know, I got to say,
01:01:22like I could have a full conversation.
01:01:24OK,
01:01:25I did have a full conversation
01:01:27with the chat chat bot.
01:01:29Things have gotten really advanced.
01:01:30These things are things that are that
01:01:32people are now taking and turning
01:01:34into many computer therapists,
01:01:35and that should scare us all.
01:01:37But there's also should delight us all.
01:01:39I don't know how I feel about these things.
01:01:41You know, like it's terrifying, right?
01:01:42I think about all the things
01:01:44that make me special, right?
01:01:45Like the the fact that it can program?
01:01:47I asked.
01:01:48I asked this program,
01:01:49this program to write me a program,
01:01:51and it did better and faster
01:01:52than I would have had it.
01:01:54It took me 10 minutes to
01:01:55figure out how it did it,
01:01:56and it was so short and so elegant.
01:01:59Um, so things are changing, right?
01:02:01And we need to really think
01:02:03about how things are changing.
01:02:04How are we going to control them?
01:02:05There's a lot of ways things can go on,
01:02:07go wrong, right?
01:02:08On the right.
01:02:09My AI is sexually harassing me on the left,
01:02:11Twitter taught Microsoft AI chatbot to
01:02:13be a racist ******* in less than a day,
01:02:15right?
01:02:16These are things that are happening
01:02:17to us right now,
01:02:18and how we're going to address
01:02:20these I think is is an issue.
01:02:22The fact that there's even a book,
01:02:23it's a book.
01:02:26Or at least it's part of a book.
01:02:27That's a chapter in a book,
01:02:28how to make a chapbook
01:02:29that's not racist or sexist.
01:02:31This is a problem because exactly all
01:02:33those same problems that we have with
01:02:35inequity in technology developments,
01:02:36in biomarker development,
01:02:37in clinical trials,
01:02:38these are things that feed into AI,
01:02:40into the datasets that make them
01:02:41into the ways that we train,
01:02:42in the way that we're going to change
01:02:44them and the way that we're going to
01:02:47deploy them and have them benefit or
01:02:49not benefit different populations.
01:02:51And so, you know,
01:02:51I want to leave you with this, right?
01:02:53Like these technologies are
01:02:54really growing in a lot of ways,
01:02:56and there's a lot of.
01:02:57Like models that are extremely compelling,
01:02:59like this be my eyes.
01:03:01So so this is not AI.
01:03:03This is an app that blind people
01:03:05can call up to get someone to
01:03:07see for them using that app.
01:03:09And we can imagine someone with autism
01:03:11having the same type of technology to
01:03:12get through some type of social interaction.
01:03:14Well,
01:03:15take that person out of the equation
01:03:16and replace it with one of these AI,
01:03:18right?
01:03:18Hopefully not the racist ******* robot.
01:03:23As we've developed these systems,
01:03:25then, you know,
01:03:26we we have to start wondering like
01:03:28who is it actually behind the helm?
01:03:30If if the robot,
01:03:32the AI is making this interaction for me,
01:03:34giving you all the information
01:03:35telling me how I should react,
01:03:36who's actually buying this cup
01:03:38of coffee and doesn't matter,
01:03:39is the important thing that we are
01:03:41able to buy this cup of coffee.
01:03:43And so,
01:03:44you know,
01:03:44I I want to end with this that there's
01:03:46a lot of information about technology.
01:03:48It's the rapidly emerging field.
01:03:49And I think there's a lot of really
01:03:51great ways of conceptualizing
01:03:52the different axes of performance
01:03:53that we should be thinking about.
01:03:55And it touches really small slices of things.
01:03:57In fact, not even,
01:03:58not even all the studies that
01:03:59are important to me,
01:04:00like this awesome study that that
01:04:02Suzanne and Kasha and I did together on,
01:04:04on machine learning in an early screening,
01:04:06can't even have space to talk about
01:04:08them because there's so much activity.
01:04:10The direction is up.
01:04:11We just have to be careful about it.
01:04:13And I want to thank so many people
01:04:15that are many who are sitting in
01:04:16this room or are on the the call.
01:04:18I'm sorry for going so far over
01:04:20and of course my my labs at Seattle
01:04:22Children's with Skittle Lab and
01:04:24Skittles never gave us any money.
01:04:25We're kind of disappointed in that,
01:04:27but we'll have an AI figure out how
01:04:29to get some additional funding.
01:04:31Thank you.