Speaker:
00:00:00
Hello, ladies.
Speaker:
00:00:01
Welcome to this week's episode of the hourly to exit podcast.
Speaker:
00:00:04
I have a very special guest today.
Speaker:
00:00:06
My law school classmate, joy Butler joy.
Speaker:
00:00:10
Welcome.
Speaker:
00:00:10
And thank you so much for joining us.
Speaker:
00:00:12
Thank you, Aaron.
Speaker:
00:00:13
I am honored to have been asked to be a guest.
Speaker:
00:00:17
Well, we're very excited to have you because AI could not be more
Speaker:
00:00:23
top of mind, for this audience.
Speaker:
00:00:25
And so as someone who has written extensively and spoken
Speaker:
00:00:29
about AI, I definitely wanted to have you on to, , go deep.
Speaker:
00:00:33
So before we get started, would you introduce yourself to the audience?
Speaker:
00:00:37
Sure.
Speaker:
00:00:37
so as you already shared, I am an attorney and in my law firm practice,
Speaker:
00:00:42
I provide product counsel services.
Speaker:
00:00:45
that essentially means I provide a combination of strategic and legal advice.
Speaker:
00:00:52
To companies that are, going into new lines of business or launching new
Speaker:
00:00:58
products new features of existing products or forming strategic partnerships.
Speaker:
00:01:04
And I come by that from, two areas of wall where I have a deep, in depth
Speaker:
00:01:11
knowledge, and that includes the technology side where I have worked on
Speaker:
00:01:18
and, help to structure probably literally, over 1000 contracts over the course
Speaker:
00:01:25
of my career for, all the contracts 1 would need when 1 is doing business.
Speaker:
00:01:30
Online and in digital technology, including end user license arrangements
Speaker:
00:01:36
and terms and conditions and the other prong of my in depth legal
Speaker:
00:01:41
knowledge concerns entertainment and copyright and this is where
Speaker:
00:01:46
you and I overlap quite a bit.
Speaker:
00:01:48
so I work on a lot of, creative content contracts also advise companies on.
Speaker:
00:01:56
Protecting their, copyrights and trademarks and, work with companies
Speaker:
00:02:02
that want to use, someone else's content, doing a lot of work
Speaker:
00:02:06
in the rights clearance area.
Speaker:
00:02:08
just to give your audience a little more of a flavor of the
Speaker:
00:02:12
types of projects I might work on, most of them are in the digital
Speaker:
00:02:16
technology and entertainment space.
Speaker:
00:02:18
for So a couple of projects include helping an entertainment social media
Speaker:
00:02:24
network launch, working with an commerce retail site that was incorporating a lot
Speaker:
00:02:32
of album cover work and original artwork.
Speaker:
00:02:37
another was, an ad supported, stock simulation game.
Speaker:
00:02:41
And here's something that may resonate with your audience, helping
Speaker:
00:02:46
a professional in the finance area take, this niche financial service
Speaker:
00:02:51
he was offering and, convert it into an online software as a service.
Speaker:
00:02:59
product.
Speaker:
00:03:01
so, that is me in a nutshell.
Speaker:
00:03:03
Awesome.
Speaker:
00:03:04
When did you first when I think even tell you the day I first heard about AI.
Speaker:
00:03:08
Where were you when you first heard about it?
Speaker:
00:03:10
What was the context and what were your initial thoughts?
Speaker:
00:03:14
I don't remember, the first time I heard about, JATGPT.
Speaker:
00:03:20
Right.
Speaker:
00:03:20
That may be what you're referring to.
Speaker:
00:03:22
Yeah.
Speaker:
00:03:22
but, actually within my practice, I have for quite some time been experimenting
Speaker:
00:03:29
with, trying to take some of my knowledge And, develop it into, digital tools,
Speaker:
00:03:36
making it more accessible to people.
Speaker:
00:03:39
as you know, I've written a couple of books on my areas of in depth knowledge.
Speaker:
00:03:44
So, one of the things I've been experimenting with is, taking some of that
Speaker:
00:03:49
knowledge and offering it in a digital format, one, experiment I believe I shared
Speaker:
00:03:54
with you was A contest and promotion tool, which asked a number of questions
Speaker:
00:04:01
and then gave you kind of a checklist.
Speaker:
00:04:04
of the legal questions you might ask before going forward with that.
Speaker:
00:04:09
and I've asked, actually used a tool, that a lot of, attorneys and,
Speaker:
00:04:15
well, it is a, Interview construction tool targeted to the legal space.
Speaker:
00:04:20
It's called Doc.
Speaker:
00:04:21
Assemble.
Speaker:
00:04:22
It's actually open source and spent a little bit of time.
Speaker:
00:04:26
tinkering around with that, is a long way to answer your question.
Speaker:
00:04:31
I was familiar with automation and artificial intelligence
Speaker:
00:04:35
through that process.
Speaker:
00:04:36
But when chat GPT came to my attention, that may have been around the same time
Speaker:
00:04:41
as it came to everyone else's attention.
Speaker:
00:04:43
I kept hearing about it and
Speaker:
00:04:46
right.
Speaker:
00:04:46
Right.
Speaker:
00:04:47
I guess I'd heard about it, but it was just noise to me kind
Speaker:
00:04:50
of like block train or crypto.
Speaker:
00:04:52
don't need to know that.
Speaker:
00:04:53
I don't want to know it, until finally I could no longer
Speaker:
00:04:56
ignore it, which was during.
Speaker:
00:04:58
Yeah.
Speaker:
00:04:58
And MCLE where I needed to get some credits, so I wasn't delinquent.
Speaker:
00:05:02
And so I'm listening to this one about AI and it's describing, they were talking
Speaker:
00:05:07
about chat, GBD in particular, and they're describing what you could do.
Speaker:
00:05:10
And they're having these samples and I'm like, what it can do.
Speaker:
00:05:13
What?
Speaker:
00:05:14
And so while I'm still in there, you know, it was just online.
Speaker:
00:05:17
I'm silly.
Speaker:
00:05:17
God forbid I go someplace in person.
Speaker:
00:05:18
and then I'm on my computer, like.
Speaker:
00:05:21
Doing stuff with it.
Speaker:
00:05:21
I'm like, Oh my God, this is bad.
Speaker:
00:05:24
And that was, well, it was February, 2023 and that was my initiation.
Speaker:
00:05:28
So what the last year has been, actually a fire hose of information
Speaker:
00:05:33
and changes in that time.
Speaker:
00:05:34
so I think chat GPT, it's the AOL of our times.
Speaker:
00:05:40
It's this technology that's been around for a while, but we finally have this
Speaker:
00:05:44
application that has made it a much more, accessible and user friendly
Speaker:
00:05:49
for a much wider group of people.
Speaker:
00:05:51
Yeah.
Speaker:
00:05:51
I mean, I guess, you know, when you think artificial intelligence has
Speaker:
00:05:54
been a while, I mean, obviously we've always had autocorrect and things
Speaker:
00:05:57
like that, or, all those things were artificial intelligence, right?
Speaker:
00:06:00
things like Alexa and Siri, right?
Speaker:
00:06:03
I mean, those Versions of it.
Speaker:
00:06:04
We just didn't think of it the way that we think of AI now.
Speaker:
00:06:08
Exactly.
Speaker:
00:06:09
It's been around for a while.
Speaker:
00:06:10
We just finally got a killer app in chat GPT.
Speaker:
00:06:15
Right.
Speaker:
00:06:15
Awesome.
Speaker:
00:06:16
So a lot of questions that I get are around, where's this data coming from?
Speaker:
00:06:22
what is the black box of, generative AI in particular we're talking about.
Speaker:
00:06:26
and what do I need to worry about?
Speaker:
00:06:29
are they taking my prompts and what are they doing with it?
Speaker:
00:06:33
client who is, utilizing signing an agreement to utilize the contract
Speaker:
00:06:39
review a I like, what are the issues regarding using 1 of those?
Speaker:
00:06:43
So everybody has questions about, What happens when I use AI and
Speaker:
00:06:48
what do I need to worry about?
Speaker:
00:06:49
And where does that, data come from and what is my exposure?
Speaker:
00:06:54
So I would just like to start from the top.
Speaker:
00:06:57
I think most of the audience is familiar, AI, but let's talk
Speaker:
00:07:01
about what training data is.
Speaker:
00:07:04
Like does it get its information from?
Speaker:
00:07:06
How does it get in there?
Speaker:
00:07:07
And, yeah, just start there with a general.
Speaker:
00:07:10
Yeah.
