Latest / CLEs You Actually Want to Hear / Synthetic Reality & Authentication: New Problems in Evidence
Transcript
- 0:03Welcome to Synthetic Reality and Authentication,
- 0:08New Problems and Evidence. I'm Lori Critt -Martin,
- 0:11a Certified Family Law Specialist, representing
- 0:14all genders in marital dissolutions, parentage
- 0:17matters, contested trial custody actions, and
- 0:19other issues. I'm delighted to introduce our
- 0:23speaker. the Honorable Curtis Carnell, retired.
- 0:26Judge Curtis Carnell, retired, served on the
- 0:29California Superior Court in San Francisco, including
- 0:33the complex litigation department, and held numerous
- 0:36leadership roles within the judiciary, contributing
- 0:39to statewide rules and policy on appellate practice,
- 0:44discovery, and court administration. A Harvard
- 0:47College graduate with a JD from the University
- 0:50of Pennsylvania, he is a former federal law clerk
- 0:53and an assistant U .S. attorney, as well as an
- 0:58award -winning judicial educator and prolific
- 1:01author and co -author of leading civil procedure
- 1:05treatises. Judge Carnow has lectured at major
- 1:08law schools nationwide, taught civil litigation
- 1:11to California judges, and is a member of the
- 1:14American Law Institute. And now let's get started.
- 1:21I am not Morgan Freeman, and what you see is
- 1:25not real. Well, at least in contemporary terms,
- 1:29it is not. What if I were to tell you that I
- 1:33am not even a human being? Would you believe
- 1:36me? What is your perception of reality? Is it
- 1:41the ability to capture, process, and make sense
- 1:45of the information our senses receive? If you
- 1:49can see... Here, taste or smell something. Does
- 1:53that make it real? Or is it simply the ability
- 1:57to feel? I would like to welcome you to the era
- 2:03of synthetic reality. Now, what do you see? Well,
- 2:21good afternoon or good morning whenever it is
- 2:23that you're watching this video. For those of
- 2:26us who are live, it's about 1 .30 in the afternoon
- 2:29California time. And I really appreciate the
- 2:32invitation from this Bar Association to talk
- 2:35to you about some of these emerging issues in
- 2:38evidence that have been sparked by the technology
- 2:41that you've just had a peek at. My timing on
- 2:45this is about 40 or 45 minutes. I will do my
- 2:49best to answer questions after that for those
- 2:51who remain, and you can always contact me through
- 2:54my website, which is in your materials, with
- 2:57any other questions. I do want to just briefly
- 2:59go through the materials just for a minute to
- 3:02let you know what's there. It's all been provided
- 3:04to you in advance. It appends two items. There
- 3:10is, number one, an article that I wrote about
- 3:12self -authentication. That's in the materials.
- 3:15And there's also The draft of the proposed Federal
- 3:19Rule of Evidence 707, which I'm going to talk
- 3:21about briefly in today's session, but the full
- 3:25draft and committee comments are in your materials
- 3:28at the end. In terms of citations, the materials
- 3:32include links to the Morgan Freeman video that
- 3:37you just saw, plus a bunch of other videos as
- 3:41well. And you can find them easily on YouTube.
- 3:47Other materials, the other sites there are a
- 3:50bunch of cases and sites under state authority,
- 3:53of course, Justice Mark Simon's volume on evidence,
- 3:56and some cases that I'm going to be mentioning
- 3:59of the citations if you want them are in the
- 4:01materials. And then there's a list of articles,
- 4:04most of which I won't be mentioning in my presentation,
- 4:08but they're pretty interesting, and almost all
- 4:10of them are available online. Not all of them
- 4:13are behind paywalls, for example, and they will
- 4:16bring you up to speed on some of the confusion
- 4:18that's been wrought by highly, highly convincing
- 4:22artificial materials that are being published.
