Tacoma City Council Study Session on AI Policy and Cushman Street Fair – August 27, 2024

I'd like to call to order the study session of August 27th, 2024.

Clark, will you please call the role?

Councilmember Bushnell.

President.

Councilmember Daniels.

Councilmember Diaz.

Deputy Mayor Hines.

Council Member Rumba.

Here.

Councilmember Sidalgay.

Here.

Council Member Scott.

Councilmember Walker.

Here.

Mayor Woodards.

Here.

Mayor and Council members.

This study session material is sent to you at 11.

All right.

Our first agenda item today is an update on the City of Tacoma's artificial intelligence policies.

And I'd like to call on information technology director Daniel Keith to begin the presentation.

Good afternoon, Mayor and Council.

I'm Daniel Key, IT Director for the City of Tacoma, and I'm here to give you an overview on artificial intelligence as a general technology and its use and policies at the City of Tacoma.

And so if we can go ahead and get our presentation started, thank you.

That's good, we're coming up.

So our agenda today, let's see if this is gonna work.

Next slide, please.

And then I'm on my own, I believe.

All right, a very brief history of artificial intelligence and the current uh excitement around generative AI, and we'll go as deep into the science and the math as you want to go, but I think we're gonna keep it relatively brief today.

Uh, general technology and governance strategy regarding generative AI specifically and AI broadly, adapting to the community's usage of the technology, and then ongoing policy and administrative considerations.

See where this is gonna work.

Next slide, please.

So a brief history in three points.

For about 40 or 50 years, uh, this math and the concepts behind the current wave of generative AI have existed in the academic context, and it's really met unmet expectations due to data volume, compute power, and general networked computers.

Really, this growth of the global internet and the power of computing behind it made the current revolution possible.

Many of us do remember Clippy and the annoyance of early attempts at this particular technology, and I think we've seen it continue to evolve over time.

Starting approximately 20 years ago, we had a next uh change in the technology around machine learning and deep neural networks, which is a really fancy sounding term for figuring out how certain parts of information are related to other parts of information and letting a computer tell you that.

And so we saw this in things like autocorrect, and we also remember how annoying that was when it started, and how it's still slightly uh annoying going forward.

Things like Siri, things like image recognition, object recognition, hey, is that a cat or the mailman at the door?

Um, those came out of machine learning and deep learning, figuring out what information means to people.

And then for the last two years, or approximately two years, the commercial availability of generative AI, which is all of that, and then one more layer, significant layer of technology and uh innovation beyond it.

And this is still not working, so next slide, please.

Sorry.

IT guys know how to turn remote on.

There we go.

Excellent.

Uh so we're really down in the depths of artificial intelligence with the current sets of tools.

There's a very broad academic field, there's a very specific technology around machine learning, there's an even more specific technique called deep learning, and now with generative AI, we're really having what's called transformer engines in idea.

And that word's gonna become important as we understand what the technology can and can't do.

It's very capable at understanding the form of information and taking that form of information and translating it into something else.

So any use case that really is about make this thing like that, or tell me something kind of like this, is very good at producing that form in a different context, and we see that as novel and innovative and creative.

So some uh mentioned a couple of these examples already.

Um, autocorrect, facial recognition.

If you ever logged in your phone by looking at it, that's enabled by deep neural networks and artificial intelligence.

The streaming recommendations, why is it so interesting to page through seven pages of things you might watch instead of actually watch something?

Because it's had a pretty good idea of what might be interesting to you, and those pictures are fascinating.

You could learn more.

Um fraud detection, you know.

If you've ever had your card questioned at the pump because you bought something that you didn't think you were gonna buy that week, the day before, and that was a little bit out of the ordinary.

Those sorts of patterns and systems self-learning those patterns is enabled by these deep rural networks and generative AI.