Speaker:
00:07:10
when we talk about a I models and some of the copyright and licensing
Speaker:
00:07:15
issues, there are kind of 2 categories.
Speaker:
00:07:19
category is the input.
Speaker:
00:07:21
And the 2nd category is the output.
Speaker:
00:07:24
When we're talking about generative, a, I, So when you mention training
Speaker:
00:07:28
material, you're talking about the first category of input.
Speaker:
00:07:32
And there has been a lot of controversy over whether or not, the training
Speaker:
00:07:37
material that is required to train these models, can be used without permission.
Speaker:
00:07:43
Because what the foundation models do, when I say foundation models, I mean
Speaker:
00:07:48
that Maybe eight or 10 models are around that, literally have millions of pieces
Speaker:
00:07:56
of content that they take into their kind of black box and, analyze it so that it
Speaker:
00:08:03
can be a general use large language model.
Speaker:
00:08:07
many of these models do is they source that data by getting data from
Speaker:
00:08:12
anywhere that they can, including, scraping the Internet for millions
Speaker:
00:08:17
and millions of pieces of data.
Speaker:
00:08:20
So, there's been, as I said, a lot of controversy around
Speaker:
00:08:23
whether or not permission is required for them to do that.
Speaker:
00:08:27
and many of these models are relying on now is an argument that, their use of that
Speaker:
00:08:34
material, as training material qualifies as a fair use to the Copyright Act.
Speaker:
00:08:41
I believe there number of, Areas a number of factors that will gradually
Speaker:
00:08:47
push these AI foundation models towards licensing that material.
Speaker:
00:08:53
1 of them is, is that there have been a number of lawsuits that have been filed
Speaker:
00:08:58
against them, charging them with copyright infringement and other related infractions
Speaker:
00:09:04
over their use of this material.
Speaker:
00:09:06
and.
Speaker:
00:09:07
A lot of those suits while all of those suits are still pending and they may
Speaker:
00:09:11
take a very long time to play out.
Speaker:
00:09:14
I think we're going to see progress towards more licensing prior to that.
Speaker:
00:09:19
And that's because people are very anxious.
Speaker:
00:09:22
to, use generative A.
Speaker:
00:09:24
I.
Speaker:
00:09:25
And, before they use that, though, they want some comfort level that
Speaker:
00:09:30
their use that material is not going to subject them to any type of a
Speaker:
00:09:35
copyright infringement or other claim.
Speaker:
00:09:38
So in order to make their customers, comfortable, With the fact that they
Speaker:
00:09:44
can use this material without taking on any legal liability, we are seeing more
Speaker:
00:09:49
and more of these AI companies gradually move towards licensing the content.
Speaker:
00:09:56
want to follow up on that before I'm going to step back just a second,
Speaker:
00:09:59
because you said large language models, and then we have machine
Speaker:
00:10:03
learning and we have generative AI.
Speaker:
00:10:05
Are those synonyms?
Speaker:
00:10:06
Or are they all different elements?
Speaker:
00:10:09
Transcribed Okay,
Speaker:
00:10:11
not the expert here, but I'll share with you my understanding.
Speaker:
00:10:15
So, the large language models, they are, the general models that can process.
Speaker:
00:10:23
the generated output, so that means they take all of the input,
Speaker:
00:10:29
all of that training material, and they basically analyze it to see
Speaker:
00:10:34
what the relationship of each data point is to this other data point.
Speaker:
00:10:39
So, when you ask it to produce something, it is, estimating or.
Speaker:
00:10:46
Putting forth, it's analysis of what word should come next or what
Speaker:
00:10:52
should come next in this particular graph, which is why it needs so much
Speaker:
00:10:57
training material from which to learn.
Speaker:
00:10:59
Got it.
Speaker:
00:11:00
Okay.
Speaker:
00:11:00
Now, you mentioned going back to where it's going towards licensing because users
Speaker:
00:11:06
of AI, I want to know that they're not going to get sued they use the output.
Speaker:
00:11:11
what does that mean for all of the current data that has been scraped
Speaker:
00:11:16
from the Internet and all these places?
Speaker:
00:11:18
previously, I mean, isn't the data sets.
Speaker:
00:11:21
and our use of AI as is almost like too big to fail.
Speaker:
00:11:26
what could happen with these lawsuits that are happening right now, if there
Speaker:
00:11:30
are billions of pieces of, let's say pirated information and say, the chat
Speaker:
00:11:36
GBT, open AI is training data set.
Speaker:
00:11:38
what could the possible remedy be if they lose?
Speaker:
00:11:43
Okay, so I do want to separate this into 2 categories again, because
Speaker:
00:11:48
when we talk about infringement, there are 2 separate questions.
Speaker:
00:11:51
The 1st question being whether or not just the process of.
Speaker:
00:11:56
Of the, a I companies, taking in data as training material and using
Speaker:
00:12:01
it to train their model, whether or not that's copyright infringement.
Speaker:
00:12:05
That's one question.
Speaker:
00:12:06
And then the second question is if you as a user of these models,
Speaker:
00:12:12
if you produce content and.
Speaker:
00:12:15
Use it to produce generated content.
Speaker:
00:12:17
Is there any legal liability for you?
Speaker:
00:12:19
Now, there are circumstances that could be imagined where, it's
Speaker:
00:12:25
possible for the, models, training data to be considered a fair use.
Speaker:
00:12:33
But maybe the way you've used it in creating output, is infringing
Speaker:
00:12:39
or violating in some way.
Speaker:
00:12:41
I'm not saying that scenario has actually come up or may come up
Speaker:
00:12:45
often, but 1 can imagine a set of circumstances where that might be true.
Speaker:
00:12:50
So, back to your original question, where is all this going?
Speaker:
00:12:53
What are the potential remedies?
Speaker:
00:12:55
well, 1 remedy with respect to these lawsuits is that they will settle with
Speaker:
00:13:02
a lot of these companies because the companies that have sued them have been
Speaker:
00:13:06
the largest companies with the most resources and very large organizations.
Speaker:
00:13:12
Like the author's guild, so they may settle, come to some agreement
Speaker:
00:13:18
on what a settlement fee should be.
Speaker:
00:13:21
And it's also possible that part of their settlement might be a
Speaker:
00:13:25
licensing agreement going forward.
Speaker:
00:13:27
that resolves matters for, the large organizations that have sued and.
Speaker:
00:13:33
The large private companies, if it's an organization or association, representing
Speaker:
00:13:37
much smaller players, it remains to be seen how much might flow to them
Speaker:
00:13:43
as part of any judicial settlement.
Speaker:
00:13:45
It may be that as opposed to a private settlement, we might get
Speaker:
00:13:50
some sort of a judicial settlement.
Speaker:
00:13:53
I think.
Speaker:
00:13:54
It's perhaps less likely, but it might be one of the outcomes and that
Speaker:
00:13:59
might be a settlement like something that was proposed in Google Books.
Speaker:
00:14:04
Now, the Google Books lawsuit, if anyone remembers his lawsuit from 2015.
Speaker:
00:14:10
This is the lawsuit that came out of Google Books starting its program
Speaker:
00:14:14
where it digitized millions of books and use them and still uses
Speaker:
00:14:18
them today to give a snippet of books in response to our search.
Speaker:
00:14:23
So that is one of the cases on which a lot of these AI model companies
Speaker:
00:14:27
rely when they argue that their use of the training material is a fair use.
Speaker:
00:14:33
For those who remember, the Google Books case initially, tried to resolve
Speaker:
00:14:39
itself, via a judicial settlement agreement that would have permitted
Speaker:
00:14:44
the snippets of those books and allowed the digitization, but that judicial
Speaker:
00:14:49
settlement, or the private settlement that was proposed, went to court.
Speaker:
00:14:53
Very much beyond just providing snippets, which is, one of the reasons
Speaker:
00:14:59
that it was ultimately, not approved by the court and kept going on and
Speaker:
00:15:05
ultimately said, okay, well, we're stripping out all this information.
Speaker:
00:15:09
you try to do in the settlement.
Speaker:
00:15:11
But, as consolation, we decided Google books that your use is a fair use.
Speaker:
00:15:16
So, it might be that some of the parties, try to move in that direction of some
Speaker:
00:15:21
type of a settlement that encompasses both small and larger players.
Speaker:
00:15:27
some of the other types of resolutions that have been thrown
Speaker:
00:15:31
out include kind of a collective.
Speaker:
00:15:34
that would be parallel to, the way we collect, public performance
Speaker:
00:15:39
royalties in the music industry.
Speaker:
00:15:41
So, for example, when a song is performed on the radio, all songwriters,
Speaker:
00:15:48
receive some income from their songs they've written being played.