- 4:25There is some materials on provenance, the concept
- 4:29I'll be talking about, and then other materials,
- 4:32including the citation to the Walter Benjamin
- 4:34essays that I will mention and other things that
- 4:37I'll be mentioning. And then on page four of
- 4:39the materials, finally, are citations and links
- 4:42to other articles that I've written about artificial
- 4:46intelligence. But let me just... drop right into
- 4:50today's topic. I was interested in this material.
- 4:53It really arose from a combination of two things.
- 4:56First of all, the entirely fake video clip that
- 4:59you had a look at another fake videos and the
- 5:02fact that they are indistinguishable, at least
- 5:04to my eyes and probably to your eyes if you're,
- 5:07unless you're an expert, indistinguishable from
- 5:09the real thing. They are deemed Secondly, the
- 5:14notion of self -authentication, which a lot of
- 5:17people think means that you can have an item
- 5:19in your hand, show it to the judge, and it will
- 5:22literally self -authenticate in the sense that
- 5:24it doesn't need a sponsoring witness. So some
- 5:28people say, I don't think that's probably true,
- 5:32but it's a fascinating area for us to explore
- 5:34in the context of some of these emerging technologies.
- 5:38A couple of disclaimers. Number one, obviously
- 5:40nothing I say here is an indication of how I
- 5:43might actually rule a trial. I am still sitting
- 5:45on the member of the Chief Justice's Assigned
- 5:50Judges Program, so I may show up in various counties
- 5:53around the state to help out our local courts.
- 5:58As I mentioned, my rulings in those cases will
- 6:02depend on the facts and the law that's presented
- 6:04to me in those cases. The last caveat is that
- 6:07I'm not going to be so much summarizing the law
- 6:09or telling you what the law is today. There'll
- 6:11be a little tiny bit of that. But what I really
- 6:14want to do today is hopefully suggest new ways
- 6:16of thinking about the issues, given the developments
- 6:20in technology. Three points that I hope to make.
- 6:24First, even the most powerfully convincing documents
- 6:28and even those that are said to be self -authenticating
- 6:32will not get admitted. unless there is oral testimony
- 6:36that lays the foundation for those things. And
- 6:40if the jury disbelieves some of that oral testimony
- 6:45after the judge has admitted the item, even the
- 6:51most apparently powerful document may backfire
- 6:54and have no effect at all. And as I'll mention
- 6:56later on, it's exactly what happened in one of
- 6:59my cases. The second thing that I want to just
- 7:02note is the obvious fact that almost all of the
- 7:04evidence or much of the evidence introduced in
- 7:06today's trials is digital. And that really challenges
- 7:10our old notions of authenticity and the role
- 7:13of the original, problems that we never had when
- 7:16we had a handwritten agreement or a handwritten
- 7:18document. The fake videos that are created by
- 7:22Sora 2, which is what you saw, are really, I
- 7:25think, just the tip of a very nasty iceberg.
- 7:29With digital things the distinction between testing
- 7:33for authenticity on the one hand and the secondary
- 7:36Evidence rule on the other hand these things
- 7:39I think really get blurred in an interesting
- 7:41way and thirdly at the end of this I'm going
- 7:45to conclude with a series of very practical suggestions
- 7:48about how to handle and how to object to the
- 7:52admission of digital evidence first, I think
- 7:55we really need to acknowledge the power of of
- 7:59the document, the power of the email, the power
- 8:02of the text that comes up at trial. Photos and
- 8:05evidence are far more persuasive very often than
- 8:09the witness's oral testimony at trial. It's routine
- 8:12to do pretty severe cross -examination of a witness
- 8:15with an email chain, or perhaps with a video.
- 8:20And the video, we think, really shows the way
- 8:22it is, as opposed to the way that a witness might
- 8:26describe it to a jury. A witness perhaps who
- 8:30can't remember all that well or can't speak all
- 8:32that well or just doesn't come across as articulate
- 8:36and knowledgeable, even though they may well
- 8:39be telling the truth and are being accurate.
