And my title is How Might We Teach AI What Goes On Set?

Introduction: Framing the “How might we” Question

And does any of you from design background?

From design?

So maybe not.

But so how might we question is actually often used in design field for brainstorming.

So my intention here is that this talk is not to provide answer,

answer, but to think about this question together.

And maybe Toyashi would speak a little more of the what could be solution side or the

maybe not.

I'm going to just leave it to Toyashi after my talk.

So that's my intention.

And of course, my slide is made by AI Cloud.

Who I Am and Why This Matters

And OK, so who am I?

Yu Sugiyama is my name.

and I've been in design field for the past 12 years and after I took masters of

system design and management I combined system thinking and design thinking to

tackle complex problems such as started user research for genetic

testing service which is very data sensitive for human so human sensitive

human data sensitive involvement and after that moved to design firm and

engaged in autonomous driving self -driving car project which also

involves data and recently I worked for Toyota's connected city project which is

a living lab, but also the testbed for well -being for all and new products.

And currently my target or the interest lies in digital legacy and digital identity.

And I'm gradually launching my research in that field.

but in the meantime I worked for an AI company and created AI rules for some period of time

and I'm not in a position to talk about any detail of that

but I learned about AI creation and what it would take to create ethical AI

So that is my interest and today I want to talk about what subjectively I think about it.

A Small Experiment: Context Changes Meaning

Okay, so let me start from a small experiment.

Reading the Same Words Differently Across Places

This is a question of your kids play piano very well.

What if your neighbor comes up to you with this question?

and since I lived only in Japan for and in the Netherlands for my life I let's okay let me start

from Amsterdam though stereotypical I would take it very direct okay this guy thinks that my kids

play piano very well, right?

Okay, I lived in Tokyo for a long, long time.

What I would

think if someone from my neighborhood was in Tokyo, I would just take it, okay, compliment,

I don't know what they are talking about.

What about Osaka?

Osaka, people in Osaka are

always joking, so they would probably think, okay, when they get famous, you'd better treat

me to dinner.

What about Kyoto?

I lived in Kyoto for six years when I was in

university.

Can you guess?

I would be a little bit tensed up if my neighbors

tell me in Kyoto because it's very very likely that they just say it's

too loud, do something about it.

Otherwise there is going to be a poster in the

the apartment that tells you to stop it, right.

So this is a collective knowledge.

This, so like Kyoto example, there are literatures that Kyoto's people are

sophisticated, have sophisticated way of talking or communication, but so these

There are data, not just on the internet, literature, or X,

or whatever.

So you can teach these things to AI.

But, OK, so another example.

Language Variations and Data Dominance

OK, so I'm studying Dutch, of course,

since I'm living in the Netherlands.

Lopen, lopen.

In the Netherlands, this means to walk.

In Belgium or Flanders, it means to run, right?

So this is different, but how do you teach this difference to AI?

It is very likely that since there are more people in the Netherlands and less people

who is speaking Belgian way of Ropen, it is likely that the AI would more take Ropen as to work

because there are more data in the public.

And if you want AI to output Ropen as to run, you need to give some sign in the prompt

or the okay I'm living in the Belgium or you share your GPS data to AI right so

what I want to say is AI is not neutral and so whose data dominates really

reflects the world of AI answer this can also be taught to AI because it's the

there are different it's obvious that there are difference but what if we go

go to more local, say hyper -local, say your family or your neighbor's communication.

Okay, let's go this way.

Hyper-Local Meaning in Everyday Relationships

How about pasta tonight, you ask your partner.

What does pasta say?

Maybe they say, that's good.

But what does that mean?

Day one, say of your marriage,

Yes, I'd like pasta, of course.

But 10 years in, this could mean a lot of things.

Of course it can mean, yes, I'd like pasta.