Speaker:
00:15:51
Well.
Speaker:
00:15:52
the radio station is not going out and I'm entering into license agreements
Speaker:
00:15:56
with the millions of songwriters.
Speaker:
00:15:58
They have collected is in the case of music, as cap and BMI and couple others.
Speaker:
00:16:03
that, have these collective agreements where they issue blanket licenses.
Speaker:
00:16:08
So something like that has been proposed, potentially for,
Speaker:
00:16:13
the training material space.
Speaker:
00:16:14
So that brings in both, rights owners with very large catalogs and rights
Speaker:
00:16:19
owners with very small catalogs.
Speaker:
00:16:22
The copyright office had a comment period where it asked a bunch of industry
Speaker:
00:16:27
players what they thought of this and most of the people who commented were
Speaker:
00:16:31
very much in favor with, direct licensing, or perhaps even aggregated licensing
Speaker:
00:16:38
now that may be in part because that's where, larger companies are going to
Speaker:
00:16:43
get kind of the premium licensing.
Speaker:
00:16:46
Um, because the direct licenses we've seen, today have been between, you
Speaker:
00:16:50
AI model companies and very large organizations for millions of dollars.
Speaker:
00:16:55
just like the, what we're talking about in the collective licensing, example,
Speaker:
00:17:00
those very large companies are not going to enter into license agreements
Speaker:
00:17:05
with, millions of small players.
Speaker:
00:17:09
there'll be some balancing where they too can participate.
Speaker:
00:17:12
1 potential example of how this might be alleviated is through aggregators.
Speaker:
00:17:18
So, one aggregator we have right now is the Copyright Clearance Center,
Speaker:
00:17:22
which is aggregating, scientific papers for use in training material, and
Speaker:
00:17:28
that allows, smaller rights owners to participate in, having their material
Speaker:
00:17:33
and being paid for their material to be used as training material,
Speaker:
00:17:37
if that's what they choose to do.
Speaker:
00:17:39
An example in, this space I've seen, come forward as a startup is called Dappier,
Speaker:
00:17:46
and that is a startup that is, dedicated to getting those smaller, rights owners,
Speaker:
00:17:54
giving them the opportunity to participate in being a part of training material.
Speaker:
00:17:59
and making that training material more accessible to both the large, AI models,
Speaker:
00:18:07
and, you know, the smaller, AI companies that might have fewer resources and not
Speaker:
00:18:12
be as able to, compete when, you know, license, you know, is Agreements are going
Speaker:
00:18:16
for millions and millions of dollars.
Speaker:
00:18:18
Yeah.
Speaker:
00:18:18
Yeah.
Speaker:
00:18:18
I mean, it sounds like this would all have to be perspective.
Speaker:
00:18:21
I mean, if, the AI companies have been scraping the Internet for we
Speaker:
00:18:25
don't know how long and is it even able to distinguish 1 piece of data?
Speaker:
00:18:30
In the data set from another, I don't know, like, how would you compensate all
Speaker:
00:18:35
the information that's already in there.
Speaker:
00:18:37
and in order to, parcel out payments, whatever fraction of a penny that, I might
Speaker:
00:18:43
get for, something, and going forward.
Speaker:
00:18:46
If you are a small content creator, kind of your everyday content creator,
Speaker:
00:18:50
like the audience here, it would then be on you to make sure that
Speaker:
00:18:55
your content is registered somewhere.
Speaker:
00:18:57
So you'd be part of some aggregator that has a license
Speaker:
00:19:00
who is getting paid by the AI, AI
Speaker:
00:19:03
Okay.
Speaker:
00:19:04
So several issues in that question.
Speaker:
00:19:07
okay.
Speaker:
00:19:08
Let's go with, That first part where you talk about kind of the provenance,
Speaker:
00:19:12
what was the source of the data?
Speaker:
00:19:14
Is it even traceable?
Speaker:
00:19:16
And this is one of the pain points.
Speaker:
00:19:18
And this is also where that analysis about whether your output subjects
Speaker:
00:19:25
you to any type of liability.
Speaker:
00:19:28
so back up for a second.
Speaker:
00:19:30
if you are any type of content creator, and you are trying to determine
Speaker:
00:19:35
whether or not the content you've created, is violating any rights,
Speaker:
00:19:40
you need to know its source, right?
Speaker:
00:19:43
So, For the output that they have, if that provenance is not available,
Speaker:
00:19:49
you using the generative AI, you can't even do that analysis.
Speaker:
00:19:53
So that's part of the pressure on the AI model companies.
Speaker:
00:19:57
in not just waiting for these lawsuits to play out, but making their
Speaker:
00:20:02
potential customers comfortable that you can use our AI models and it can
Speaker:
00:20:07
produce output that you can then use.
Speaker:
00:20:11
And so part of having to do that is knowing the provenance.
Speaker:
00:20:16
Now, the extent to which they currently do that.
Speaker:
00:20:19
I don't know.
Speaker:
00:20:20
I model companies have often.
Speaker:
00:20:22
Been quite opaque and not very transparent about how the sausage is being made.
Speaker:
00:20:28
on the outside, though, like again as another example of where the industry
Speaker:
00:20:32
is going, there has cropped up another, kind of startup in this space called
Speaker:
00:20:38
barely trained, which is offering certification for AI model companies.
Speaker:
00:20:44
that have, produced their models relying solely on an authorized data set.
Speaker:
00:20:52
And then, you know, theory is, you are, a company that wants to leverage AI,
Speaker:
00:20:58
you can get more comfort in knowing that, you're relying on AI model, an
Speaker:
00:21:03
AI company that is fairly trained.
Speaker:
00:21:04
And the last time I checked, there were only a few dozen companies that had
Speaker:
00:21:08
that certification, but, that may grow.
Speaker:
00:21:11
So, maybe enterprise users would go for the fairly trained type,
Speaker:
00:21:15
because they're much more concerned, frankly, than most everyday users
Speaker:
00:21:20
about the quality of that output.
Speaker:
00:21:22
it seems like if they're using it, to create public facing
Speaker:
00:21:26
materials, they would want that fairly trained data set behind it.
Speaker:
00:21:31
they did, they also give reps and warranties when you go through them
Speaker:
00:21:33
regarding the quality of the output.
Speaker:
00:21:35
Do they give representations and warranties fairly trained provided
Speaker:
00:21:39
anyone who well, fairly trained or someone who has licensed their
Speaker:
00:21:44
data from fairly trained would they then in their terms of use have.
Speaker:
00:21:49
Represent fairly,
Speaker:
00:21:51
trained doesn't license data.
Speaker:
00:21:53
Fairly trained is a certification program.
Speaker:
00:21:56
So if an AI company, wants this certification to show everyone, that.
Speaker:
00:22:04
They have relied on an authorized data set, then this is a
Speaker:
00:22:07
certification that they can apply for.
Speaker:
00:22:10
Got it.
Speaker:
00:22:10
Okay.
Speaker:
00:22:11
because I believe that there are some platforms that do provide
Speaker:
00:22:15
indemnification, although they have a bunch of provisos, where's that going?
Speaker:
00:22:19
So that users feel more.
Speaker:
00:22:21
Comfort there, right?
Speaker:
00:22:22
So, I mean, I think that's part of their responding to, this pain point
Speaker:
00:22:27
of needing to make their customers more comfortable with using their product.
Speaker:
00:22:32
they are, providing certain indemnifications, it remains to be seen,
Speaker:
00:22:37
how effective those indemnifications would be if a customer were actually sued.
Speaker:
00:22:44
And as you mentioned, they do have, a lot of exclusions, personally, I think that is
Speaker:
00:22:51
just sort of an intermediate stop gap and they are going to be pushed more towards,
Speaker:
00:22:57
More licensing of their data sets.
Speaker:
00:23:01
and I would say, while we wait for this to play out, I mean, as you
Speaker:
00:23:05
know, this lawsuit could take and probably will take a very long time.
Speaker:
00:23:09
the Google books case, for example, on which the AI companies are
Speaker:
00:23:12
relying to 10 years before finally reaching, that conclusion that, Google
Speaker:
00:23:18
books digitization was a fair use.
Speaker:
00:23:21
So I would say in the interim, AI companies, and those.
Speaker:
00:23:26
producing AI models should look more to, using, authorized data sets or
Speaker:
00:23:33
construction of their models and authorized data sets with a traceable
Speaker:
00:23:38
provenance so that, their customers, when using the, output or wanting
Speaker:
00:23:44
to put the output into use, can.
Speaker:
00:23:47
Know what the source is and do that analysis of is this
Speaker:
00:23:51
violating any copyright?