- 8:42They might forget details and the jury itself
- 8:45knows the frailty of memory. They're probably
- 8:48going to get reminded about it in the jury instructions.
- 8:52But when it comes to the picture or the actual
- 8:55email, they will think, well, that doesn't lie.
- 8:58The documents don't lie. They don't have a motive
- 9:02to fib, so we say. We tend to believe the documents,
- 9:08by documents I mean videos and photos as well,
- 9:11over the testimony. They are the gold standard.
- 9:15They are very often what makes or breaks a case.
- 9:19Not fallible witness testimony, so we say. So
- 9:22we'll get back to that apparent dichotomy in
- 9:24a bit, because the truth is that the admission
- 9:27of almost every single document, photo or video,
- 9:31actually does depend on oral witness testimony
- 9:34on the stand, whether a trial or in a pre -trial
- 9:38402 hearing. So a little background on the law,
- 9:42just so we're all on the same page. First, a
- 9:45quick refresh. There are three legal hurdles
- 9:48that every document really needs to pass before
- 9:53it's admitted. First, authentication. Second,
- 9:56the secondary evidence rule. And thirdly, hearsay.
- 10:00I'm going to be focusing on authentication. Is
- 10:03the evidence what the proponent claims it to
- 10:06be? We should be careful here, because the test
- 10:09is not exactly whether the item is the original.
- 10:16turning to what the role of the original is later
- 10:19on. The question is whether the document is what
- 10:22the proponent says it is. So if he claims it's
- 10:26a copy or a rough approximation of something,
- 10:30then that's the question. Is it really that?
- 10:33The secondary evidence rule, which replaced the
- 10:37best evidence rule in California around 1999,
- 10:41generally allows in copies of documents, copies
- 10:44of what many think of as an original. It's premised
- 10:48on being a faithful copy of the original. We're
- 10:52going to explore the extent to which that continues
- 10:55to be a useful way to think about it. By the
- 10:58way, just an aside here, I still get best evidence
- 11:02objections at trial. I don't know if you've seen
- 11:04this. Hopefully not from you, but from other
- 11:07lawyers, perhaps. I don't know why people still
- 11:09do this. I think it's maybe a carryover from
- 11:12watching too much Perry Mason. which is something
- 11:15I am certainly guilty of from time to time. But
- 11:19best evidence is not an objection that should
- 11:22be made. And then, of course, the third hurdle
- 11:25for documents is hearsay. Documents which are
- 11:28authentic, they're proper under the secondary
- 11:31evidence rule. Nevertheless, they might contain
- 11:33hearsay and not be admitted. I'm really not going
- 11:36to be talking about hearsay today. Now, as for
- 11:40secondary evidence, traditionally the issue only
- 11:43arises after the authentication issue has been
- 11:47handled. It's a completely separate new issue
- 11:50under evidence code 1401. But I think we're going
- 11:54to find that the issues of authenticity on the
- 11:57one hand and secondary evidence on the other
- 11:59are really closely mixed in many situations.
- 12:02So let's briefly talk about the legal test focusing
- 12:05again on authenticity, just to set part of the
- 12:08background. In federal court, we have federal
- 12:11rule of evidence 901 and all the way through
- 12:13903 with the admissibility of secondary evidence
- 12:17and federal rule of evidence 1001 and following.
- 12:24In state court, we have authenticity, which is
- 12:26handled in evidence code 1400 and following.
- 12:31And with authenticity, as you probably know,
- 12:33it's a very low threshold. What does that mean?
