Responsible AI belongs next to HR and finance, not on the side
Even Microsoft, which takes responsible AI extremely seriously, has faced reputational damage from its own AI, and the EU AI Act is set to make financial liability very real. Didem Un Ates, who sits on the Responsible AI Governance Board for Microsoft Customer and Partner Solutions, hears the same question from chief data and AI officers: once the principles are written, where do they actually start?
Get a practical plan to make responsible AI standard, from executive backing to shared toolkits
Didem Un Ates, Head of Applied Strategy for Data and AI in Microsoft's CDO Office, argues that responsible AI must be operationalized as a standard, ongoing business function rather than discussed as principles or treated as a side experiment. She outlines four building blocks (principles, practices, tools and independent governance), walks through Microsoft's six responsible AI principles, and frames the EU AI Act as "GDPR for AI" that organizations must prepare for now.
“Operationalizing means we make it standard.”
“The easiest way to describe it, I think, is GDPR for AI. That is what it is.”
What you will learn
Frame responsible AI as an essential business function, as core as HR, finance or legal.
Structure a responsible AI program around principles, team-specific practices, standard tools and independent governance.
Explain the EU AI Act to stakeholders as GDPR for AI and plan for years of preparation.
Build a cross-functional responsible AI team and track its plan with OKRs or a scorecard.
Apply Microsoft's six principles (fairness, reliability and safety, privacy and security, inclusiveness, transparency, accountability) as a starting point to adopt or adapt.
Key takeaways from the speaker
1. Accept RAI as an essential business function
2. Enable the broader acceptance and cultural change across your organization.
3. Form a passionate cross-functional team and build your RAI chart.
4. Build your RAI operationalization plan.
5. Adopt RAI tools, toolkits, checklists, and frameworks.
Details
Level
Beginner
Format
Talk
Technical depth
Introductory technical
For
People manager, Executive, Senior individual contributor
Length
20:30
Published
July 2022
Questions this talk answers
Where should an organization start when operationalizing responsible AI?
With four building blocks: agreed principles, practices tailored to each team, standard tools and checklists, and ongoing governance by an independent body.
What is the EU AI Act, in simple terms?
Didem describes it as GDPR for AI, and expects preparation to be as long as the years spent on GDPR, or even more complicated.
What are Microsoft's six responsible AI principles?
Fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability.
Why is operationalizing responsible AI so hard?
It requires a cultural change, not just impact assessments and toolkits, and many organizations still treat AI as an experiment rather than an ongoing program.
Who should oversee responsible AI in a company?
An independent, overarching body rather than sales or engineering; at Microsoft this is the Office of Responsible AI, reporting directly to the senior leadership team.
About this talk
The EU AI Act is expected to be implemented by 2023, and it will change how businesses develop and utilize AI. Any business that offers an AI application to EU citizens or organizations will need to ensure its applications are aligned with EU Charter of Rights and various other laws, including ones in relation to discrimination and privacy. These applications will have to abide by the framework of “trustworthy AI” set forth by the Act. One major area of change will be responsible AI (“RAI”).
Respond, don't react: starting girls in AI hackathons (79 s)
What operationalizing responsible AI really means (50 s)
The EU AI Act is GDPR for AI (61 s)
I'm Didem Un Ates. I Head Applied Strategy for Data AI in Microsoft's Chief Data Office for our sales organization, which is called Microsoft Customer and Partner Solutions Team. It's fascinating to be here another year with you all. I am absolutely privileged. And before I get going on how we operationalize responsible AI and responsible metaverse, I should say, at Microsoft, I would like to share a bit about myself because I think we share a lot in this forum.
I do think, I mean, it's not just responsible AI or responsible metaverse, but we have so much to share, so much questions to learn from. I'm always here. If this time is not enough for us, feel free to reach out to me in any topic, whether diversity inclusion or, you know, responsible AI or education upskilling, I'm here. For me, I've had the pleasure to have 25 years in technology space and, excuse me, and about 10 years in data AI and recently metaverse.
For me, I was, I think, for me, I was very privileged to see all of these as a fantastic opportunity to give back, to make a positive, meaningful impact, not just on the business, not just on the revenue numbers, etc., but our overall society. And that's what really gives me happiness, fulfillment and so on, so, a bit of a look in this 10-year journey in data AI and why I'm so excited about leading operationalization of responsible AI at Microsoft right now, sales organization.
