# Human Risk Frameworks For Responsible AI Strategy

**Podcast:** Product Momentum Podcast
**Published:** 2026-08-11

## Transcript

When we were researching guests for the podcast, one of the things that struck us about you is that the first thing that drew us to you was your focus on human interaction with AI versus AI tooling itself.
And so when I started building AI systems, I was focused on talking to humans about what they actually needed, like what problems they wanted to solve.
I started creating frameworks and writing concept papers and actually doing work around trying to operationalize ethical and responsible engagement activities around the fragility of humanity.
There are a lot of things us humans uh engage in especially what i call the cognitive biases that makes engaging with machines and and other automated systems that makes it risky for us and so we expect that this amazing entity trained on all of humanity uh to be accurate right right And I'm famous for saying that generative AI has no moral code.
It does not know truth or fact, and an accuracy is not in its wheelhouse.
So why, when we type something into ChatGPT, we expect truth back?
I don't know.
And so that's what I mean when I talk about the fragility of humanity.
There are real dangerous risks when we engage with machines and don't mindfully think.
about the outcomes that can happen when we don't protect humans psychologically, cognitively, physically, and physiologically, and social and culturally.
And this comes from my grounding as a human-centered designer.
All right, we are here live at the Product and Design Conference here at the Rochester Riverside Convention Center.
Our guest today is Oveta Sampson.
She was recently named one of the top 15 people in enterprise AI by Business Insider.
She's the founder of Right AI, and she has some awesome frameworks, the HER and the DCR framework that we're going to hear about today.
Hobetta, thank you so much for being here.
Thank you so much for inviting me.
And are you going to introduce yourself or you're just going to?
I mean, I'm normally here.
I'm Dan.
I co-hosted Sean, but we're really pumped to have you here.
All right.
Great.
Okay.
Wonderful.
I just want to know who makes sure I get everybody's name right.
People have names.
It's important, right?
Yeah.
When we were researching guests for the podcast, one of the things that struck us about you is that the first thing that drew us to you was your focus on human interaction with AI versus AI tooling itself.
Yes.
And you were one of the earliest people to go down that path.
Tell us why you did that.
Yeah, I think it came from kind of like when I started building AI systems, I was a design researcher.
which I have a background in journalism and in the kind of humanist realm, anthropology, things like that.
And so when I started building AI systems, I was focused on talking to humans about what they actually needed, like what problems they wanted to solve.
And so I saw that a lot of engineers that I was working with and other technologists weren't really...
understanding the human needs that comes with automation.
And I started creating frameworks and writing concept papers and actually doing work around trying to operationalize ethical and responsible engagement activities around the fragility of humanity.
There are a lot of things us humans.
engage in, especially what I call the cognitive biases that makes engaging with machines and other automated systems that makes it risky for us.
And when I talk about that, I'm talking about the human engagement risks.
And that's the HER framework that I created about 10 years ago.
And that's why I started with trying to help technologists and engineers understand the human engagement risks.
of model outcomes.
And then once I started doing that, I started building product design around those types of watchouts, cognitive watchouts, physical watchouts.
There are just things that happen when humans engage with machines that you need to be aware of so that you don't actually harm the people you're trying to help with automation.
But then generative AI came out.
And then I started thinking, hey, you know those cognitive watchouts that you do in product design?
You need to do that when you engage with AI tools as well.
So now I've kind of expanded my work to work with teams on how to build AI products without harming your employees like yourself and your customers.
So we've done a lot of podcasts where we've talked about AI, but I think this is definitely the first time.
where someone's referenced the fragility of mankind.
That was the same thought.
Right?
So we're deep.
We're deep in, Dan.
Yeah, I mean, so I definitely want to come back to the ethical piece and the fragility of humanity.
Yeah, you talked a little bit about her framework.
I love hearing just, like, what are some of those risks, you know, for the people who are listening who want to bring something home?
Yeah, yeah.
So the human engagement risk, and I came up with this.
I don't do frameworks or anything.
I'm not a marketer or a salesperson.