- 12:37What it means practically is that garbage sometimes
- 12:42gets in. Fake stuff might actually get admitted
- 12:45because the authenticity barrier is so low. And
- 12:50you may have to argue the issue ultimately to
- 12:52the fact finder. The rules of evidence are not
- 12:56a guarantee that only good stuff, valid stuff,
- 13:00truthful stuff gets in. They're just constraints
- 13:03that make it hopefully likely that only good
- 13:06stuff gets in. So here's what specifically makes
- 13:10authenticity a low threshold test. First of all,
- 13:15judges are going to resolve competing inferences
- 13:19about a document in favor of the proponent. of
- 13:22the document. Conflicts are going to be said
- 13:24to go to the weight of the evidence, in other
- 13:27words, for the fact finder, for the jury. Circumstantial
- 13:31evidence is going to be enough. We can look to
- 13:34see where, let's say, the document was found,
- 13:37who seems to have written it, the contents of
- 13:40the document itself, and so on. Thirdly, prima
- 13:44facie case is going to be just fine. Let me pause
- 13:47here to illustrate by what I mean by prima facie
- 13:49case being more than enough or being enough.
- 13:53Let's say a witness, Alice, gets on the stand
- 13:57and she says, this is the letter that I wrote
- 14:00to Bob. A witness says that, and then the proponent
- 14:07moves to admit the document based on that testimony
- 14:10and the other side jumps up. And I've seen this
- 14:12repeatedly. The other side jumps up and says,
- 14:15no, wait, judge, before you admit the document,
- 14:17I want to cross -examine that witness. Maybe
- 14:20to show that she's lying, or something like that.
- 14:24And the answer is usually no. If we interrupted
- 14:27direct testimony every time a witness provided
- 14:30the foundation for a document, I think we would
- 14:32all go mad. But more importantly, the weight
- 14:36of Alice's testimony, as to that, as a judge,
- 14:40I don't care. Argue it to the jury. For me, The
- 14:44low threshold of a prima facie case is if somebody
- 14:47believes that witness, would it be enough to
- 14:50authenticate? And if the answer is yes, it's
- 14:53admitted. It doesn't matter if I personally believe
- 14:57Alice or not, unless, of course, I'm a fact finder
- 15:00and I handle that issue down the road. Lastly,
- 15:04we can rely on presumptions to get things authenticated.
- 15:08Generally, they are rebuttable presumptions.
- 15:11I say this because there are different kinds
- 15:13of presumptions. They work like this. The presumptions,
- 15:17some of which are found in evidence code 1552
- 15:20through 1553, they can presume the accuracy of
- 15:24computer information and of the digital images
- 15:27that are printed versions that are purported
- 15:30to represent that information. If there's no
- 15:33contrary evidence to any of these presumptions,
- 15:36then The presumption wins and the presumption
- 15:39alone is enough to have the item authenticated.
- 15:43If there is some contrary evidence, then the
- 15:45proponent has to provide some actual other evidence
- 15:48and cannot just rely on the presumption. So bottom
- 15:51line here, I think is when it's really hard to
- 15:53figure out whether or not something is authentic
- 15:56in the case of computer data, for example, if
- 15:59it's really an authentic printout of what's inside
- 16:02the computer, that is going to be a problem for
- 16:05the party. objecting to the document. Presumptions
- 16:11are going to allow often the proponent to win
- 16:13unless the other side has at least some evidence
- 16:16to the contrary. Presumptions though actually
- 16:19often win out and the document gets admitted.