So, I started my journey with our first algorithms for content. I worked with AI startups, incubating industry solutions, AI solutions, different geographies for different industries, which was fantastic. And that role led me to join Microsoft research, where we had our first business AI team, incubating our first AI algorithms for sales, for customer service, for marketing and so on. So, again, amazing journey.
And it was, and it was, and then we moved over to engineering to launch those solutions as official products, which again was fantastic, super exciting. And it was around that time Accenture reached out to ask me to set up their Microsoft data AI practice. So, I did that again, and a few weeks after I joined, actually the COVID happened, which I'll come back to that a little bit.
But it also accelerated as I'm sure most of you, if not all of you are aware, the technology development and adoption and obviously Web3 metaverse took off around that time. We also saw acceleration in super intelligence, like open AI, deep mind, which keeps launching their new models and language models, etc. And amongst all of this acceleration, all of this very fascinating technology development and adoption, we see EU AI Act coming into force in a few months.
I'm absolutely delighted that this regulatory progress is taking place, not just in EU, of course, we are seeing it in Brazil, China and so on, because it is, we do need this. We really need the regulation to catch up. So, I also want to emphasize something that I think will resonate with most of you. When we have our career journeys, it's never a walk in the park, right?
We have sometimes some shocks, sometimes some challenges, and I would, if I may strongly recommend, when you get to experience such situations, do not react, but try to respond. Reaction is a bit like quick, right? Responding is being thoughtful about the challenge or opportunity and how you can make it a fantastic opportunity for yourself and for the society.
So, I'll give three examples. When I joined Microsoft Research, I was absolutely shocked by the lack of diversity in AI. You know, of course, diversity in technology has always been a challenge, right? That's not new. But in AI, it was even worse. I'm not going to go into the details as we have limited time, but that's what, when we, when I thought, okay, I need to do something about this.
So, we formed a volunteer team of 80 people, 80 friends around Microsoft. We hadn't even known each other, but we believed in the vision. And then we put in girls in AI hackathons and so on. This was five, six years ago. Now, this program is global, Accenture, Avanade, Microsoft, of course, and many other large companies are involved.
The curriculum feeds into other fantastic programs like Imagine Cup and so on. So, it's one of the most proudest impact responses I believe I have given in my career. Another thing was when I joined Accenture and the pandemic happened. So, I had responsibility for solution delivery, go to market, et cetera. But amongst all that, we have this global shock.
So, we said, okay, within days, we put together a humanitarian program to help together with Microsoft, Avanade, and Accenture. Forgive me, it's not, it wasn't only Accenture. We said, we are going to help any client with any call volume challenges in education, in health, of course, government, and so on. And one of the implementations that I'm most proud of happened in India with the Ministry of Health.
We were basically helping 1.3 billion citizens in 10 days, where our client, digital CEO of India, said, we are saving lives. We are really saving lives. So, another response. And I'll come to Responsible AI and why I feel incredibly proud and excited about this. With everything going on in our space, amazing, you know, we wouldn't be here if it weren't for data AI, right?
We wouldn't be almost getting out of the pandemic or managing it so fast. So, these technologies are amazing. But how we manage them, how we make sure they are responsible and inclusive are important. And I also want to emphasize metaverse, you know, AI took a while for regulators to start developing regulations. And I spoke at the EU Commission panel a few months ago, said the same thing.
We don't have to wait for the same time for metaverse, right? Responsible metaverse needs to happen hand in hand with responsible AI. So, let's get going. I want to just recap this introduction because I think we will all continue to face this. Viktor Frankl is a Holocaust survivor and he wrote an amazing book, one of my favorite books called Man's Search for Meaning.
And I love this quote from his book, "Between stimulus and response lies a space. In that space lie our freedom and power to choose a response, not a reaction. In our response lies our growth and our happiness." I couldn't agree more. And I think this really applies perfectly well with technology. So, I would like to say each of us must consciously not just develop and shape technology, but we need to shape its impact as well.
We are all on the hook, if I may say so. We are all responsible. So, let's get going. Why? What are the risks of this? I mean, we all have our jobs. I love these technologies and I love developing them, scaling them and so on. But there are risks, right? There are unintended consequences and it's not just a one-off.
