Summary
Notes
Transcript
This is a lot for the individuals, but if you take one step back and you look at maybe 20 of the 28 people or whatever are from BC and stuff. We're going to be having a high concentration of this knowledge here and it will probably continue to grow and spread and stuff. Looking for the community collective up-level of consciousness. you know, here. So that's kind of where I'm at.
So a little bit more about me. I'm Sarah Downey. I'm in Victoria, BC, and that's on the unceded territories of the Songhees, Senekotan, and Wissanic nations. And I work at the intersection of AI governance and nonprofit leadership. And I've spent the last couple of years helping Canadian nonprofits figure out how to engage with AI responsibly. So in the nonprofit work that I'm in, I help people use AI to lighten their workload, protect their values and their privacy and their data. And really the part I love is to help them use AI to strengthen their work culture and their human connections. So that's the lens that I'm bringing to this course for my work. But really, I'm not here to tell anyone how to feel about AI in any way.
All right, awesome. Thank you so much, Sarah. Thanks everyone for joining for your time. Really appreciate the folks here. My name is Martin Lopatka. I'm joining you from sunny North Vancouver as well. See that there's folks in the cohort here. I've been working and studying and alternating between working and studying in the space of AI since the early 2000s. I did my master's in AI way back before any of these sort of generative AI and buzzwords Back in 2006, I did a PhD in statistics, subsequently looking at the generalization of AI from graph theoretical approaches and had the privilege of working in a lot of large scale open source projects at Mozilla for just under six years, focused mostly on privacy preserving machine learning techniques.
So at a moment where there was sort of a divide in the general consensus that data AI type things weren't compatible with a privacy guarantee. So that's been a professional trajectory of mine. For the last several years, I've been working in professional services and worked with a Big variety of really interesting, amazing clients at the enterprise scale, working on impactful systems, high-risk systems, and a little bit on the audit side as well. So Sarah and I really connected over this course.
Great intro. Well, today's a big day for me personally because I've been dreaming of doing this for like six months. So over the last three years, I've led 26-week courses for professionals called AI Upgrade Programs. I've run eight cohorts of the Creative Pros AI Upgrade, six of the Media and Comms Pros, three of the sales ones. And so I've been in close relationship with a whole bunch of learners who are just trying to figure out where to go in this moment. And those courses are all a lot more like skills and tools based.
And everyone kind of needs as a next intermediate level, people keep coming back to me asking for the ethical frameworks, the responsible frameworks, the governance and deployment frameworks, because a lot of the people that were early adopters have gone on to be change agents inside their organizations. And I've been leading a lot of these, adoption efforts of AI inside their orgs. So I've been percolating this. It's been an energetic build in my notion over the last six months. And then Martin and I started Becoming Friends about a year ago. And he's a first order technical thought leader when it comes to responsible and ethical AI usage.
We're going to have to cut way more out of the weekly talks than we're able to actually give. But we've set up all sorts of other mechanisms, including the Notion and the Discord and stuff like that, in order to kind of supplement your learnings. You will get as much out of this course as you put into it in terms of watching the videos beforehand, doing the readings.
If we all commit to doing that, we can all expect that we're going to be showing up into a room with people with this on the top of their mind who have also invested themselves in it. The last thing I'll say before flipping it over, and the reason why we're going to go around the room and do intros, even though it's going to be a little hard to keep it to 20 seconds apiece, is because this is a very powerful cohort of people.
is people you'll likely stay connected with after this learning journey and will be collaborators on implementation or other types of things related to this. So I've learned the value of these courses and the reason why it's worth it to pay for them instead of just watching YouTube videos or something like that is the value of the relationships and the connections we make along the way. So yeah, thanks for being here.
So for the next four weeks, we are meeting on Zoom at three o'clock. Is anybody here not in? this time zone, Pacific time zone. Maybe put it in the chat if you're not, just want to acknowledge you. And then except for the last class is going to be an extra 90 minutes because we're going to have our capstone presentations. Okay, Bruce is not in this time zone. Well, please adjust. Yes, and Chris will say a little bit more about that after.
