Five Pillars of Trustworthy AI
Key Points
- The speaker’s three biggest night‑time worries are climate change, the hidden impact of AI on personal decisions (loans, jobs, college admissions), and the mistaken belief that AI is inherently unbiased or ethically perfect.
- Over 80 % of AI proof‑of‑concept projects stall during testing, mainly because decision‑makers don’t trust the model’s outcomes.
- Trust in an AI system can be built on five pillars: fairness to all groups, explainability of data and methods, robustness against manipulation, transparency about usage and metadata, and protection of data privacy.
- IBM proposes three guiding AI principles: AI should augment—not replace—human intelligence; data and the insights derived from it belong to their creator; and AI systems must be designed and used responsibly.
- To successfully deploy AI, organizations must proactively address these trust pillars and adhere to principled guidelines, ensuring that AI decisions are fair, understandable, secure, transparent, and privacy‑respecting.
Sections
- Three Nighttime Worries: Climate, AI, Trust - The speaker highlights three sleepless concerns—climate change, AI-driven decisions that affect personal outcomes, and the mistaken belief that AI is inherently unbiased—explaining how mistrust stalls most AI proof‑of‑concepts and introducing a five‑pillar trust framework.
- AI Trust: Principles & Challenges - The speaker outlines five trust pillars, IBM’s three AI principles—augmenting human intelligence, creator‑owned data, and transparent explainability—and stresses that building trustworthy AI is a socio‑technical, holistic effort centered on people, culture, and team diversity.
Full Transcript
# Five Pillars of Trustworthy AI **Source:** [https://www.youtube.com/watch?v=aGwYtUzMQUk](https://www.youtube.com/watch?v=aGwYtUzMQUk) **Duration:** 00:06:09 ## Summary - The speaker’s three biggest night‑time worries are climate change, the hidden impact of AI on personal decisions (loans, jobs, college admissions), and the mistaken belief that AI is inherently unbiased or ethically perfect. - Over 80 % of AI proof‑of‑concept projects stall during testing, mainly because decision‑makers don’t trust the model’s outcomes. - Trust in an AI system can be built on five pillars: fairness to all groups, explainability of data and methods, robustness against manipulation, transparency about usage and metadata, and protection of data privacy. - IBM proposes three guiding AI principles: AI should augment—not replace—human intelligence; data and the insights derived from it belong to their creator; and AI systems must be designed and used responsibly. - To successfully deploy AI, organizations must proactively address these trust pillars and adhere to principled guidelines, ensuring that AI decisions are fair, understandable, secure, transparent, and privacy‑respecting. ## Sections - [00:00:00](https://www.youtube.com/watch?v=aGwYtUzMQUk&t=0s) **Three Nighttime Worries: Climate, AI, Trust** - The speaker highlights three sleepless concerns—climate change, AI-driven decisions that affect personal outcomes, and the mistaken belief that AI is inherently unbiased—explaining how mistrust stalls most AI proof‑of‑concepts and introducing a five‑pillar trust framework. - [00:03:09](https://www.youtube.com/watch?v=aGwYtUzMQUk&t=189s) **AI Trust: Principles & Challenges** - The speaker outlines five trust pillars, IBM’s three AI principles—augmenting human intelligence, creator‑owned data, and transparent explainability—and stresses that building trustworthy AI is a socio‑technical, holistic effort centered on people, culture, and team diversity. ## Full Transcript
I want to start off with talking to you about three things that keep me up at night, right?
Three things: the first, and it may be, you know, very common for you too, is climate change.
Climate change absolutely keeps me up at night. The second thing that keeps me up at night
is that people may have no idea that an artificial intelligence is making a decision that directly
impacts their lives - what percentage interest rate you get on your loan, whether you get that
job that you applied for, whether your kid gets into that college that they really want to go to.
Today AI is making decisions that directly impact you.
The third thing that keeps me up at night is: even when people know that an AI
is making a decision about them, they may assume that because it's not a fallible human with bias,
that somehow the AI is going to make a decision that's morally or ethically squeaky clean,
and that could not be farther from the truth.
So, if you think about organizations and what happens over 80% of the time
proof of concepts associated with artificial intelligence actually gets stalled in testing
and more often than not it is because people do not trust the results from that AI model.
So, we're going to talk a lot about trust,
and when thinking about trust (I’m going to switch colors here) there's actually five pillars.
OK, when you're thinking about what does it take to earn trust in an artificial intelligence
that's being made by your organization or being procured by your organization: five pillars.
The first thing to be thinking about is fairness. How can you ensure
that the AI model is fair towards everybody in particular historically underrepresented groups.
OK, the second is explainable is your AI model explainable such that you'd be able to tell
somebody, an end user, what data sets were being used in order to curate that model, what methods,
what expertise was the data lineage in provenance associated with, how that model was trained.
The third: robustness. Can you assure end users that nobody can hack such an AI model such that
a person could disadvantage willfully other people and or
make the results of that model benefit one particular person over another?
The fourth is transparency. Are you telling people, right off the bat, that the AI model
is indeed being used to make that decision and are you giving
people access to a fact sheet or metadata so that they can learn more about that model?
And the fifth one is: are you assuring people's data privacy?
So, those are the five pillars. OK, now IBM has come up with three principles
when thinking about AI in an organization. The first being that the purpose of artificial
intelligence is really meant to be to augment human intelligence not to replace it.
The second is that data and the insights from those data belong to the creator alone
OK, and the third is that AI systems, and I would opine the entire AI life cycle, really should be
transparent and explainable, right? So, so, those are the five pillars.
Now, the next thing I want you to remember as you're thinking about this space of earning
trust and artificial intelligence is that this is not a technological challenge. It can't be solved
with just throwing tools and tech over some kind of fence. This is a socio-technological challenge.
"Social" meaning people, people, people. Socio-technological challenges because
it's a socio-technological challenge it must be addressed holistically, okay?
"Holistically" meaning there's three major things that you should think about. I mentioned people,
people the culture of your organization, right? Thinking about the diversity of your teams,
you know, your data science team. Who is curating that data to train that model? How many women are
on that team? How many minorities are on that team, right? Think about diversity. I don't
know if you've ever heard of the the "wisdom of crowds". That's actually a proven mathematical
theory: the more diverse your group of people, the less chance for error, and that is absolutely
true in the realm of artificial intelligence. The second thing is process or governance, right?
What is it that use your organization what are you going to promise your
both your employees as well as the market with respect to what standards you're going to stand by
for your AI model in terms of things like fairness and explainability accountability, etc., right?
And the third area is tooling, right? What are the tools, AI engineering methods, frameworks
that you can use in order to ensure these things, ensure those five pillars, and we're
gonna do a deep dive into that as well, but the next show that I’m going to be running with you
we're actually going to be talking about this one. About people and culture. So, stay tuned.
If you like this video and series, please comment below stay tuned for more videos
that are part of this series and to get updates please like and subscribe.