These technologies have a life of their own. So, we need to manage them on an ongoing basis. There are risks of reputational damage as even we as Microsoft who take this extremely seriously and I think we are thought leaders, if not the leader here. We have faced these challenges. You will, some of you will remember the tie bot.
Financial liability, of course, is very real and it will become even more significant once we have these EU AI Act and similar regulations in place in a matter of months. And then operationalizing this mindset actually is not easy. It is not easy because it really includes a cultural change, not just impact assessments and tools and toolkits, but actually a massive change in how we think of technology and it's how we feel responsible for its impact at scale, not just as individuals or groups of people.
And of course, all of this may cause customer partner friction resulting in very real business damage. So, you know, why people, when people ask why operational, I mean, hopefully nobody asked this question. I would like to think, but operationalizing is different than talking about responsible AI or having a reading something and getting upset about it. Operationalizing means we make it standard.
We make it part of life. We make it systematic. And that's not easy. That's why I wanted to take a bit of your, let's say, brain power to share best practices and join forces if you're interested. The drivers are, we all want to innovate. We love this stuff. We want to do it responsibly, obviously. But, you know, it may be you, me and whoever, DDAM thinking about this, but not all of the population.
So we need to also empower others on how to do that, encourage and empower others so that we can foster a positive impact with these amazing technologies and shape their solutions. So where to start? In my day job, I work with chief data officers, chief AI officers, and analytics officers. And usually we co-create roadmaps. We say, what are we going to do together in the next five years?
And that's my job. And very naturally, most, if not all of these executives ask me the question, DDAM, where do I start? I don't want to be talking about principles. Where do I execute responsible AI? How do I do it? So here is how principles are essential. I'm going to share Microsoft's principles. We have six of them.
We have been discussing, agreeing, debating them since, you know, 2015, 16. You're most welcome to adopt them or adapt them to your needs. Then practices are also, of course, very important. Why? Because we are finding one product team needs to implement responsible AI, operationalize responsible AI, is actually very different than another product team or a business team.
Because let's say one is B2C consumer, another one is B2B. So there are very simple realities of each business, each team in terms of operationalizing it. Next, tools are, of course, very important. Why? Because we need to standardize. We need to make sure they serve the purpose and responsible AI doesn't become an ad hoc, one-off experimentation. So tools, toolkits, checklists, frameworks, all of these, we have them on our responsible AI website.
If you need to use any of them, feel free to go. They are all open. And if anything is unclear, feel free to reach out to me or any one of us on that website. And lastly, of course, governance. This is an ongoing effort. It's not a one-off. That's one of the most important mindset changes that we need to really internalize.
It needs to be overseen by an independent body, not, you know, by sales team, by engineers, but somebody who's overarching. So this space in our world is called Office of Responsible AI reporting directly to our senior leadership team. So I am incredibly, incredibly grateful to the regulators. I cannot wait for the EU AI Act, for example, and other similars to come into force.
So we join forces and everything becomes more formal, but these are all in development. And in the meantime, in parallel, we need to, in our organizations, we need to enable and inspire these very important changes. One is the mindset change. What I mean here is responsible AI is part of business. Like any other function, HR, finance, legal, you can't do, you can't have a company without these functions.
So it's the same with responsible AI, responsible metaverse, if you're in the AI business, of course. And the cultural change is most often in our partners, customers, and even at Microsoft, we're seeing this AI thing is some kind of experimentation. I mean, obviously it's impact and it's very serious and so on. You know, many are becoming AI platforms rather than AI applications, vertical, you know, implementations.
But it's still seen as an experimentation rather than an ongoing effort, ongoing massive program. So these two things are very important for success in operationalizing responsible AI. And when I ask, when I get the question, what is EU AI act? What, you know, how would you describe it? The easiest way to describe it, I think, is GDPR for AI.
That is what it is. And some of you will remember, we worked, we got, we prepared for years, at least three years for GDPR. It's the same, maybe even more complicated for AI, you know, responsible AI and responsible metaverse to happen. I had promised to share our principles, so we have six of them. Fairness, of course, is very important.
AI systems need to be developed so that they treat everyone fairly and they avoid affecting similar groups in different ways. So that shouldn't happen. Another very important area is reliability and safety. They must operate reliably, safely, and consistently under normal circumstances and also unexpected conditions. This is easier said than that. So very important we have multiple work streams under this principle.