Speaker:
00:23:52
Is this violating any right of publicity or trademark or anything else?
Speaker:
00:23:57
I would say for the companies that want to leverage a I, when you're looking for
Speaker:
00:24:01
partners, you do want to look at partners who are using authorized data sets.
Speaker:
00:24:08
Right now, what I see is that a lot of companies, brands.
Speaker:
00:24:13
companies in the film and television industry that are actually leveraging AI
Speaker:
00:24:19
and it is being leveraged, but they're using it for a first draft or a proof
Speaker:
00:24:26
of concept for things that are iterative and you'll need to be turned around
Speaker:
00:24:31
very quickly, but they're not using it.
Speaker:
00:24:34
As part of the final consumer facing output, just due to those copyright
Speaker:
00:24:42
reasons, both the reasons we just discussed, fear of having any type
Speaker:
00:24:47
of legal liability, but also, because there are limitations on the degree to
Speaker:
00:24:53
which you can protect, output that's generated by, artificial intelligence.
Speaker:
00:24:59
Right, so when they have a.
Speaker:
00:25:01
Authorized data set.
Speaker:
00:25:03
And does the output come with footnotes with What does that look like?
Speaker:
00:25:07
Do you know?
Speaker:
00:25:07
Have you seen, what that looks like
Speaker:
00:25:09
tell us what the sources are with it?
Speaker:
00:25:12
Like, does it identify?
Speaker:
00:25:13
Yeah.
Speaker:
00:25:13
Oh, oh, oh, I see.
Speaker:
00:25:15
When is it authorized?
Speaker:
00:25:15
It is right.
Speaker:
00:25:16
To my knowledge, it is not coming with anything.
Speaker:
00:25:21
And you're talking about the fairly trained component, right?
Speaker:
00:25:23
Yeah, is not coming with anything.
Speaker:
00:25:26
but that does need to be a path toward which we're traveling.
Speaker:
00:25:31
And there has been like a lot of conversation about that in this space that
Speaker:
00:25:36
it needs to be, you marked, needs to be traced in terms of, what was the source?
Speaker:
00:25:42
What did you rely on to do that?
Speaker:
00:25:46
Yeah.
Speaker:
00:25:46
As far as, the magic that happens inside of generative AI
Speaker:
00:25:50
platform, do we know what that is?
Speaker:
00:25:53
Or is that kind of the trade secrets of each companies?
Speaker:
00:25:56
Or is there general technology that Makes the magic happen?
Speaker:
00:26:00
I am not the expert on the technology inside of the AI models.
Speaker:
00:26:06
I can share what I know.
Speaker:
00:26:08
In part, it does depend on the approach that they've used, whether it's supervised
Speaker:
00:26:14
learning or unsupervised learning.
Speaker:
00:26:16
Which to make it very simple depends on how much you assisted the machine,
Speaker:
00:26:21
like, did you mark things and tell them, this is a dog and this is a cat
Speaker:
00:26:26
or did you just give them like kind of millions of pictures and kind of let
Speaker:
00:26:29
them figure it out when you let them figure it out when it's unsupervised,
Speaker:
00:26:35
it is more of a black box in terms of how they got to that answer,
Speaker:
00:26:40
which brings up all sorts of other.
Speaker:
00:26:43
societal issue, right?
Speaker:
00:26:45
I think it's a time to play it about
Speaker:
00:26:47
a, uh, interesting.
Speaker:
00:26:48
can we wrap up with some best practices just for your everyday kind of chat, GBT,
Speaker:
00:26:53
Janet, what is the Google on Genesis?
Speaker:
00:26:55
What does it, user like when they're using it, for this audience, the expertise
Speaker:
00:26:59
based business, maybe they're using it to create first drafts or to help them with
Speaker:
00:27:03
social media posts or something like that, like just some general best practices.
Speaker:
00:27:08
Sure, you want to be, circumspect about any, confidential or proprietary
Speaker:
00:27:15
information you include in a prompt, may want to anonymize it.
Speaker:
00:27:21
you need to keep in mind that, whatever output you get from, the
Speaker:
00:27:26
AI model, may not be eligible for copyright protection if this is, um.
Speaker:
00:27:31
something, material or output that you are passing on to a client or to a
Speaker:
00:27:38
customer, you may need to disclose that use, and you have to make sure that,
Speaker:
00:27:45
you're using the output Depending on the extent to which you're using it, are
Speaker:
00:27:51
you using it just for, a little bit of assistance in, modifying a few sentences?
Speaker:
00:27:57
Or are you actually producing images with it or producing an entire report with it?
Speaker:
00:28:02
You gonna want to make sure that you're procedures for like using generative
Speaker:
00:28:07
AI are consistent with the, contract that you have with your customer.
Speaker:
00:28:12
If you want to know whether or not, your material, your prompts are being
Speaker:
00:28:17
incorporated into the training data and being used to further train that AI model,
Speaker:
00:28:25
take a look at the terms and conditions.
Speaker:
00:28:28
to give you an example for chat, if you're using the free model,
Speaker:
00:28:34
and it is, recording your history.
Speaker:
00:28:36
Of your prompts, then, the prompts that you put in.
Speaker:
00:28:40
There are subject to being included as part of future training data.
Speaker:
00:28:45
Yeah.
Speaker:
00:28:46
So, if it along the left hand side there, you have scroll
Speaker:
00:28:49
through and see all your graphs.
Speaker:
00:28:51
I have.
Speaker:
00:28:51
that means it is going into the training data.
Speaker:
00:28:54
That is excellent.
Speaker:
00:28:55
can't say with certainty that it is going into the training data.
Speaker:
00:28:58
But I would say, it is susceptible to being used.
Speaker:
00:29:02
It's like they have not provided you, CHAT2P has not provided
Speaker:
00:29:06
you any representation that they will not use it for training.
Speaker:
00:29:09
Right.
Speaker:
00:29:10
Very good.
Speaker:
00:29:10
Thank you for making that distinction.
Speaker:
00:29:12
thank you for this.
Speaker:
00:29:13
this podcast is to help create a society that, and an economy
Speaker:
00:29:18
that works for more of us.
Speaker:
00:29:20
So I love to ask my guests, if there is an organization or a person who is doing the
Speaker:
00:29:26
good and hard work to help make an economy that works for more of us, is there
Speaker:
00:29:31
one that you'd like to share with us?
Speaker:
00:29:32
Sure, I really like organizations whose mission it is to
Speaker:
00:29:37
bridge the digital divide.
Speaker:
00:29:39
And one of my favorite is girls who code that has as part of its
Speaker:
00:29:44
mission, introducing more women.
Speaker:
00:29:47
into the technology field, and that's very apropos to our conversation
Speaker:
00:29:51
today, because as part of making AI, you beneficial for all humankind, we
Speaker:
00:29:58
really do need, a diverse perspective.
Speaker:
00:30:02
Yeah, I mean, we know that just when we talk about the Trina dating sets, like
Speaker:
00:30:06
what data is going in there, obviously.
Speaker:
00:30:08
what the output is only as diverse as the input, right.
Speaker:
00:30:11
And how it's being trained.
Speaker:
00:30:13
And I know that has come up in a number of, controversial ways as well, but
Speaker:
00:30:17
whether something's leaning this way or that way, but we definitely want to make
Speaker:
00:30:20
sure everyone has a voice in the future.
Speaker:
00:30:23
Thank you for that one.
Speaker:
00:30:24
And we will put that in the show notes along with how people can reach you.
Speaker:
00:30:29
where do you hang out, Joy?
Speaker:
00:30:30
And how can people get in touch with you to find out more?
Speaker:
00:30:32
Sure, so I'm always, , happy to, chat with, people doing innovative
Speaker:
00:30:38
things with technology, especially in the digital technology, online and
Speaker:
00:30:43
entertainment space, so they can find me through my website, which is www.
Speaker:
00:30:47
joybutler.
Speaker:
00:30:49
com.
Speaker:
00:30:49
And I'm awesome.
Speaker:
00:30:50
So, on LinkedIn, I have Joy Butler.
Speaker:
00:30:53
Awesome.
Speaker:
00:30:53
Well, thank you so much.
Speaker:
00:30:55
And, yes, everyone, please, follow Joy and, let us know if you have
Speaker:
00:30:59
any other questions about AI.
Speaker:
00:31:01
I know it's constantly evolving.
Speaker:
00:31:03
There's always going to be something new and we can continue
Speaker:
00:31:06
this conversation in the future.
Speaker:
00:31:08
Thanks again, Joy.
Speaker:
00:31:09
Thank you.