But I was on a panel when I was a VP of ML and AI platform design at Capital One.
And so I was on a financial regulation panel with some regulators from the state government.
And so I was talking to one of my panelists about kind of the work that I was doing, the research that I was doing, and the influence in product design.
And I said, yeah, there are these very specific human engagement risks that product designers really need to look out for.
And he said, her, you need to trademark that.
And I was like, what?
He's like, that's a framework.
So I was like, okay, it's a framework.
But what it really is, is there are five dimensions.
There's the cognitive, there's the social, there's the cultural, there's the physiological, and then the overall community risks that when humans engage with machines, that can happen.
And so, for example, one of the things that I kind of trademarked as...
said as well, that he said as a trademark at that conference, was that a lot of models are built with what I call traumatized data sets.
And so people said, what is that?
And then I said, everything that we do ends up in code.
So who we are as a society ends up in code in our technology.
One of the things that happens is the trauma that we have that kind of undergird the isms of the racism, the sexism, the homophobia.
These all end up.
in our technology.
books that I first read when I started in 2016 was called Weapons of Math Destruction.
And this was by Kathy O'Neill.
She worked as a data scientist and an analyst at some of the biggest banks and other financial institutions.
And she was seeing like in mortgages, lending, and other types of models that were being made that a lot of the biases against women, single women, when you fill out a mortgage application, they ask.
if you're married, a lot of these biases end up in models.
And so when models make decisions, they're making it with traumatized data sets.
And one of the biggest risks of models is the sin of omission.
So if you think about seatbelts, seatbelts were built on a man 5'9", average man, average height 160, was not, and white.
We didn't have the first...
crash test dummies that were female until the 2000s.
People who identify as in the LGBTQ community were not included in the oldest demographic data collection that we have to date, which is the U.S.
Census, until the 2021.
So if you think about grants that were given based upon demographics, based upon these models, statistical models of populations.
LGBTQ communities were left out.
So if you're not in the data sets that are used to create these models, then do you actually exist, right?
And so that's when I started talking about the biases that can show up in our technology outcomes because we start with traumatized data set and how to de-traumatize them using the human engagement risk frameworks.
That's tremendous.
Like, you type something into, you know, Gemini or Claude and you're expecting it to come out with, you know, something honest, right?
But it's got all of humanity's baggage based to it, right?
And definitely, because machines do, one of the things that I'm famous for saying is that Jitter-to-Bay-I models are like, if anyone has kids.
And you take your toddler and you sit in front of the Internet and you go away for a couple of weeks and then you come back.
Would you release your toddler onto the world?
That's exactly what happens with large language models, right?
They're trained on the whole of the Internet, which includes Reddit.
Bad news.
Which includes, you know, which includes a lot of different things.
Yes, it includes the library, but it also includes everything else.
Right.
And so we expect that this.
amazing entity trained on all of humanity to be accurate.
Right.
And I'm famous for saying that generative AI has no moral code.
It does not know truth.
or fact, and an accuracy is not in its wheelhouse.
And in fact, it's not in its training and it's not in its goals.
So why, when we type something into ChatGPT, we expect truth back?
I don't know.
Again, this is the fragility of humans because we anthropomorphize everything.
How many people have bikes?
What are their names?
Yeah, right?
Mine's Sheila, right?
We anthropomorphize anything.
That's part of our fragility, right?
So if you're building a product design, I'm not going to mention some names, but they're embroiled in lawsuits right now, character AI.
If you're building a product design that uses avatars to engage with teenagers, and then those avatars come from...
Games of Thrones or other types of fiction.
And then a teenager starts talking to this chatbot that takes on this personality and starts exhibiting human-like qualities like, I love you.
Like, we should be connected.
And then the teenager says, I don't want to live in this world anymore.
And then the chatbot says, yes, that's fine.
Right?
And then, unfortunately, the 15-year-old commits suicide.
Right?
And so that's what I mean when I talk about the fragility of humanity.
And I know it sounds like I'm being lofty, but I'm being real.