Privacy and security, of course. These systems need to comply with privacy laws, need to respect our data, how we collect, how they collect it, how they store it, how they use it. And make sure we, as the consumers, have the appropriate controls to choose how our data is used. Inclusiveness, very close to my heart, obviously. So these systems have to incorporate and address a broad range of diverse needs, diverse barriers, capabilities, experiences.
And transparency allows, I mean, anybody dealing, engaging with an AI system to understand so that, understand the behavioral functional components so that we can identify performance issues and risks, privacy concerns, biases, et cetera, and tackle them effectively and rapidly. And finally, accountability, of course. And if we're designing and deploying an AI system, we are accountable for how they operate, which again requires ongoing maintenance, ongoing attention.
So, how to start or call to action, whatever we, and I think we need to accelerate, not just responsible AI, but also responsible metaverse. It has to come from the top. Let's face it. If there is no support from executives, most things in business do not happen. Same thing with responsible AI. And then we need to treat responsible AI as an essential business function.
As I mentioned, just like HR, finance. It has to be there. It's not like a volunteer project. It's not on the side. It's core. The cultural change we discussed, that has to happen. We need to ensure this is ongoing. This is, there will be experimentation mode. There will be unintended consequences. Even if we take all the impact assessments, if we take all the checklists, things could still go wrong.
But the cultural change will embrace that and still move forward with it. And of course, cross-functional responsible AI team is incredibly important to make this real. Why? Because we need to include many diverse perspectives to be able to operationalize it effectively. We need to include different business units, different teams, different stakeholders in the journey, internally and externally, I should say.
And as with anything else in business, we have to have a plan. If you're using all KRs, objectives and key results, or if you're using a scorecard, it has to have your goals, your milestones, your timeline, and check for progress with the key important metrics for your business. And finally, tools, toolkits, checklists, frameworks, these save a lot of time.
They also enable systematic approach, standardization. So we are most welcome to share these with you. If you have any questions, feel free to reach out anytime. We believe every AI solution must be a responsible AI solution. So, and as I mentioned in the beginning, with great power comes great responsibility. Thank you so much. I'm always here. Even if our time is up right now, feel free to ask any questions today or afterwards.
I'm just checking the chat very quickly. Again, I'm sorry for the delay. I think there was a mix up with the links because I ended up in another session and took me a while. Thanks for staying in. Hi, Jennifer. Nice to see you. Savita, I don't know if you're still around. Very happy to connect, of course. If there are no more questions, I'm going to leave.
And wonderful. Thank you so much for your time. Wishing you a fantastic conference. Thank you all very much.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
How do you make responsible AI a standard part of business?
Microsoft's Didem Un Ates explains why responsible AI must move from principles to a systematic, governed business function.
In short
Responsible AI only works when it stops being a conversation and becomes standard practice. Didem Un Ates, Head of Applied Strategy for Data and AI in Microsoft's CDO Office, argues that companies should treat it as an essential business function, built on principles, team-specific practices, shared tools and independent governance. With the EU AI Act approaching, which she calls "GDPR for AI", she urges leaders to start the cultural change now.
Key takeaways
Accept RAI as an essential business function.
Enable broader acceptance and cultural change across your organization.
Form a passionate cross-functional team and build your RAI chart.
Build your RAI operationalization plan.
Adopt RAI tools, toolkits, checklists and frameworks.
Respond, do not react
Didem opens with her own path: 25 years in technology, about ten of them in data and AI, from incubating AI startups to Microsoft Research's first business AI team and then setting up Accenture's Microsoft data and AI practice. Along the way she learned a lesson she applies to technology as much as to careers. When a shock or challenge arrives, she recommends pausing before acting.
"Reaction is a bit like quick, right? Responding is being thoughtful about the challenge or opportunity and how you can make it a fantastic opportunity for yourself and for the society." (4:03)
She gives two examples. Shocked by the lack of diversity in AI when she joined Microsoft Research, she and a volunteer team of 80 colleagues launched girls in AI hackathons, which she says are now global and involve Accenture, Avanade, Microsoft and other large companies. And when the pandemic hit weeks after she joined Accenture, her team built a humanitarian program within days; she describes an implementation with India's Ministry of Health that helped 1.3 billion citizens in 10 days. Responsible AI, she suggests, is the next such response. Quoting Viktor Frankl's Man's Search for Meaning, she argues that each person must shape not only technology but its impact.