00:00:00
Hello, ladies.
Speaker:
00:00:01
Welcome to this week's episode of the hourly to exit podcast.
Speaker:
00:00:04
I have a very special guest today.
Speaker:
00:00:06
My law school classmate, joy Butler joy.
Speaker:
00:00:10
Welcome.
Speaker:
00:00:10
And thank you so much for joining us.
Speaker:
00:00:12
Thank you, Aaron.
Speaker:
00:00:13
I am honored to have been asked to be a guest.
Speaker:
00:00:17
Well, we're very excited to have you because AI could not be more
Speaker:
00:00:23
top of mind, for this audience.
Speaker:
00:00:25
And so as someone who has written extensively and spoken
Speaker:
00:00:29
about AI, I definitely wanted to have you on to, , go deep.
Speaker:
00:00:33
So before we get started, would you introduce yourself to the audience?
Speaker:
00:00:37
Sure.
Speaker:
00:00:37
so as you already shared, I am an attorney and in my law firm practice,
Speaker:
00:00:42
I provide product counsel services.
Speaker:
00:00:45
that essentially means I provide a combination of strategic and legal advice.
Speaker:
00:00:52
To companies that are, going into new lines of business or launching new
Speaker:
00:00:58
products new features of existing products or forming strategic partnerships.
Speaker:
00:01:04
And I come by that from, two areas of wall where I have a deep, in depth
Speaker:
00:01:11
knowledge, and that includes the technology side where I have worked on
Speaker:
00:01:18
and, help to structure probably literally, over 1000 contracts over the course
Speaker:
00:01:25
of my career for, all the contracts 1 would need when 1 is doing business.
Speaker:
00:01:30
Online and in digital technology, including end user license arrangements
Speaker:
00:01:36
and terms and conditions and the other prong of my in depth legal
Speaker:
00:01:41
knowledge concerns entertainment and copyright and this is where
Speaker:
00:01:46
you and I overlap quite a bit.
Speaker:
00:01:48
so I work on a lot of, creative content contracts also advise companies on.
Speaker:
00:01:56
Protecting their, copyrights and trademarks and, work with companies
Speaker:
00:02:02
that want to use, someone else's content, doing a lot of work
Speaker:
00:02:06
in the rights clearance area.
Speaker:
00:02:08
just to give your audience a little more of a flavor of the
Speaker:
00:02:12
types of projects I might work on, most of them are in the digital
Speaker:
00:02:16
technology and entertainment space.
Speaker:
00:02:18
for So a couple of projects include helping an entertainment social media
Speaker:
00:02:24
network launch, working with an commerce retail site that was incorporating a lot
Speaker:
00:02:32
of album cover work and original artwork.
Speaker:
00:02:37
another was, an ad supported, stock simulation game.
Speaker:
00:02:41
And here's something that may resonate with your audience, helping
Speaker:
00:02:46
a professional in the finance area take, this niche financial service
Speaker:
00:02:51
he was offering and, convert it into an online software as a service.
Speaker:
00:02:59
product.
Speaker:
00:03:01
so, that is me in a nutshell.
Speaker:
00:03:03
Awesome.
Speaker:
00:03:04
When did you first when I think even tell you the day I first heard about AI.
Speaker:
00:03:08
Where were you when you first heard about it?
Speaker:
00:03:10
What was the context and what were your initial thoughts?
Speaker:
00:03:14
I don't remember, the first time I heard about, JATGPT.
Speaker:
00:03:20
Right.
Speaker:
00:03:20
That may be what you're referring to.
Speaker:
00:03:22
Yeah.
Speaker:
00:03:22
but, actually within my practice, I have for quite some time been experimenting
Speaker:
00:03:29
with, trying to take some of my knowledge And, develop it into, digital tools,
Speaker:
00:03:36
making it more accessible to people.
Speaker:
00:03:39
as you know, I've written a couple of books on my areas of in depth knowledge.
Speaker:
00:03:44
So, one of the things I've been experimenting with is, taking some of that
Speaker:
00:03:49
knowledge and offering it in a digital format, one, experiment I believe I shared
Speaker:
00:03:54
with you was A contest and promotion tool, which asked a number of questions
Speaker:
00:04:01
and then gave you kind of a checklist.
Speaker:
00:04:04
of the legal questions you might ask before going forward with that.
Speaker:
00:04:09
and I've asked, actually used a tool, that a lot of, attorneys and,
Speaker:
00:04:15
well, it is a, Interview construction tool targeted to the legal space.
Speaker:
00:04:20
It's called Doc.
Speaker:
00:04:21
Assemble.
Speaker:
00:04:22
It's actually open source and spent a little bit of time.
Speaker:
00:04:26
tinkering around with that, is a long way to answer your question.
Speaker:
00:04:31
I was familiar with automation and artificial intelligence
Speaker:
00:04:35
through that process.
Speaker:
00:04:36
But when chat GPT came to my attention, that may have been around the same time
Speaker:
00:04:41
as it came to everyone else's attention.
Speaker:
00:04:43
I kept hearing about it and
Speaker:
00:04:46
right.
Speaker:
00:04:46
Right.
Speaker:
00:04:47
I guess I'd heard about it, but it was just noise to me kind
Speaker:
00:04:50
of like block train or crypto.
Speaker:
00:04:52
don't need to know that.
Speaker:
00:04:53
I don't want to know it, until finally I could no longer
Speaker:
00:04:56
ignore it, which was during.
Speaker:
00:04:58
Yeah.
Speaker:
00:04:58
And MCLE where I needed to get some credits, so I wasn't delinquent.
Speaker:
00:05:02
And so I'm listening to this one about AI and it's describing, they were talking
Speaker:
00:05:07
about chat, GBD in particular, and they're describing what you could do.
Speaker:
00:05:10
And they're having these samples and I'm like, what it can do.
Speaker:
00:05:13
What?
Speaker:
00:05:14
And so while I'm still in there, you know, it was just online.
Speaker:
00:05:17
I'm silly.
Speaker:
00:05:17
God forbid I go someplace in person.
Speaker:
00:05:18
and then I'm on my computer, like.
Speaker:
00:05:21
Doing stuff with it.
Speaker:
00:05:21
I'm like, Oh my God, this is bad.
Speaker:
00:05:24
And that was, well, it was February, 2023 and that was my initiation.
Speaker:
00:05:28
So what the last year has been, actually a fire hose of information
Speaker:
00:05:33
and changes in that time.
Speaker:
00:05:34
so I think chat GPT, it's the AOL of our times.
Speaker:
00:05:40
It's this technology that's been around for a while, but we finally have this
Speaker:
00:05:44
application that has made it a much more, accessible and user friendly
Speaker:
00:05:49
for a much wider group of people.
Speaker:
00:05:51
Yeah.
Speaker:
00:05:51
I mean, I guess, you know, when you think artificial intelligence has
Speaker:
00:05:54
been a while, I mean, obviously we've always had autocorrect and things
Speaker:
00:05:57
like that, or, all those things were artificial intelligence, right?
Speaker:
00:06:00
things like Alexa and Siri, right?
Speaker:
00:06:03
I mean, those Versions of it.
Speaker:
00:06:04
We just didn't think of it the way that we think of AI now.
Speaker:
00:06:08
Exactly.
Speaker:
00:06:09
It's been around for a while.
Speaker:
00:06:10
We just finally got a killer app in chat GPT.
Speaker:
00:06:15
Right.
Speaker:
00:06:15
Awesome.
Speaker:
00:06:16
So a lot of questions that I get are around, where's this data coming from?
Speaker:
00:06:22
what is the black box of, generative AI in particular we're talking about.
Speaker:
00:06:26
and what do I need to worry about?
Speaker:
00:06:29
are they taking my prompts and what are they doing with it?
Speaker:
00:06:33
client who is, utilizing signing an agreement to utilize the contract
Speaker:
00:06:39
review a I like, what are the issues regarding using 1 of those?
Speaker:
00:06:43
So everybody has questions about, What happens when I use AI and
Speaker:
00:06:48
what do I need to worry about?
Speaker:
00:06:49
And where does that, data come from and what is my exposure?
Speaker:
00:06:54
So I would just like to start from the top.
Speaker:
00:06:57
I think most of the audience is familiar, AI, but let's talk
Speaker:
00:07:01
about what training data is.
Speaker:
00:07:04
Like does it get its information from?
Speaker:
00:07:06
How does it get in there?
Speaker:
00:07:07
And, yeah, just start there with a general.
Speaker:
00:07:10
Yeah.
Speaker:
00:07:10
when we talk about a I models and some of the copyright and licensing
Speaker:
00:07:15
issues, there are kind of 2 categories.