So yeah, between our sessions, really everything lives in the Notion Hub. And all of the course content is nested inside toggles. under the content toggle. You're also going to find all your pre-reading in there, exercises, artifacts, week by week. We are building it live in the course hub. So you will see things change and update as we add things. And one thing that we want to know is that everything you share is valuable and you can use the chat here anytime to ask questions.
And we will respond during class if we can. I know Chris is going to be all over the chat and getting to that. we likely won't get to everything. And so we're going to actually share a log of the chat in the Notion dashboard week by week. And so we'll be able to answer things that we've missed there. So our intention is that we're actually going to get to all of your questions. And then our Discord, we can go deeper into conversations in Discord.
If we want to nerd out on a topic, that would be the place to do that. What did I miss? Chris Martin, want to fill in any gaps?
So I've also learned that in addition to the cohort-based learning, this capstone thing is a really valuable part of this process. And the idea here is that over the course of these four weeks, as you're doing exercises, as you're building little tools and stuff, that you have in your mind some final project that you can create. It could be like a dashboard or an assistant. It could be an ethics framework or ethics evaluation matrix.
It could be a remix of a lot of the tools that we're building here. But I want everyone to think about a capstone project. Probably in week three, we'll go around the room and I might ask what you might build for a capstone project. And you'll say that in front of other people because you'll find there's a lot of people that are kind of exploring similar ideas, kind of give each other some feedback and stuff.
So just start thinking now, like what's something that would be useful to your job or your work around AI?
Last logistic I think is, I shouldn't call it a logistic, it's our agreements for our weekly sessions. And I'm gonna pop them in the chat and just name them really quickly. Oh, I can only pop three at a time because my chat's too long. So first one, this time belongs to all of us. So just being mindful before you, you know, share a question or a comment, useful gut check-in, like, is this something that the whole group will benefit from? If it is a conversation that can be one-on-one with us after, just know that we're accessible outside of the sessions for after and that group time that we have together, it's a shared resource.
Let's spend it well. This group is unique. We come from different sectors, different roles. lived experiences and we are all in the same room at the same time, which is amazing. And we want to make our time count. The second, your questions are gold. The chat is your parking lot. We're gonna track them. We're gonna follow up. Nothing disappears. Nothing is a bother. We wanna hear from you. We will keep us on track and we are asking for your support. So if we need to move the conversation forward, you might hear us say, let's hold that and keep going because we're taking responsibility for the timing and for covering what the whole group needs.
Again, we have, that's why we've created this discord space so we can dive into things more further in between the sessions. The last two, which I still have to paste. One second. There you go. cameras on we love that we love to have your presence uh it creates conditions for trust and real exchange and learning that sticks Of course, things happen. Like me last week in a meeting who needed to turn off my camera because I realized my...
shirt was on backwards and I wanted to fix that. You know, sometimes there are things you don't have to explain it. We trust you if you need to turn off your camera, but just know that the intention is camera's on. And then lastly, really important, different perspectives are a given. We all come from different backgrounds and that is such a huge value, right? So disagreement here isn't disruption, it's engagement.
And the bar we want to hold for each other is generous listening and genuine curiosity across difference. So just, yeah, come ready to have your thinking stretched. So for everyone, if you can show you are in agreement, I want you to respond in the chat with an emoji.
We post the videos and the transcripts on the Notion after the course.
And so like if you missed a course, you should watch the video, but you should also take the transcript, drop it into an LLM and be like, yo, what's the five most important things I missed from this course here today? Okay. Um.
I know there's some folks that aren't in discord yet and have asked about how to, how to get in there. I'm not very helpful troubleshooting that, but at the end of the class today, we're going to do a discord onboarding.
Yeah, that's planned for just at the end. Yeah, those of you who need a little bit of extra help, we'll hang back and get folks individually set up in the Discord. A little bit of tech support moment. I know it's a new ecosystem for a lot of folks.
