WEBVTT

00:00:05.760 --> 00:00:08.080
Hi, everyone.
My name is Mihaela Vorvoreanu.

00:00:08.240 --> 00:00:12.080
I am a responsible AI researcher
and educator at Microsoft,

00:00:12.160 --> 00:00:16.040
and I am super happy to introduce you
to the topic of responsible AI.

00:00:16.120 --> 00:00:21.640
One thing you might not have guessed
about me is that I like to drive,

00:00:21.840 --> 00:00:25.440
and I actually like to drive
fast, powerful cars,

00:00:25.960 --> 00:00:28.840
even though I recognize they're
not that good for the environment.

00:00:29.320 --> 00:00:33.920
Now, it is highly unlikely that
I will ever be driving a car like the one

00:00:34.000 --> 00:00:35.560
in the picture, this Lamborghini.

00:00:36.040 --> 00:00:37.480
You know, one can daydream.

00:00:37.960 --> 00:00:43.760
But let's think a little bit about this
beautiful, powerful piece of machinery,

00:00:44.080 --> 00:00:46.000
and let's think about safety.

00:00:46.560 --> 00:00:51.400
Of course, it's a lot of fun to drive,
but you also need to stay safe.

00:00:51.880 --> 00:00:55.800
And here, safety is
a shared responsibility

00:00:55.880 --> 00:00:58.880
between the auto manufacturer who needs

00:00:58.960 --> 00:01:03.520
to make sure that car is safe in
different conditions, has safety features

00:01:03.600 --> 00:01:05.920
like airbags and seatbelts.

00:01:06.360 --> 00:01:11.840
But it is also the driver's responsibility
not only to put the seatbelt on,

00:01:11.920 --> 00:01:18.320
but also to learn how to use this kind
of car, how to maneuver this kind of car

00:01:18.400 --> 00:01:22.560
in different types of conditions,
to learn how to drive it in the rain,

00:01:22.640 --> 00:01:26.120
for example, or to learn that maybe
you shouldn't drive it in the rain.

00:01:26.200 --> 00:01:30.000
Now, the same is true of AI systems.

00:01:30.520 --> 00:01:34.320
They are a very, very,
very powerful piece of technology.

00:01:34.400 --> 00:01:41.040
And a lot of the responsibility for safety
rests with the companies that build

00:01:41.120 --> 00:01:43.840
these AI systems and AI applications.

00:01:43.920 --> 00:01:47.920
There is an entire discipline
of responsible AI where we work

00:01:48.000 --> 00:01:49.680
to make these systems safer.

00:01:49.760 --> 00:01:53.720
But a little bit of that responsibility
also sits with the users

00:01:53.800 --> 00:01:55.480
of these technologies.

00:01:55.560 --> 00:02:00.880
So, let's think for a second, what
might that user responsibility entail.

00:02:00.960 --> 00:02:04.880
It might entail things
like understanding the system's

00:02:04.960 --> 00:02:07.280
capabilities and limitations.

00:02:07.520 --> 00:02:11.880
It might also entail learning
how to use the system effectively.

00:02:12.440 --> 00:02:17.240
And this is why I am very pleased
to have partnered with Founderz

00:02:17.320 --> 00:02:22.480
to get off the ground this Toolkit
for the Responsible Use of AI.

00:02:22.560 --> 00:02:25.840
You can see that its initial
version is already online.

00:02:25.960 --> 00:02:31.760
And this toolkit is an interactive guide
that we designed it for empowering

00:02:31.840 --> 00:02:38.120
individuals to interact with AI systems
responsibly, effectively, and safely

00:02:38.320 --> 00:02:41.040
within their business environment.

00:02:41.360 --> 00:02:45.440
So, let's look a little bit at this
toolkit and what you can find in it.

00:02:45.760 --> 00:02:50.200
You can find right now
four types of information.

00:02:50.280 --> 00:02:52.640
First of all, there's foundations.

00:02:53.120 --> 00:02:56.800
These are basic things,
basic principles that

00:02:57.000 --> 00:03:00.160
everyone who uses AI
for work should know.

00:03:00.600 --> 00:03:04.440
Then you can find a catalog
of prompting techniques.

00:03:04.920 --> 00:03:09.000
And these are techniques that you can use
to get more out of generative AI,

00:03:09.080 --> 00:03:15.200
to get more out of interacting with large
language model-powered applications.

00:03:15.760 --> 00:03:18.640
You can also find applied use guides

00:03:18.840 --> 00:03:23.200
that will tell you, for example, how you
can make the best out of generative AI

00:03:23.360 --> 00:03:25.880
if you're working in a certain area

00:03:25.960 --> 00:03:30.240
like marketing or HR or customer service.

00:03:30.320 --> 00:03:35.560
And then you will also be able to find
case studies of teams or companies

00:03:35.640 --> 00:03:39.640
who have used AI effectively
so you can learn from them.

00:03:39.720 --> 00:03:46.320
Now, I would like you to keep in mind that
the Responsible Use of AI Toolkit

00:03:46.400 --> 00:03:48.560
right now is a starting framework.

00:03:48.800 --> 00:03:53.880
We put it in place as a framework
for others to build an add-on so we can

00:03:53.960 --> 00:03:58.480
collectively learn from each other and
create a useful library of information.

00:03:58.560 --> 00:04:02.600
I also need you to keep
in mind that if you follow the steps

00:04:02.680 --> 00:04:06.280
or a certain prompting technique,
it's not a guarantee that everything

00:04:06.360 --> 00:04:07.720
is going to work out fine.

00:04:07.800 --> 00:04:10.800
As we know, AI systems are probabilistic.

00:04:10.880 --> 00:04:15.720
The same prompt or the same input
can generate various outputs.

00:04:15.800 --> 00:04:20.160
And so, what might this user
responsibility entail?

00:04:20.240 --> 00:04:24.880
It might entail understanding the system's
capabilities and limitations

00:04:25.120 --> 00:04:28.520
and learning how to use
the system effectively.