Speaker:
00:07:19
category is the input.
Speaker:
00:07:21
And the 2nd category is the output.
Speaker:
00:07:24
When we're talking about generative, a, I, So when you mention training
Speaker:
00:07:28
material, you're talking about the first category of input.
Speaker:
00:07:32
And there has been a lot of controversy over whether or not, the training
Speaker:
00:07:37
material that is required to train these models, can be used without permission.
Speaker:
00:07:43
Because what the foundation models do, when I say foundation models, I mean
Speaker:
00:07:48
that Maybe eight or 10 models are around that, literally have millions of pieces
Speaker:
00:07:56
of content that they take into their kind of black box and, analyze it so that it
Speaker:
00:08:03
can be a general use large language model.
Speaker:
00:08:07
many of these models do is they source that data by getting data from
Speaker:
00:08:12
anywhere that they can, including, scraping the Internet for millions
Speaker:
00:08:17
and millions of pieces of data.
Speaker:
00:08:20
So, there's been, as I said, a lot of controversy around
Speaker:
00:08:23
whether or not permission is required for them to do that.
Speaker:
00:08:27
and many of these models are relying on now is an argument that, their use of that
Speaker:
00:08:34
material, as training material qualifies as a fair use to the Copyright Act.
Speaker:
00:08:41
I believe there number of, Areas a number of factors that will gradually
Speaker:
00:08:47
push these AI foundation models towards licensing that material.
Speaker:
00:08:53
1 of them is, is that there have been a number of lawsuits that have been filed
Speaker:
00:08:58
against them, charging them with copyright infringement and other related infractions
Speaker:
00:09:04
over their use of this material.
Speaker:
00:09:06
and.
Speaker:
00:09:07
A lot of those suits while all of those suits are still pending and they may
Speaker:
00:09:11
take a very long time to play out.
Speaker:
00:09:14
I think we're going to see progress towards more licensing prior to that.
Speaker:
00:09:19
And that's because people are very anxious.
Speaker:
00:09:22
to, use generative A.
Speaker:
00:09:24
I.
Speaker:
00:09:25
And, before they use that, though, they want some comfort level that
Speaker:
00:09:30
their use that material is not going to subject them to any type of a
Speaker:
00:09:35
copyright infringement or other claim.
Speaker:
00:09:38
So in order to make their customers, comfortable, With the fact that they
Speaker:
00:09:44
can use this material without taking on any legal liability, we are seeing more
Speaker:
00:09:49
and more of these AI companies gradually move towards licensing the content.
Speaker:
00:09:56
want to follow up on that before I'm going to step back just a second,
Speaker:
00:09:59
because you said large language models, and then we have machine
Speaker:
00:10:03
learning and we have generative AI.
Speaker:
00:10:05
Are those synonyms?
Speaker:
00:10:06
Or are they all different elements?
Speaker:
00:10:09
Transcribed Okay,
Speaker:
00:10:11
not the expert here, but I'll share with you my understanding.
Speaker:
00:10:15
So, the large language models, they are, the general models that can process.
Speaker:
00:10:23
the generated output, so that means they take all of the input,
Speaker:
00:10:29
all of that training material, and they basically analyze it to see
Speaker:
00:10:34
what the relationship of each data point is to this other data point.
Speaker:
00:10:39
So, when you ask it to produce something, it is, estimating or.
Speaker:
00:10:46
Putting forth, it's analysis of what word should come next or what
Speaker:
00:10:52
should come next in this particular graph, which is why it needs so much
Speaker:
00:10:57
training material from which to learn.
Speaker:
00:10:59
Got it.
Speaker:
00:11:00
Okay.
Speaker:
00:11:00
Now, you mentioned going back to where it's going towards licensing because users
Speaker:
00:11:06
of AI, I want to know that they're not going to get sued they use the output.
Speaker:
00:11:11
what does that mean for all of the current data that has been scraped
Speaker:
00:11:16
from the Internet and all these places?
Speaker:
00:11:18
previously, I mean, isn't the data sets.
Speaker:
00:11:21
and our use of AI as is almost like too big to fail.
Speaker:
00:11:26
what could happen with these lawsuits that are happening right now, if there
Speaker:
00:11:30
are billions of pieces of, let's say pirated information and say, the chat
Speaker:
00:11:36
GBT, open AI is training data set.
Speaker:
00:11:38
what could the possible remedy be if they lose?
Speaker:
00:11:43
Okay, so I do want to separate this into 2 categories again, because
Speaker:
00:11:48
when we talk about infringement, there are 2 separate questions.
Speaker:
00:11:51
The 1st question being whether or not just the process of.
Speaker:
00:11:56
Of the, a I companies, taking in data as training material and using
Speaker:
00:12:01
it to train their model, whether or not that's copyright infringement.
Speaker:
00:12:05
That's one question.
Speaker:
00:12:06
And then the second question is if you as a user of these models,
Speaker:
00:12:12
if you produce content and.
Speaker:
00:12:15
Use it to produce generated content.
Speaker:
00:12:17
Is there any legal liability for you?
Speaker:
00:12:19
Now, there are circumstances that could be imagined where, it's
Speaker:
00:12:25
possible for the, models, training data to be considered a fair use.
Speaker:
00:12:33
But maybe the way you've used it in creating output, is infringing
Speaker:
00:12:39
or violating in some way.
Speaker:
00:12:41
I'm not saying that scenario has actually come up or may come up
Speaker:
00:12:45
often, but 1 can imagine a set of circumstances where that might be true.
Speaker:
00:12:50
So, back to your original question, where is all this going?
Speaker:
00:12:53
What are the potential remedies?
Speaker:
00:12:55
well, 1 remedy with respect to these lawsuits is that they will settle with
Speaker:
00:13:02
a lot of these companies because the companies that have sued them have been
Speaker:
00:13:06
the largest companies with the most resources and very large organizations.
Speaker:
00:13:12
Like the author's guild, so they may settle, come to some agreement
Speaker:
00:13:18
on what a settlement fee should be.
Speaker:
00:13:21
And it's also possible that part of their settlement might be a
Speaker:
00:13:25
licensing agreement going forward.
Speaker:
00:13:27
that resolves matters for, the large organizations that have sued and.
Speaker:
00:13:33
The large private companies, if it's an organization or association, representing
Speaker:
00:13:37
much smaller players, it remains to be seen how much might flow to them
Speaker:
00:13:43
as part of any judicial settlement.
Speaker:
00:13:45
It may be that as opposed to a private settlement, we might get
Speaker:
00:13:50
some sort of a judicial settlement.
Speaker:
00:13:53
I think.
Speaker:
00:13:54
It's perhaps less likely, but it might be one of the outcomes and that
Speaker:
00:13:59
might be a settlement like something that was proposed in Google Books.
Speaker:
00:14:04
Now, the Google Books lawsuit, if anyone remembers his lawsuit from 2015.
Speaker:
00:14:10
This is the lawsuit that came out of Google Books starting its program
Speaker:
00:14:14
where it digitized millions of books and use them and still uses
Speaker:
00:14:18
them today to give a snippet of books in response to our search.
Speaker:
00:14:23
So that is one of the cases on which a lot of these AI model companies
Speaker:
00:14:27
rely when they argue that their use of the training material is a fair use.
Speaker:
00:14:33
For those who remember, the Google Books case initially, tried to resolve
Speaker:
00:14:39
itself, via a judicial settlement agreement that would have permitted
Speaker:
00:14:44
the snippets of those books and allowed the digitization, but that judicial
Speaker:
00:14:49
settlement, or the private settlement that was proposed, went to court.
Speaker:
00:14:53
Very much beyond just providing snippets, which is, one of the reasons
Speaker:
00:14:59
that it was ultimately, not approved by the court and kept going on and
Speaker:
00:15:05
ultimately said, okay, well, we're stripping out all this information.
Speaker:
00:15:09
you try to do in the settlement.
Speaker:
00:15:11
But, as consolation, we decided Google books that your use is a fair use.
Speaker:
00:15:16
So, it might be that some of the parties, try to move in that direction of some
Speaker:
00:15:21
type of a settlement that encompasses both small and larger players.
Speaker:
00:15:27
some of the other types of resolutions that have been thrown
Speaker:
00:15:31
out include kind of a collective.
Speaker:
00:15:34
that would be parallel to, the way we collect, public performance
Speaker:
00:15:39
royalties in the music industry.
Speaker:
00:15:41
So, for example, when a song is performed on the radio, all songwriters,
Speaker:
00:15:48
receive some income from their songs they've written being played.