So here's our first collaborative group activity. We are going to try to do around this whole room group intros without anybody rocking the mic too, too hard. So let's say you have 20 to 30 seconds. your name and where you work or what you're affiliated with. And then I just want to know how you're showing up today and one thing that you hope to get out of this course. So I'm Chris, I'm the BC AI ecosystem executive director. I'm showing up pretty excited. Like I said, there's been a lot of energy going into this and here we are. And I'm just really excited to hear first order what people's, you know, like where the rubber hits the road for a lot of you guys.
Well, believe it or not, we only ran one and a half minutes over the allotted time so far.
And hopefully we can go into a little bit more of a self-reflective exercise from there. Andrew walked us through the sort of evolution and sophistication, right? Starting out with that video, we saw a growth and the development of increasingly sophisticated methods for performing data processing tasks from fundamental simple things around statistical methodology all the way through to deep learning methods.
And The starting point for that was essentially like statistical methods just for characterization of data sources, right? So we had the ability to compress and identify patterns and signals and hopefully discard out some of the noise, right? Essentially that was the starting point of that video. And what really got people excited down this learning trajectory and development of more and more sophisticated AI tooling?
The ability to capture some fundamental characteristic within data And then I describe it in a compressed or simple form. So like a model. From that, we saw the increasing capabilities to be able to do classification. So traditional machine learning models started to become really, really good at extrapolating those patterns to unseen data. And we had the advent of, let's say, classification methodologies in more and more sophisticated machine learning.
And that's essentially an extrapolation of that earlier primitive sort of ability. to characterize and throw away the noise. So compress and characterize, which gave us sort of, I would say what we refer to as sort of one of the verbs of classic machine learning. So classification. Subsequently, we reached like a finer level of precision. We had a whole category of models that allowed us to not only classify in terms of groups, but actually predict specific values.
So these are things that you might have encountered in your space, like prediction models, churn models, forecasting models that seek to narrow it down beyond just a set of classes, but actually into like a specific predictive value. And again, abstract away the noise and predict on future unseen examples. So this is the sort of baseline of machine learning that Andrew walked us through that sort of got us onto the trajectory that we are today.
So the ability to identify a pattern, distill it down, and then apply that as a classification for as of yet unseen and like new possible data points. With the advent of deep learning, we see something pretty profound change. One, models start to emerge that are at a level of complexity where we're no longer really able to keep track of how the machinery works. And from a responsible ad perspective, that's actually pretty profound.
So everything that I've discussed up to this point, we could audit it pretty precisely. You could say when the model was wrong, why was it wrong? When the model had a particular margin of error, we could explain that it was maybe a data lineage problem. We needed to source more data, data representativeness, et cetera. But with deep learning, which underpins a lot of what I think has driven the generative AI right now, we reach a level of complexity and a level of data hunger, let's say, for these models that we're no longer able to really, really diagnose how it is working.
Generative AI building on top of that fundamental deep learning sort of paradigm is truly and fundamentally a paradigm shift for several reasons. So building upon that chain of innovation We haven't changed anything in terms of our approach, but the complexity and the scale continue to sort of run ahead of that. And we introduce a new verb. So beyond just being able to characterize, and forecast we now have the ability to use algorithms to use machine learning and AI systems to generate, to generate something novel. And we can come back, I think we'll come back in week two.
scale has been a huge, huge part of that. So, Neural networks have existed since the 70s. There has not been a moment in time up until very, very recently where they've been actually viable for solving real world problems. And that's been a consequence of commoditization of hardware and innovation. that also requires fuel. So the massive hype that we're experiencing right now is not only a consequence of innovations in terms of the technology and the algorithms and the hardware, But it is also a cycle where that investment needs to be justified with the increasing prevalence of AI products ubiquitous in the technologies that you interact with. And we're going to come back to that topic.
Cool. So we've lost the ability to both examine and diagnose the inner workings of the model, but even the output now falls into a space that has a bit of ambiguity around the way that we evaluate it. For example, a poem that is output by the latest Claude model, could be an exceptional poem, it could be a very mediocre poem, or it could be a very poorly written poem.