But it could mean that after 10 years, you know that your partner loves carbonara, and

that's good means carbonara, and your partner will get angry if you make tomato pasta, right?

or that good could mean I prefer pizza but I don't say anything about it

or that good is a sign of I don't want to talk to talk with you today

so these things are very very local right only you and your partner knows this conversation

conversation, and yes, you can teach these things to AI if you input lots of your conversation

to AI, but you don't do it, basically.

You don't want to do it.

I mean, I don't personally

want to do it.

What AI Can Learn Depends on What We Input

So, AI conversation involves input data, and only things you input to AI

AI can AI output, right?

And how that is functioning, so how AI company builds AI is actually human, right?

So behind every AI, there are thousands of humans who create AI.

There are engineers who create models and there are people who actually prepare such

data.

Okay, this is a good data.

this is if we put too much of open from the Netherlands background it will only

this AI will only output AI as walk and they people from Belgium would get

angry that's not good so we should maybe add a little bit more of data from

Belgium but can we hire should we hire Belgian people more or so those are the

the realistic problems or the problem they are fighting.

So someone can put Kyoto and Namsu as the customers

to the AI -trained data, but it's really hard for AI

to grasp a really individual context of your life

unless you teach them individually.

A Framework for What AI Can and Can’t Know

So what can AI learn and cannot learn?

So I made, I mean, I just borrowed a framework from Donald Rumsfeld, who used to be a secretary of the Defense Department of the United States.

So there are four kinds of knowledge out there in the world, he said.

So the left side is whether the question is no or not, and the upper part is the answer is no, no, not.

and no known is what we know we know so that is fact or knowledge right so that

there are of course factor knowledge there are I mean most of the most of

many data that are facts and knowledge right so it's very easy to input to AI

and no unknown question or no unknown field is something you don't know okay

OK, what would happen with agentic AI in coming 10 years?

That is no unknown question.

People are talking about it, but we don't know the answer yet.

So those are the knowledge and information

that is easy to discuss on input AI.

But the bottom part, it will be a little bit tricky.

tricky.

The bottom left part, unknown, is something we know but haven't put into words.

This is something, for example, intuition.

If you worked for a factory for 30 years and

you step into the factory and you have a lot of knowledge and you somehow sense a subtle

to sign that something is wrong today.

That is probably not verbalized.

That is no unknown

field.

And there are, OK, maybe someone verbalized all of that and input AI.

That can be AI knowledge,

so it goes to the upper side.

But there are what Gozan said from the title from today

today is actually, I think, in this field.

And there are also unknown, unknown questions.

This is the question we don't know yet.

So for example, maybe 100 years later, there

are questions that human beings are tackling.

And that question, what would the question

would be, is totally unknown, unknown question.

And today I want to talk, go a little bit

“Unknown Known”: Tacit Knowledge, Place, and Work

more deep dive into this unknown known field and I think there are three elements in this area

and one is place right so really hyperlocal question or really individual questions are

it's it's uh there are things you don't know that you know you intuitively unconsciously do

do and there are still things that you cannot verbalize and AI knows, they don't know them.

But it's also in the workplace, there are so many tacit understanding people do intuitively

or like operational excellence that happens in the company that people don't verbalize

it but maybe with AI it solves a lot of problems or it can be automatized so

Time: Living With AI Over Years, Not Moments

there are a lot of opportunities here and time and will I'm gonna go a little

bit more deep dive and let's talk about time so this is a kind of talk that I

mean encouraging to let's stop a little bit instead of you know we are using AI

crazily these days so what if we stop and think about the meaning of using AI

this was a question thought I came up when I was working with the AI company

so think about the kindest person in your life it can be your sports coach or

guardian or parents and I know there are cultural differences but for me they're

sometimes not so nice right they can be really silent for years or they can

easily get angry and being nice and friendly and being really kind is

sometimes different because parents are really strict sometimes and you need to

endure and live with those kind person for years and only after years

you notice that okay that harshness was love and that meant something for me but

But can AI do that?

If AI becomes silent for, OK, my cloud

becomes silent for three years, I

would just switch to Gemini, of course.

I wouldn't wait for three years.