Why operationalizing is different from talking
Didem is candid about the risks. AI technologies "have a life of their own" and need ongoing management. She lists unintended consequences, reputational damage (recalling Microsoft's own Tay bot), financial liability that will grow once the EU AI Act takes effect, and friction with customers and partners that can cause real business damage.
"We are all on the hook, if I may say so. We are all responsible." (8:19)
The hard part, she stresses, is that this is a cultural change, not just a matter of impact assessments and toolkits. Reading about AI harms and feeling upset is not the same as building a system that prevents them.
"Operationalizing means we make it standard." (9:45)
Four building blocks: principles, practices, tools and governance
In her day job, Didem co-creates multi-year roadmaps with chief data, AI and analytics officers, and they all ask where to begin. Her answer has four parts. Principles come first: Microsoft has six, debated since around 2015, and she invites organizations to adopt or adapt them. Practices come next, because a B2C product team operationalizes responsible AI very differently from a B2B or business team. Tools, toolkits, checklists and frameworks keep the work standardized so it does not become an ad hoc experiment; she notes Microsoft publishes these openly on its responsible AI website. Finally, governance must be ongoing and independent, not owned by sales or engineering. At Microsoft that role belongs to the Office of Responsible AI, reporting directly to the senior leadership team.
"It's not a one-off. That's one of the most important mindset changes that we need to really internalize." (12:51)
Mindset, culture and the EU AI Act
While regulation develops, Didem says organizations must drive two changes themselves. The first is mindset: responsible AI is a business function like HR, finance or legal, for any company in the AI business.
"What I mean here is responsible AI is part of business." (13:24)
The second is culture. Many customers, partners and even teams at Microsoft still see AI as experimentation rather than an ongoing, large-scale program. For anyone wondering how to explain the coming regulation, she offers a simple comparison, recalling that organizations prepared for at least three years for GDPR.
"The easiest way to describe it, I think, is GDPR for AI. That is what it is." (14:57)
Six principles and a call to action
Didem walks through Microsoft's six principles: fairness, so similar groups are not affected differently; reliability and safety under normal and unexpected conditions; privacy and security, with user control over data; inclusiveness of diverse needs and abilities; transparency, so people can spot performance issues, risks and biases; and accountability for how deployed systems operate.
Her call to action follows. Responsible AI needs support from the top, because without executive backing most things in business do not happen. It must be treated as a core function, not a volunteer project.
She adds that culture must accept that things can still go wrong even with every checklist in place, that a cross-functional team brings the diverse perspectives needed, that the work needs a plan with goals, milestones and metrics (using OKRs or a scorecard), and that tools save time and enable standardization. She also asks that responsible metaverse develop hand in hand with responsible AI, rather than waiting years for regulation.
"We believe every AI solution must be a responsible AI solution." (19:20)
Watch the talk
Operationalizing Responsible AI, Didem Un Ates, 21 minutes: https://www.womentech.net/video/operationalizing-responsible-ai-didem-un-ates
About the speaker
Didem Un Ates is Head of Applied Strategy, Data & AI, CDO Office at Microsoft, and sits on the Responsible AI Governance Board for Microsoft Customer and Partner Solutions. She directly supported the birth and incubation of Microsoft Business AI Solutions at Microsoft Research, led the Alice Envisions the Future/Girls in AI Hackathons, and previously led Accenture's Microsoft Data & AI Business in Europe.
Questions this talk answers
Where should an organization start when operationalizing responsible AI?
What is the EU AI Act, in simple terms?
What are Microsoft's six responsible AI principles?
Why is operationalizing responsible AI so hard?
Who should oversee responsible AI in a company?
Drafted with AI from the talk transcript. Editor review pending.
Frameworks and ideas mentioned
Microsoft responsible AI principles
EU AI Act
GDPR
OKRs
Scorecard
Resources mentioned
Man's Search for Meaning (Viktor Frankl)
Microsoft responsible AI website (tools, toolkits, checklists and frameworks)
Cite this talk
Didem Un Ates. "Operationalizing Responsible AI." WomenTech Network, 2022. https://www.womentech.net/video/operationalizing-responsible-ai-didem-un-ates