And aside from like the ability to assess its grammatical correctness, the quality of that output is in the eye of the beholder. So The foundations, like all of the tools that we've used to go from simple descriptive methods all the way up to modern LLMs, we no longer have access to those tools when it comes to assessing the output of the systems that we've built. We don't have the ability to assess the accuracy of that poem.
We don't have the ability to assess the quality of that except against very primitive things. And that ambiguity is also super interesting for the conversation around responsible AI. Um, We're at a stage And the innovation or in the development of these methods where the tools that we have to assess their quality has been outpaced by the capabilities that they've developed. And that's both incredibly exciting, but also really, really concerning from a responsible AI perspective.
I'd like to leave this little piece of the course on the first pillar that is going to be a recurring topic in terms of looking at the system. One thing that I really got from that video, that Andrew video, regardless of the sophistication of the AI tool. And this applies to everything from a descriptive statistical methodology regression, a simple classifier, a deep learning model that's being used for any application all the way through to modern frontier LLMs.
data quality and the understanding of your data lineage is fundamental to being able to have a conversation about the responsibility of that. So what data was sourced? What are the implications for the performance? What measures have gone into ensuring that, studying that, monitoring that? And that is as a system designer, paramount to both the functionality and the responsibility. And it doesn't require engineering skills.
And so much in my life has changed in that time. So I'd love us to dive. We're going to actually go into little breakout groups and we're going to work with the question. How has generative AI unexpectedly changed something in your life? And I want you to consider things like maybe how you research or you write or you think through problems. Perhaps you have a habit or a skill or a daily ritual that has shifted.
Perhaps there's something that you used to do one way and now you do it completely differently. Or maybe you want to think about a relationship with information or creativity that just feels different now. So we're going to break into groups of two or three and each person's going to share. I want you to track your own time. We're going to have about eight minutes, eight to 10 minutes in the chat. So just make sure that you each get a chance to share. And I want you to be as you're.
And for me, the most surprising use of AI has been the personal. aspects of sort of facilitating my hobbies, like There's sometimes songs, melodies stuck in my head, but I don't know how to play any instruments. And now I can sort of use AI to make the melody that's stuck in my head. Or I joined the AI Film Club with DCAI. And this summer, I'm going to put out a video, which is something I never would have done if I had to learn how to create a video by myself, like a digital video.
But with AI, it feels more like whatever is in my head can just come out without me. having the practical skills to do it. or things like journaling in the Instead of just writing in my journal and then never looking at it again, I can say the thing to AI and get some feedback and have it back and forth. And what's surprising, I think, is with AI taking away all the things that I usually waste a lot of time on, I now have a lot of space in my life to focus on the things that I really care about.
One of the most surprising things that I found over the last year or so is that AI has allowed me to become more independent. And by that, I mean that I'm able to do things at work that I never considered myself capable of doing before. So I can produce a Excel spreadsheet with very complex formulas across multiple tabs that provides a phenomenal pivot table and summary and, you know, all of that kind of stuff. Excel is not my happy place in any way, shape or form.
But I can do that myself now by working with Claude to be able to prompt through that, which is phenomenal. It allows me to, you know, to not have to, you know, people's time frames for supporting or, you know, work quicker or more nimbly. But the kind of downstream consequence of that is that I find that there's less collaboration on things like that. Whereas before I'd have gone to other people adjacent to me and said, hey, could you help me with this? Hey, can you provide that particular activity or action or service? And can we collaborate on this? And so that loss of collaboration kind of concerns me a little bit.
that I'm still getting extra sets of eyes on the outputs of what I'm producing. So there's the upside of being able to do a lot of stuff that I never imagined myself capable of doing, but the downsides of being super watchful about that lack of bringing in other perspectives, right? And I think...