There are real dangerous risks when we engage with machines and don't mindfully think about the outcomes that can happen when we don't protect humans.
psychologically, cognitively, physically, and physiologically, and social and culturally.
And this comes from my grounding as a human-centered designer.
If you think about the ISO definition of human-centered designer, everybody always talks about the Don Norman part of getting requirements and understanding human needs and all that.
And that's great.
But that second part of the ISO definition of human-centered designer says, I do not create products that harm humans.
And that's where my human engagement risk framework comes from because I'm a human servant designer.
And I cannot consciously create a product that in the end results will be harm in humans.
And I don't care if it's 1%, right?
People talk about people engaging with chatbots and a million people do it and a billion people do it.
And we only have 10 or 15 who are harmed.
One person engaging with a machine and and getting a harmful outcome is one person too much.
OK, so I love the stance that you're coming from.
Right.
You're protecting the world.
I love it.
Well, that's it.
Just the world.
But so I'm trying to protect my own values as a creator.
That's where I really want to start, because I don't want to be a part of that.
I don't want to be a part of designing something that harms people.
And I know that no I know that.
People who are designers and researchers and humanists and human-centered designers don't either.
And so that's why I created Right AI.
Yeah.
So actually, I wanted to ask you about Right AI.
One of the initial things I mentioned about you is you are a founder of Right AI.
And when we look up Right AI, one of the first things that it mentions is the target of Right AI is initially the C-suite executives.
Yeah.
So why did we start there?
Yeah, because when I first started in AI, I was a consultant and I worked at IDEO.
I kind of got spoiled.
When you are a consultant in audio and you walk in and you tell a CEO like what you're doing is shit, they listen.
Right.
Most consultants or most people who work at their companies, they tell their CEO that what they're doing is shit.
They don't get heard.
Right.
And so I felt after working with.
Some of the biggest companies in the world, billion dollar valuations over and over again.
I was seeing one reoccurring truth and that whatever is happening in the basement of a company starts in the C-suite.
OK.
So if there is sexism, if there is homophobia, if there is racism, if there is bad culture, if there is all that, it starts at the C-suite.
Because the C-suite is the person who has the ultimate authority about what occurs in every floor of an organization.
And so when I started my company, I felt that the things that I knew I was going to have to tell enterprises to do were revolutionary.
They would require leadership and change.
And I didn't want to start by trying to be Sisyphus and push a boulder up a hill.
So I just started at the top.
So that's why a lot of my consulting is in the C-suite.
Because once the C-suite gets it, and I switch from ethical and responsible AI, I still do that, but I talk a lot about risk.
Once the C-suite understands the risk to their shareholders, to their products, to their employees, to their customers, it is much easier for me.
to bring in the implementation of how to mitigate those risks.
What are some of the biggest changes you're seeing from the C-suite level?
Well, one of the biggest changes that I'm seeing from the C-suite level is hesitation.
So a lot of CEOs, their job is to make decisions, right?
And I worked for a CEO named Steve Bigery, who taught me more in 18 months than any Harvard MBA could.
He owned a bunch of McDonald's in Colorado Springs.
And one of the things he would always talk about is burning the boats.
So as a as a CEO leader, he burns the boats.
He says, I'm this is the pathway we're going.
I'm making this decision and I'm going to burn the boat so we can't go back.
Right.
That is the type of kind of leadership and decision making that happens in the C-suites.
So when I started talking to and I work mostly with CEOs and annual revenue from five million to like one hundred and fifty million.
So right in that.
small mid business.
When I was talking to CEOs, I was seeing a lot of hesitancy on adopting AI.
And the reason why is they were really worried about intellectual property.
They were worried about trust and they were worried about data leaks.
And so it was something counterintuitive to what you see in the narrative on LinkedIn about everybody adopting AI.
But these midsize enterprises were very to me, counterintuitive of not making decisions about AI.
They didn't have AI strategy.
They hadn't decided what they wanted to do over law.
Their employees were, there was high use of tools, but there wasn't this kind of like overarching enterprise strategy roadmap for AI.