- 16:23Next by way of background, let me just make a
- 16:25contrast between what I'm going to call classical
- 16:28evidence and digital evidence. Classical things
- 16:34come into existence at a particular point in
- 16:37time and they have a unique path through time
- 16:41and through space. They are things like a painting,
- 16:44like a car, a specific piece of paper, or a specific
- 16:49book, or a specific hand -signed letter, or maybe
- 16:53a hand -signed agreement. Copies can be made
- 16:56of these things, but whether we can detect whether
- 17:01there is a copy or whether it's an original,
- 17:04this is something that might depend on the kind
- 17:06of tools that we have to analyze the item. And
- 17:09there is a classic distinction between an original
- 17:12and a copy of a classic item. This completely
- 17:16coherent concept that we have that kind of a
- 17:19distinction, original versus the copy. And it's
- 17:22what underlies the rationale of the secondary
- 17:25evidence rule. Now the notion of an original
- 17:29I suggest really only applies to these sorts
- 17:31of classical objects classical objects like the
- 17:3545 .52 karat Deep blue Hope diamond. There's
- 17:40only one in the universe Maybe the pipe the actual
- 17:44pipe that Einstein Smoked or the hat the actual
- 17:47hat that Lincoln wore. These are all classic
- 17:50objects But digital objects are not like that
- 17:55i think at all digital copies are usually very
- 17:58very easy to make and often they are entirely
- 18:02indistinguishable from anything that you might
- 18:06want to call a copy they can be perfect fakes
- 18:10the point was made a little bit differently by
- 18:13the critic and essayist walter benjamin uh in
- 18:16the 30s the 1930s that's a site in your materials.
- 18:20What Benjamin did is he drew this distinction
- 18:22between classic works of art like a painting
- 18:26or a sculpture and the works of what he called
- 18:29mechanical reproduction such as prints and offsets
- 18:33and importantly photographs. Classic items have
- 18:36a uniqueness, an order. Walter Benjamin said,
- 18:39tied to their history through time and space
- 18:42and a physical location. And that is what established
- 18:46the authenticity of that classical item. He said,
- 18:51the presence of the original is the prerequisite
- 18:54to the concept of authenticity. But when he turned
- 18:58to mechanical reproductions, he said, by making
- 19:02many reproductions, it substitutes a plurality
- 19:04of copies for a unique existence. These are the
- 19:09endless copies, one made, every one perhaps,
- 19:12a kind of an original. These are the things that
- 19:15come off the printing press or developed from
- 19:17a negative, as many of them as you like. There
- 19:20is no authentic original in a way, and in a way
- 19:23there are no originals at all. The text of an
- 19:26email to you on my machine Is not literally not
- 19:31literally the same as the text of the email on
- 19:34your cell phone It's a copy but both could be
- 19:37termed an original as we pass them up to the
- 19:40judge to look at as authentic documents My iPad
- 19:47and my cell phone and my PC all may have more
- 19:52copies of exactly the same thing not exactly
- 19:55the same thing perhaps but in maybe the important
- 19:58ways, the same thing. And there's another original
- 20:01or another copy stationed somewhere else on Yahoo
- 20:04servers, on Google servers, and so on and so
- 20:06forth. With these things, as with all digital
- 20:10files, like text emails and online agreements
- 20:13and so on, we really cannot use the classic concept
- 20:19of an original to analyze either authenticity
- 20:22or secondary evidence. So we now live in this
- 20:26digital universe and often the things that we
- 20:28want to introduce at trial are these digital
- 20:31things, not the classic objects. The rules on
- 20:35computer generated data now are a result of a
- 20:40general acceptance that computers are reliable
- 20:42and don't make mistakes. And our Supreme Court
- 20:46said in Goldsmith, you've got the site in the
- 20:48materials, which is a 2014 case, the Supreme
- 20:52Court said two interesting things. which at the
- 20:55time were really not in any serious dispute.
- 20:58First, quote, we declined to require a greater
- 21:01showing of authentication for the admissibility
- 21:05of digital images merely because in theory they
- 21:10can be manipulated. And the second thing the
- 21:13court said was we have not required testimony
- 21:16regarding the acceptability, the accuracy or
- 21:19maintenance and reliability of computer hardware
- 21:22and software. The trouble is that now we do find
- 21:27this. It may still be true that computers are
- 21:30reliable in the sense of their maintenance and
- 21:34accuracy and the way their CPUs function is not
- 21:37something that we have to get into. But the possibilities
- 21:40of manipulation are no longer just in theory.