And I think it compounds with the fact that work sort of changed. A lot of us are independent people. Before we would have been a department inside of a bigger organization where we would have had other teams and collaborators. So already we're like isolated into these kind of independent work worlds or whatever. And now you get all these capabilities to do all these new things.
you couldn't have done yourself before. And those two things, I think, compound each other.
I don't think it is a great use of time to our first tenant of like making sure that the time that we spend in the room is beneficial to everyone. But I know that some folks who perhaps are interested in things like risk assessment, risk mitigation within their organizations might be interested to know some of the frameworks that are already in existence that are out there. So that is included in the notion.
We've got a good overview of the existing recognized risk management frameworks. and I won't go through all of them, but just more in general, the reason that there are so many, there's not a global standard as you would expect. Part of that is that they come from very different design philosophies. So like what sorts of risks they're meant to manage, who their target audiences are, what they're meant to accomplish.
So we have, for example, the UNESCO AI sort of Tactical AI guideline, which really comes from this primary orientation of human rights, like human dignity central to the sort of like the norms that should govern AI products. Obviously something developed at the scale of UNESCO, it's going to be very much the, not a compromise, but like a conversation between culturally distinct participants coming to something that's like universally accessible. But at the same time, it's not compelled upon anyone. It is a voluntary guideline.
Conversely, you have stuff like the European AI Act, which can be very, very prescriptive and comes from the perspective of gating market entry. So that's going to be something that is very prescriptive in terms of you have risk categories. If you would like to do business within Europe and with European citizens, you must comply with a very prescriptive set of guidelines around risk. AI applications that fall into the risk categories as described within that piece of legislation.
And then there's an operationalization framework for that. Um, I know that William, for example, and a couple other folks talked about how they work a little bit more on the, let's say, adversarial testing, like security compliance side. So not surprisingly, folks from that mindset also have come up with a framework. I'm a very big fan of the MITRE Atlas framework, which is also listed within the Notion page.
Um, NIST is universally accepted as probably the most comprehensive. It's usually where I would start with enterprise level clients because it does sort of cover the full gamut of that. It is widely, widely recognized as like a good starting point. But for for small, medium enterprises, it is very, very wieldy. So I believe the last time that I looked at NIST, it was 697 dimensions, each of which requires like an actual, an actual like piece of work.
So for comprehensive coverage, it's phenomenal for scale where you are looking at resource constraints. It is perhaps a little bit too much to bear. Again, I don't want to do a comprehensive walkthrough, but I just wanted to point you all to the fact that those frameworks are listed within our course content on the Notion. They're designed differently. I'm happy to revisit them throughout the course, but definitely take a look at those. And we've tried to link to a very, very lightweight high-level summary of each of them.
I know I'm excited to just have that Notion dashboard to be able to go back to when I'm looking for frameworks. It's such a great resource to have that there. And I'm happy that you all get to check them out. I'm going to just, yeah, go for it.
So a capstone project could be like if you're into vibe coding and you don't know what you want to vibe code yet and you want to explore this stuff, build a tool that compares and contrasts and sorts and explores the five different AI risk assessment frameworks and compares them or something. That'd be a great project. You'll get to know the frameworks, you'll get to know vibe coding, something like that.
I'm going to jump to chat about our artifact for the week. And that's our homework. And so the way that we're looking at the homework is we want to provide you with something that's actually really useful for you. So this is for your benefit. We're not checking your homework. This is, yeah, this is for you. And so the project this week, it's a personal AI inventory. you And so knowing what is already being used is the first step to using it responsibly.
So the instructions are to basically list all the AI tools and systems that you interact with, personal, at work, And, you know, this is something that you can do anytime. in a data, in a spreadsheet? Where do you use Notion? Do you use, you want to write it in your sketchbook? This is up to you. But to track, you know, the task, what you use it for, The input, you know, what you provide, the text, files, personal data, what it is, and then what's the output? What does it give back to you?
So these instructions are a, well, they're in the chat, but they're also notion under the toggle that says artifact. And, um, Yeah, to be mindful of the tools you use is to stay awake to their influence and to keep Your choices in your hands. I know I'm going to do this homework because I don't know what tools I use. There's so many, there's so many, and it's a great, a great first step.
