Article 21. AI Risk Training: Role-based tailoring

Spoken by a human version of this article.

AI literacy is growing in importance (e.g., EU AI Act, IAIS).
AI literacy needs vary across roles.
Even "AI professionals" need AI Risk training.

Links

• EU AI Act: The European Union Artificial Intelligence Act - specific expectation about “AI literacy”.
• IAIS: The International Association of Insurance Supervisors is developing a guidance paper on the supervision of AI.

To subscribe to the weekly articles: https://riskinsights.com.au/blog#subscribe

About this podcast

A podcast for Financial Services leaders, where we discuss fairness and accuracy in the use of data, algorithms, and AI.

Hosted by Yusuf Moolla.
Produced by Risk Insights (riskinsights.com.au).

Episode Transcript

Yusuf (00:14):
This article was published in January 2025.
It's titled AI risk trainingrole based tethering.
The background to this is we spokeabout AI literacy or education and
awareness in a previous article.
, and so we touched on the importanceof algorithm slash AI awareness

(00:34):
and education in article 18.
So you go to the episode thatwas published in December,
article 18 algorithm integritytraining and awareness.
And so this is a follow up to that.
Talking about how AI literacyis growing in importance.
So for example, the EU AI Act andthe, , International Association
of Insurance Supervisors.

(00:56):
But AI literacy needs willvary across roles, and even AI
professionals need AI risk training.
So that's the shortdiscussion in this article.
, Here we go.
Growing expectations.
Article 4 of the EU AI Actis dedicated to AI literacy.

(01:16):
In a recent podcast episode, sothat was the first guest episode,
Ryan Carrier, founder of 4Humanityexplained two important things
to note about this article.
Number one, most of the provisionsof the Act come into force in August
2025 with others in 2026 or later.

(01:38):
But AI literacy requirements willstart to apply on the 2nd of Feb 2025.
second important things to note, most ofthe provisions of the Act focus on high
risk AI systems, but the AI literacyrequirements, so again, that's Article
4, those requirements apply to all.
all AI systems.

(02:00):
Then we have the InternationalAssociation of Insurance Supervisors,
who are developing a guidance paperon the supervision of AI, and we'll
put a link to that in the notes.
And draft version of that guidancepaper contains various references to
education, training, and awareness.
And so this is an important topic.

(02:21):
Tailoring it to the audience.
To be effective, AI literacy effortsneed to be tailored to the audience.
What the data scientists need tobe aware of is quite different to
what most employees need to know.
So the training will needto be differentiated.
In that same podcast episodewhere we spoke to Ryan Carrier, he
explained one way to solve for this.

(02:42):
For humanity has developeddistinct personas, and they look
something like this, then not anexact terminology in use here,
but , something along these lines.
So we've got five AI leaders and thentop management and oversight bodies,
employees working with AI system.
all other employees, and then AI subjects.

(03:07):
These personas represent different levelsof interaction with those AI systems,
and then corresponding literacy needs.
Let's explore each of the personas, notingthat some of these may look different
for you depending on your jurisdiction,the types of systems you're implementing,
etc. The first one was AI leaders.
So AI leaders typically include AIproject managers, data scientists,

(03:30):
and machine learning engineers.
Their responsibilities includedeveloping and implementing AI models
for credit scoring, fraud detection,and risk assessment, and others,
and staying updated on the latestAI technologies and their potential
applications in financial services.
The challenge is that this groupis often neglected in terms of risk

(03:54):
focused training, possibly becauseit is assumed that they are already
aware, which is often not the case.
We've seen some data science professionalsthat are very technical focused, so
you may come across models that areaccurate but still exhibit bias.
second persona being top management andoversight bodies, and this group includes

(04:15):
c suite executives, and senior managers.
Their AI literacy needs includeunderstanding the strategic implications
of AI in financial services, overseeingAI governance and risk management
frameworks, and making informed decisionsabout AI investments and implementation.
The challenge with this group isthat they might be responsible

(04:39):
for ensuring AI literacy.
So the level of literacy across theorganization may cascade from there.
The third group are employeesworking with the AI system.
So frontline staff and middle managerswho directly interact with AI systems.
In banking and insurance, this couldinclude customer service representatives

(04:59):
using AI powered support systems, or fraudrisk analysts using the outputs of machine
learning triage systems as examples.
Their AI literacy needs may includeunderstanding how AI impacts their
daily work and decision makingprocesses, recognizing potential biases
or errors in AI outputs, Effectivelychecking outputs before using them

(05:23):
to communicate with customers.
The fourth group are all other employees.
So these are staff who may notdirectly work with algorithms.
But need a basic understandingof the impact of those
algorithms . And then AI subjects.
So these are customersand end users largely.
Their literacy needs will depend on thenature of the systems that are used.

(05:47):
It's not a one size fits all concept.
Each group has unique needsand responsibilities when
it comes to understanding AIsystems or algorithmic systems.
Even AI people need AI literacytraining or risk training.
This is especially important tobroaden awareness of risks and

(06:08):
balance out a purely technical focus.
By tailoring AI literacy effortsto these distinct personas, banks
and insurance companies can ensurethat all stakeholders have the
appropriate level of understanding.
That's the end of that article.
Thanks for listening.