Speaker:
00:15:51
Well.
Speaker:
00:15:52
the radio station is not going out and I'm entering into license agreements
Speaker:
00:15:56
with the millions of songwriters.
Speaker:
00:15:58
They have collected is in the case of music, as cap and BMI and couple others.
Speaker:
00:16:03
that, have these collective agreements where they issue blanket licenses.
Speaker:
00:16:08
So something like that has been proposed, potentially for,
Speaker:
00:16:13
the training material space.
Speaker:
00:16:14
So that brings in both, rights owners with very large catalogs and rights
Speaker:
00:16:19
owners with very small catalogs.
Speaker:
00:16:22
The copyright office had a comment period where it asked a bunch of industry
Speaker:
00:16:27
players what they thought of this and most of the people who commented were
Speaker:
00:16:31
very much in favor with, direct licensing, or perhaps even aggregated licensing
Speaker:
00:16:38
now that may be in part because that's where, larger companies are going to
Speaker:
00:16:43
get kind of the premium licensing.
Speaker:
00:16:46
Um, because the direct licenses we've seen, today have been between, you
Speaker:
00:16:50
AI model companies and very large organizations for millions of dollars.
Speaker:
00:16:55
just like the, what we're talking about in the collective licensing, example,
Speaker:
00:17:00
those very large companies are not going to enter into license agreements
Speaker:
00:17:05
with, millions of small players.
Speaker:
00:17:09
there'll be some balancing where they too can participate.
Speaker:
00:17:12
1 potential example of how this might be alleviated is through aggregators.
Speaker:
00:17:18
So, one aggregator we have right now is the Copyright Clearance Center,
Speaker:
00:17:22
which is aggregating, scientific papers for use in training material, and
Speaker:
00:17:28
that allows, smaller rights owners to participate in, having their material
Speaker:
00:17:33
and being paid for their material to be used as training material,
Speaker:
00:17:37
if that's what they choose to do.
Speaker:
00:17:39
An example in, this space I've seen, come forward as a startup is called Dappier,
Speaker:
00:17:46
and that is a startup that is, dedicated to getting those smaller, rights owners,
Speaker:
00:17:54
giving them the opportunity to participate in being a part of training material.
Speaker:
00:17:59
and making that training material more accessible to both the large, AI models,
Speaker:
00:18:07
and, you know, the smaller, AI companies that might have fewer resources and not
Speaker:
00:18:12
be as able to, compete when, you know, license, you know, is Agreements are going
Speaker:
00:18:16
for millions and millions of dollars.
Speaker:
00:18:18
Yeah.
Speaker:
00:18:18
Yeah.
Speaker:
00:18:18
I mean, it sounds like this would all have to be perspective.
Speaker:
00:18:21
I mean, if, the AI companies have been scraping the Internet for we
Speaker:
00:18:25
don't know how long and is it even able to distinguish 1 piece of data?
Speaker:
00:18:30
In the data set from another, I don't know, like, how would you compensate all
Speaker:
00:18:35
the information that's already in there.
Speaker:
00:18:37
and in order to, parcel out payments, whatever fraction of a penny that, I might
Speaker:
00:18:43
get for, something, and going forward.
Speaker:
00:18:46
If you are a small content creator, kind of your everyday content creator,
Speaker:
00:18:50
like the audience here, it would then be on you to make sure that
Speaker:
00:18:55
your content is registered somewhere.
Speaker:
00:18:57
So you'd be part of some aggregator that has a license
Speaker:
00:19:00
who is getting paid by the AI, AI
Speaker:
00:19:03
Okay.
Speaker:
00:19:04
So several issues in that question.
Speaker:
00:19:07
okay.
Speaker:
00:19:08
Let's go with, That first part where you talk about kind of the provenance,
Speaker:
00:19:12
what was the source of the data?
Speaker:
00:19:14
Is it even traceable?
Speaker:
00:19:16
And this is one of the pain points.
Speaker:
00:19:18
And this is also where that analysis about whether your output subjects
Speaker:
00:19:25
you to any type of liability.
Speaker:
00:19:28
so back up for a second.
Speaker:
00:19:30
if you are any type of content creator, and you are trying to determine
Speaker:
00:19:35
whether or not the content you've created, is violating any rights,
Speaker:
00:19:40
you need to know its source, right?
Speaker:
00:19:43
So, For the output that they have, if that provenance is not available,
Speaker:
00:19:49
you using the generative AI, you can't even do that analysis.
Speaker:
00:19:53
So that's part of the pressure on the AI model companies.
Speaker:
00:19:57
in not just waiting for these lawsuits to play out, but making their
Speaker:
00:20:02
potential customers comfortable that you can use our AI models and it can
Speaker:
00:20:07
produce output that you can then use.
Speaker:
00:20:11
And so part of having to do that is knowing the provenance.
Speaker:
00:20:16
Now, the extent to which they currently do that.
Speaker:
00:20:19
I don't know.
Speaker:
00:20:20
I model companies have often.
Speaker:
00:20:22
Been quite opaque and not very transparent about how the sausage is being made.
Speaker:
00:20:28
on the outside, though, like again as another example of where the industry
Speaker:
00:20:32
is going, there has cropped up another, kind of startup in this space called
Speaker:
00:20:38
barely trained, which is offering certification for AI model companies.
Speaker:
00:20:44
that have, produced their models relying solely on an authorized data set.
Speaker:
00:20:52
And then, you know, theory is, you are, a company that wants to leverage AI,
Speaker:
00:20:58
you can get more comfort in knowing that, you're relying on AI model, an
Speaker:
00:21:03
AI company that is fairly trained.
Speaker:
00:21:04
And the last time I checked, there were only a few dozen companies that had
Speaker:
00:21:08
that certification, but, that may grow.
Speaker:
00:21:11
So, maybe enterprise users would go for the fairly trained type,
Speaker:
00:21:15
because they're much more concerned, frankly, than most everyday users
Speaker:
00:21:20
about the quality of that output.
Speaker:
00:21:22
it seems like if they're using it, to create public facing
Speaker:
00:21:26
materials, they would want that fairly trained data set behind it.
Speaker:
00:21:31
they did, they also give reps and warranties when you go through them
Speaker:
00:21:33
regarding the quality of the output.
Speaker:
00:21:35
Do they give representations and warranties fairly trained provided
Speaker:
00:21:39
anyone who well, fairly trained or someone who has licensed their
Speaker:
00:21:44
data from fairly trained would they then in their terms of use have.
Speaker:
00:21:49
Represent fairly,
Speaker:
00:21:51
trained doesn't license data.
Speaker:
00:21:53
Fairly trained is a certification program.
Speaker:
00:21:56
So if an AI company, wants this certification to show everyone, that.
Speaker:
00:22:04
They have relied on an authorized data set, then this is a
Speaker:
00:22:07
certification that they can apply for.
Speaker:
00:22:10
Got it.
Speaker:
00:22:10
Okay.
Speaker:
00:22:11
because I believe that there are some platforms that do provide
Speaker:
00:22:15
indemnification, although they have a bunch of provisos, where's that going?
Speaker:
00:22:19
So that users feel more.
Speaker:
00:22:21
Comfort there, right?
Speaker:
00:22:22
So, I mean, I think that's part of their responding to, this pain point
Speaker:
00:22:27
of needing to make their customers more comfortable with using their product.
Speaker:
00:22:32
they are, providing certain indemnifications, it remains to be seen,
Speaker:
00:22:37
how effective those indemnifications would be if a customer were actually sued.
Speaker:
00:22:44
And as you mentioned, they do have, a lot of exclusions, personally, I think that is
Speaker:
00:22:51
just sort of an intermediate stop gap and they are going to be pushed more towards,
Speaker:
00:22:57
More licensing of their data sets.
Speaker:
00:23:01
and I would say, while we wait for this to play out, I mean, as you
Speaker:
00:23:05
know, this lawsuit could take and probably will take a very long time.
Speaker:
00:23:09
the Google books case, for example, on which the AI companies are
Speaker:
00:23:12
relying to 10 years before finally reaching, that conclusion that, Google
Speaker:
00:23:18
books digitization was a fair use.
Speaker:
00:23:21
So I would say in the interim, AI companies, and those.
Speaker:
00:23:26
producing AI models should look more to, using, authorized data sets or
Speaker:
00:23:33
construction of their models and authorized data sets with a traceable
Speaker:
00:23:38
provenance so that, their customers, when using the, output or wanting
Speaker:
00:23:44
to put the output into use, can.
Speaker:
00:23:47
Know what the source is and do that analysis of is this
Speaker:
00:23:51
violating any copyright?