- 21:44They are pervasive at this point. And we really
- 21:47need to rethink our reliance on presumptions
- 21:50of data fidelity. or come up with a way to allow
- 21:53for a thoughtful examination of these issues
- 21:56at trial. So let's outline exactly what some
- 22:00of those manipulations are. And later on at the
- 22:03end, I'll discuss what lawyers can do about that
- 22:06situation. So manipulations. The app, Sora 2,
- 22:12produces a video indistinguishable from the real.
- 22:15While many photoshopped images are obvious fakes
- 22:20and some of those are actually really bad fakes
- 22:22coming from governments It's a little surprising
- 22:24how bad some of the fakes are that come out of
- 22:27governmental entities around the world You would
- 22:30think that they would have better tools, but
- 22:32those are fakes Sometimes you can tell and sometimes
- 22:36you cannot And some of the articles that I've
- 22:39provided to you in the handout, there's about
- 22:41four or five or six of them, they relate to fake
- 22:44videos and fake photographs which have had a
- 22:49real impact on people's views of very important
- 22:53events in our recent history. And they have shaped
- 22:57the public discussion of those events. But the
- 23:01problem that I'm getting at is a little bit more
- 23:04subtle than the distinction between a fake and
- 23:06a truthful image. In many situations, the proponent
- 23:10is not actually trying to pass off a fake image
- 23:14as an honest one. He is trying to pass off what
- 23:19he thinks is a real image, but just with maybe
- 23:23some enhancements. Widely available technology
- 23:28can clean up an image, fill in the interstices
- 23:32between video frames, make a color perhaps more
- 23:36or less obvious, change the hues, change the
- 23:39values of the colors, and otherwise make an obscure
- 23:43image perhaps more clear or vice versa the other
- 23:47way, make something maybe just a little less
- 23:49clear. All images are actually in some way edited,
- 23:54enhanced, or cleaned up. perhaps with the use
- 23:57of a lens or a filter, we're in what we might
- 24:00call post -production. Indeed, if we compare
- 24:03the photo to what a human, an ordinary day -to
- 24:07-day experience using only his or her eyes, photos
- 24:11will reveal things that humans simply can't see
- 24:14or simply don't pick up on, such as by freezing
- 24:18an instant. showing a person blinking, but in
- 24:21the photo it looks like she's asleep because
- 24:24the eyes are closed. A photo showing a smile,
- 24:28which is really part of a pained grimace, or
- 24:32a person grabbing somebody, but actually in real
- 24:35life somebody just tripped. Many digital files
- 24:39are manipulated between the time that they're
- 24:41first created and the time we view them, often
- 24:45most frequently through what is called lossy
- 24:48compression. where seemingly redundant bits are
- 24:51tossed out, which conveniently reduces files
- 24:54for transmission, maybe from me to you. Lossy
- 24:57compression will occur. You may or may not know
- 24:59that it has occurred. But when it does, it irretrievably
- 25:02degrades the quality of the image. Files can
- 25:06then be reconstructed at the other end by averaging
- 25:10or interpolation and so on. AI is now using a
- 25:14process called AI upscaling, which based on the
- 25:19AI's view of millions of training images, creates
- 25:22a really beautifully nicely finished product.
- 25:25But like every other ordinary routine technique
- 25:29that I've just alluded to, combines elements
- 25:31of the so -called original item with what we
- 25:34might call new bits. What is the distinction
- 25:39between, on the one hand, the edited or translation
- 25:42or just slightly cleaned up version of an original
- 25:44and, on the other hand, what we might call a
- 25:47complete fake, a Sora 2 image? And the answer
- 25:52is not much. These are modifications that are
- 25:55just simply taking place on a scale of a little
- 25:58tiny bit to a lot. Each of these things takes
- 26:02elements of what we might want to call an original
- 26:05and recombines it. In the case of an enhanced
- 26:08video, for example, to get more contrast and
- 26:11to enable us to get an item more obviously depicted,
- 26:16we might take an original exposure and add or
- 26:18subtract pixels or other elements. With text,
- 26:22we might add color. With emails, we might include
- 26:24or cut. cut portions of what on some devices
- 26:27might be a string of emails. I had a case involving
- 26:30texts, and there were actually two versions of
- 26:33this because somebody thought that the string
- 26:36of text really started here, whereas somebody
- 26:38else included an exchange that had occurred above.