Probably day two, I would switch, right?

Will: Value Judgments and Responsibility

OK, another thing is will.

Should your parent keep their driver's license?

OK, this is a big problem in Japan.

pun, OK, after year 75, the probability of driving accidents

increase after year 75.

And AI can say that, right?

Statistically, your accident risk arises after 75.

So should you return your driverless license?

But this is a trade -off question.

Okay, should I advise my parents this or not involves really subtle decision -making because

it would hurt their dignity, or they really believe they can drive yet, and they think,

can I take driver license and opportunity to visit the graveyard of

the grandparents right so AI can answer with the optimal but

decision -making involves really value judgment here and that is not I think

this is may not be the field we can totally leave to AI so what I feel is

is that AI can behave nice, but AI cannot be.

So be here involves time.

Know how to be silent is something AI do not know yet,

because it's a commercial product, right?

And AI cannot take responsibility for,

and because they don't have the will to spend with your whole life yet.

Maybe it will change, but I think at least this is how I see AI product.

Toward AI That Walks Beside Us Across a Lifetime

Okay, so my question goes, how might we teach AI human life?

Okay, so we are not just looking for some quick answer,

and how should we make it really human -centered and really how could we make AI really for us, our life?

So that is actually the question I'm tackling right now.

And AI that walks beside us for whole life, maybe from youth to adults and passing on the AI even to the next generation,

generation, maybe people 200 years later, they look at our AI data and say, okay, this

guy 200 years later at the beginning of AI emergence was thinking and talking about such

and such things.

That might happen.

But, at youth, maybe we shouldn't encourage people to use so much AI for quick answers,

and maybe we should encourage them to explore the world instead of asking quick answers.

So it depends on the moment, and as a UX specialist, I feel that different goals should be set across a lifetime, even if it's the same AI.

And so this is a personal interest.

Digital Legacy: Inheriting Data, Identity, and Care

My research involves this inheritance history.

So, what would it take for us to inherit our data if we live a whole life with AI and pass

it on to the next generation?

It would involve, so this is a little bit detailed, but I think it will involve three

different areas.

Data Accumulation and Ownership

One is the data accumulation.

So AI, behind AI is your data.

so what data should you accumulate and how this is tricky because companies

have different so many service providers have so many different data of yours so

it's a really scattered around and who in research that your data is this your

family or should some part of your recording should be maybe kept more

public, or how can you show your will not to inherit to the next generation, and what

the ethics involves.

Not just GDPR, but also there are regulations that involve this discussion.

After Death: Handovers, Notifications, and Coordination

And so this is an e -Deaf experience, a little even more detailed, but currently I'm thinking

about, okay, what would happen after death and what does it take to accumulate a person's

data and what does it take to hand over to another company and this actually involves

multiple, if we really want to design this, this involves so many companies and even governments

because like the central part will be like government parts like DVD or digital wallets to identify this person's debt

and companies would maybe in the future companies get a notification from government that certain person is dead

But right now, bereaved families need to contact each company to get back the data or to, say, your bank account of a deceased person.

It would involve a lot of work.

So I'm believing that this might open up a new way of caring to the bereaved family.

This also shows that it involves a lot of fields, not just tech and policy, but existing

ritual and UX.

This is my last slide.

Conclusion: The Meaning of Time Spent (and Lost) With AI

For me, when I was a kid, I read a book by Antoine de Saint -Exupéry, Le Petit Prince.

In that book, it said it is the time you have spent on your roads that makes your road so important.

Actually, it's the wrong translation.

Oh, that's good to know.

The time you have lost on your roads.

So, even that one is what's making it important.

Ok, but thanks for letting me know.

That's even better.

So that means that the lost time actually matters,

and the time spent really does matter.

But we only spend, say, two or three years with AI

for most many people, like a whole month or a few years,

months to a few years maybe.

And we will live with AI for the coming decades,

case, and we don't know what it would mean yet.

So let's explore the AI time together.

OK, thank you.