Speaker:
00:23:52
Is this violating any right of publicity or trademark or anything else?
Speaker:
00:23:57
I would say for the companies that want to leverage a I, when you're looking for
Speaker:
00:24:01
partners, you do want to look at partners who are using authorized data sets.
Speaker:
00:24:08
Right now, what I see is that a lot of companies, brands.
Speaker:
00:24:13
companies in the film and television industry that are actually leveraging AI
Speaker:
00:24:19
and it is being leveraged, but they're using it for a first draft or a proof
Speaker:
00:24:26
of concept for things that are iterative and you'll need to be turned around
Speaker:
00:24:31
very quickly, but they're not using it.
Speaker:
00:24:34
As part of the final consumer facing output, just due to those copyright
Speaker:
00:24:42
reasons, both the reasons we just discussed, fear of having any type
Speaker:
00:24:47
of legal liability, but also, because there are limitations on the degree to
Speaker:
00:24:53
which you can protect, output that's generated by, artificial intelligence.
Speaker:
00:24:59
Right, so when they have a.
Speaker:
00:25:01
Authorized data set.
Speaker:
00:25:03
And does the output come with footnotes with What does that look like?
Speaker:
00:25:07
Do you know?
Speaker:
00:25:07
Have you seen, what that looks like
Speaker:
00:25:09
tell us what the sources are with it?
Speaker:
00:25:12
Like, does it identify?
Speaker:
00:25:13
Yeah.
Speaker:
00:25:13
Oh, oh, oh, I see.
Speaker:
00:25:15
When is it authorized?
Speaker:
00:25:15
It is right.
Speaker:
00:25:16
To my knowledge, it is not coming with anything.
Speaker:
00:25:21
And you're talking about the fairly trained component, right?
Speaker:
00:25:23
Yeah, is not coming with anything.
Speaker:
00:25:26
but that does need to be a path toward which we're traveling.
Speaker:
00:25:31
And there has been like a lot of conversation about that in this space that
Speaker:
00:25:36
it needs to be, you marked, needs to be traced in terms of, what was the source?
Speaker:
00:25:42
What did you rely on to do that?
Speaker:
00:25:46
Yeah.
Speaker:
00:25:46
As far as, the magic that happens inside of generative AI
Speaker:
00:25:50
platform, do we know what that is?
Speaker:
00:25:53
Or is that kind of the trade secrets of each companies?
Speaker:
00:25:56
Or is there general technology that Makes the magic happen?
Speaker:
00:26:00
I am not the expert on the technology inside of the AI models.
Speaker:
00:26:06
I can share what I know.
Speaker:
00:26:08
In part, it does depend on the approach that they've used, whether it's supervised
Speaker:
00:26:14
learning or unsupervised learning.
Speaker:
00:26:16
Which to make it very simple depends on how much you assisted the machine,
Speaker:
00:26:21
like, did you mark things and tell them, this is a dog and this is a cat
Speaker:
00:26:26
or did you just give them like kind of millions of pictures and kind of let
Speaker:
00:26:29
them figure it out when you let them figure it out when it's unsupervised,
Speaker:
00:26:35
it is more of a black box in terms of how they got to that answer,
Speaker:
00:26:40
which brings up all sorts of other.
Speaker:
00:26:43
societal issue, right?
Speaker:
00:26:45
I think it's a time to play it about
Speaker:
00:26:47
a, uh, interesting.
Speaker:
00:26:48
can we wrap up with some best practices just for your everyday kind of chat, GBT,
Speaker:
00:26:53
Janet, what is the Google on Genesis?
Speaker:
00:26:55
What does it, user like when they're using it, for this audience, the expertise
Speaker:
00:26:59
based business, maybe they're using it to create first drafts or to help them with
Speaker:
00:27:03
social media posts or something like that, like just some general best practices.
Speaker:
00:27:08
Sure, you want to be, circumspect about any, confidential or proprietary
Speaker:
00:27:15
information you include in a prompt, may want to anonymize it.
Speaker:
00:27:21
you need to keep in mind that, whatever output you get from, the
Speaker:
00:27:26
AI model, may not be eligible for copyright protection if this is, um.
Speaker:
00:27:31
something, material or output that you are passing on to a client or to a
Speaker:
00:27:38
customer, you may need to disclose that use, and you have to make sure that,
Speaker:
00:27:45
you're using the output Depending on the extent to which you're using it, are
Speaker:
00:27:51
you using it just for, a little bit of assistance in, modifying a few sentences?
Speaker:
00:27:57
Or are you actually producing images with it or producing an entire report with it?
Speaker:
00:28:02
You gonna want to make sure that you're procedures for like using generative
Speaker:
00:28:07
AI are consistent with the, contract that you have with your customer.
Speaker:
00:28:12
If you want to know whether or not, your material, your prompts are being
Speaker:
00:28:17
incorporated into the training data and being used to further train that AI model,
Speaker:
00:28:25
take a look at the terms and conditions.
Speaker:
00:28:28
to give you an example for chat, if you're using the free model,
Speaker:
00:28:34
and it is, recording your history.
Speaker:
00:28:36
Of your prompts, then, the prompts that you put in.
Speaker:
00:28:40
There are subject to being included as part of future training data.
Speaker:
00:28:45
Yeah.
Speaker:
00:28:46
So, if it along the left hand side there, you have scroll
Speaker:
00:28:49
through and see all your graphs.
Speaker:
00:28:51
I have.
Speaker:
00:28:51
that means it is going into the training data.
Speaker:
00:28:54
That is excellent.
Speaker:
00:28:55
can't say with certainty that it is going into the training data.
Speaker:
00:28:58
But I would say, it is susceptible to being used.
Speaker:
00:29:02
It's like they have not provided you, CHAT2P has not provided
Speaker:
00:29:06
you any representation that they will not use it for training.
Speaker:
00:29:09
Right.
Speaker:
00:29:10
Very good.
Speaker:
00:29:10
Thank you for making that distinction.
Speaker:
00:29:12
thank you for this.
Speaker:
00:29:13
this podcast is to help create a society that, and an economy
Speaker:
00:29:18
that works for more of us.
Speaker:
00:29:20
So I love to ask my guests, if there is an organization or a person who is doing the
Speaker:
00:29:26
good and hard work to help make an economy that works for more of us, is there
Speaker:
00:29:31
one that you'd like to share with us?
Speaker:
00:29:32
Sure, I really like organizations whose mission it is to
Speaker:
00:29:37
bridge the digital divide.
Speaker:
00:29:39
And one of my favorite is girls who code that has as part of its
Speaker:
00:29:44
mission, introducing more women.
Speaker:
00:29:47
into the technology field, and that's very apropos to our conversation
Speaker:
00:29:51
today, because as part of making AI, you beneficial for all humankind, we
Speaker:
00:29:58
really do need, a diverse perspective.
Speaker:
00:30:02
Yeah, I mean, we know that just when we talk about the Trina dating sets, like
Speaker:
00:30:06
what data is going in there, obviously.
Speaker:
00:30:08
what the output is only as diverse as the input, right.
Speaker:
00:30:11
And how it's being trained.
Speaker:
00:30:13
And I know that has come up in a number of, controversial ways as well, but
Speaker:
00:30:17
whether something's leaning this way or that way, but we definitely want to make
Speaker:
00:30:20
sure everyone has a voice in the future.
Speaker:
00:30:23
Thank you for that one.
Speaker:
00:30:24
And we will put that in the show notes along with how people can reach you.
Speaker:
00:30:29
where do you hang out, Joy?
Speaker:
00:30:30
And how can people get in touch with you to find out more?
Speaker:
00:30:32
Sure, so I'm always, , happy to, chat with, people doing innovative
Speaker:
00:30:38
things with technology, especially in the digital technology, online and
Speaker:
00:30:43
entertainment space, so they can find me through my website, which is www.
Speaker:
00:30:47
joybutler.
Speaker:
00:30:49
com.
Speaker:
00:30:49
And I'm awesome.
Speaker:
00:30:50
So, on LinkedIn, I have Joy Butler.
Speaker:
00:30:53
Awesome.
Speaker:
00:30:53
Well, thank you so much.
Speaker:
00:30:55
And, yes, everyone, please, follow Joy and, let us know if you have
Speaker:
00:30:59
any other questions about AI.
Speaker:
00:31:01
I know it's constantly evolving.
Speaker:
00:31:03
There's always going to be something new and we can continue
Speaker:
00:31:06
this conversation in the future.
Speaker:
00:31:08
Thanks again, Joy.
Speaker:
00:31:09
Thank you.