It's also absolutely the foundational step to setting up any badass automations.
People always come to me and they're like, I want to set up agentic workflows and automations. It's like, all right, let's draw some boxes and arrows of all the AI tools you use and how they talk to each other. And then once we've got all that, then we can automate the whole system or whatever. So you got to understand what you got and how it plugs together if you want to do some automation.
All right. I think I've seen a couple questions I'm going to come back to that obviously as we promised in the chat. There are some really, really interesting dimensions that we didn't get a chance to go through. I promise that we will go through this, either have them on the notion or bring them up at the beginning of next session for discussion. Because I think that, again, some of these are like super, super interesting.
So as we're approaching the end here, I'd love to do just a little one word checkout. And we're going to do it as a waterfall. I don't know if anyone's... done that method before and it's called a chat waterfall. So I want you to think of one word in your head right now That is a checkout word for where you're landing or how you're feeling in this moment at the end of the course. Everybody think of their word.
And then we're going to go. We're going to do a countdown and we're all going to. press send at the same time. Okay. So three, two, Two, one. Send. Awesome. Oh, shoot.
That helps to get us all kind of on the same page here today. If you haven't already, try to watch some of those videos and do the readings because things are going to keep moving pretty fast through here. We've got four weeks together. It's intended to be kind of a focused, structured time. No one's going to wag their finger at you if you don't meet your own personal standards. But take this moment that you've set aside to go deep. And let's meet on the chat.
I will post the LinkedIn list later. I will post this video and I will post the transcript for you. And yeah, you can get us there if you have any questions.
where they developed a new legal category called deployers of AI systems, where you can no longer say I didn't develop it and I didn't create it and I didn't do any of these other things. But if you deploy it, you have legal, ethical culpability if things go wrong.
I'm also having issues with Notion. Notion keeps saying that I don't have permission, so I know Discord is first, but if there's time to look at Notion, I'd appreciate it.
Because the page that I just shared is a public page, so you shouldn't even need any access.
That was my understanding is that you that that the all of the content and notion is being served immediately to a public facing page.
Sorry. It just takes me a minute to navigate between these things. Shit, so there's the main page And From there, if you go here to... And there's all sorts of interesting stuff here, like the week by week stuff is here, like here's a week one thing. And in week one, you'll see the pre readings, including the videos with Andrew Ning and stuff and the content and the exercise from that week. Um... And then if you click on Go ahead.
If you, If you click on... Yeah.
Also, this is like a magic trick, but control A, you know, control C, all Tab into your browser, chat GPT.
What did we learn in class today? And then that's the whole Notion page right there. And it's going to go off and chew on that and get back to me. And so like, if you're feeling like I don't want to be dropping down through toggles to be looking for stuff, just drop the whole page into a thing and ask where those frameworks are.
Yeah, I think it might be my iPad switching between, trying to switch between. So on the Discord, it tells me the wrap thing is not, it just says there's a voice option, join voice. I click that, it says, you don't have permission. but the links open in the browser, but when I click on them to go to Discord or actually turns out Notion as well, it doesn't actually open. I'm wondering if I'm clicking on the wrong link.
and it just keeps showing me BCI rules about membership.
Because I think Cause I signed up for some, like another conversation in discord and it won't accept my info at. And so I just signed it up in Chrome and, I'm a Firefox girl, but I moved to Chrome, a different browser, and I signed up in a different email address and I think I got it. Oh, for whatever reason, it won't let you. I think it's an issue with the system. Like it's not recognized. Like you can't even reset, reset your passwords or anything.
I am going to sign off, but we will be available if you have any problems. We will email those links around again to everybody, to the Discord and the Notion. And yeah. Definitely.
Oh Uh... Actually, probably part of it is both of those tools, you're able to use them through the web browser, but you shouldn't use them through the web browser. You should download the Discord app and you should download the Notion app.