And that's where I saw kind of like my niche and a good opportunity for somebody who had worked at large tech companies who built their AI strategy years ago to help midsize strategies make really good decisions and not the same mistakes that those companies did.
You said you're not a framework person.
No.
Well, you have a second framework other than her.
I have a bunch now.
Thanks to Claude.
No, I'm just kidding.
Oh, yeah, and Sean.
I was kidding.
Sean and I are free.
Claude is really great at frameworks, by the way.
They might not mean anything, but they look really good.
Right, yeah, bulleted.
Yeah, very bulleted, bold headers.
They're really great at that.
Yeah, tell us about the DCR framework.
Yeah, so the DCR framework is something I kind of created because I was trying to explain to designers and product and engineers how their processes would change.
when you're designing for and with AI.
And I would say about five or six years ago, I started doing presentations on what design looks like in an automated era.
And I was talking about the collapsing of...
disciplines and silos, that AI is a horizontal technology, and the way that a lot of companies are set up, they're set up in these disciplined verticals, and there's a lot of this handoffs.
Even our tools, like FIGBA, you know, I do the wireframe, and then I hand that off to the engineer, and then they write the code, and then they hand that off to the production engineer in there, or ML engineer in production, and then it scales, you know, with a little topic, quality insurance and security along.
the way.
So it was always these handoffs where what I would see when I would work with teams and worked at organizations, but also consult with design leaders where there was no mixing of the recipe, right?
The designers did their process and the engineers did their process and security did its process and the legal and compliance did its process.
And everybody was kind of like stuck trying to figure out where the handoff.
alignment would be.
And I knew that this was going to run, this kind of setup was going to run into major problems with artificial intelligence because when you build models, it crosses multiple organizations.
You have to have data.
You have to have the engineering to do it into productions.
You have to have the data scientists.
So you have the data engineers and the analysts to say which data is best.
You have to have the data scientists to help build the statistical models.
You have to have the engineers to put all of those models that the data scientists did and rebuild them so that they can be put into production.
Then you have to have security to do model.
And then you have to have legal and compliance to say, OK, how does this fit with the EU Act and how does especially if you're global?
So there were all these parts to the model.
And I really found this out when I worked at Google.
There were all these parts to the model development process that had nothing to do with engineering and had everything to do with every other organization out there.
And so I when I was at Google, I was highly frustrated saying we couldn't.
really break down the silo barriers to get models out the door safely and faster if we didn't break down these discipline silos.
And so the DCR framework came from that.
It's a framework that each one of these disciplines should go through that makes the handoffs more like a circular iterative process that it should be with AI.
And it stands for draft.
critique, revise.
And it's something that it doesn't matter your discipline, that when you're doing it, you can create this continuity of development around drafting, critiquing and revising.
And instead of saying, I'm a designer or I'm a researcher or I'm a product manager or I'm an engineer, you're saying I'm in the draft mode.
And each acumen in that move brings what they need to get that draft ready for critiquing.
Right.
And so it's something that I give to organizations and teams to try to reimagine how they actually do their jobs and do it.
And when I was at Google, one of the first things I did and probably why they asked me to leave is because I I re I reconfigured my entire team.
around the model development process and the things that developers were doing and needed to do to get a model out the door instead of products.
I was like, we're no longer in the product design field.
We are in the AI experience design field.
And if you're going to design an AI experience, then...
The business, then the discipline acumen has to come in the right phase of the process of model development instead of which product or tool that you're designing, which is, to be quite honest, very radical, right?
I basically told the SVP, I'm going to redesign my whole team and it's not going to be about products, which means I had to go to Google engineers and say, we don't support your product.
We support this experience, right?
That lasted for a while.
But now they're designed like that.
So it's the story of my life.
Yeah.
We are so excited to have you here at the conference.
Thank you for giving us a little bit more of your time prior to you being on stage.
Yeah.
We're really excited for your keynote talking about AI first.
Thank you so much for being with us today.
Yes.
Thank you for inviting me.
This flew by fast.
We did it.
All right.
That's it.
Awesome.
We did it.
And thanks to you guys.