- 26:42It was a modification done with complete innocence,
- 26:46but what the copies were and what the originals
- 26:50were was confusing. An agreement Is created these
- 26:56are the sorts of agreements you and I enter into
- 26:57all the time is created when you sign into a
- 27:00hospital for example when you submit your credit
- 27:02card and get an electronically created receipt
- 27:05from an online vendor when maybe you signed an
- 27:08employment agreement online Text emails all of
- 27:12these things including the online agreement.
- 27:15They are all just fleeting They're all just representations
- 27:18of what are otherwise fleeting electronic events.
- 27:22Most of these documents are just electronic,
- 27:25signed perhaps when you agree to the terms and
- 27:27conditions of a website by doing what? By using
- 27:29the website, by clicking, by ticking a box, by
- 27:33ordering something. Now, how much of that website,
- 27:36of those circumstances, is going to get produced
- 27:39in court to show what the agreement was, to show
- 27:43what it is in its entirety? all the pages that
- 27:48led up to the ticked box, all aspects of the
- 27:51page, the checkout page that says here are the
- 27:54terms and conditions, and how close to the specific
- 27:57consumer's idiosyncratic experience on his or
- 28:02her webpage has to be produced to show what the
- 28:06original agreement was. How much of that has
- 28:09to be replicated? Even what seems to be a traditional
- 28:13agreement, like, for example, a publishing agreement
- 28:15that I just signed with The Law Review a couple
- 28:17of months ago, is not exactly the original agreement.
- 28:23I've signed, quote unquote, a PDF associating
- 28:26in... one file, two different files or groups
- 28:29of data, copies of which, not the original, but
- 28:32copies of which, by the way, were sent to the
- 28:34Law Review, which then presumably copied into
- 28:37the Law Review's computer files, or perhaps were
- 28:41sent to the cloud, a third -party computer. When
- 28:44all of that is printed out, let's say two years
- 28:46from now when we have a dispute, yet another
- 28:49copy, this time in printed form, will be produced.
- 28:52The final copy, The last in a very long line
- 28:56of multitudes of copies in various formats will,
- 29:00for the first time, actually look like an original
- 29:05agreement. And the irony of that, I hope, is
- 29:09palpable. So in this case of the fake Morgan
- 29:13Freeman video that we saw, we again might start
- 29:18with an acknowledged original. the original video,
- 29:23an original video or some kind of original picture,
- 29:26maybe the original timber of his voice that somebody
- 29:28recorded, perhaps samples of the original words.
- 29:32We would then remix the words, shift pixels and
- 29:35so on, so it looked like Freeman's voice was
- 29:38coming out of the lips and the lips were synchronized
- 29:41with the words. The system is trained on and
- 29:44is fed large samples and then it issues this
- 29:47remix. And I suggest that really every digital
- 29:51file, all the ones that I have mentioned, are
- 29:54in a sense a digital remix of some sort. The
- 29:58ones that we admit and the ones that we don't
- 30:00admit, they all have this aspect. Every digital
- 30:05file in this sense is faithless. Let me just
- 30:09do a brief aside on the concept of the liar's
- 30:12dividend before I get into what we can practically
- 30:15do with some of these situations. The liar's
- 30:18dividend Now, given what we know and what I think
- 30:25the general population is sensitive to about
- 30:28the possibilities of fake documents, lawyers
- 30:31can inject doubt even when there's no reason
- 30:34for doubt. Can we always believe our eyes, a
- 30:38lawyer will say. And there are subtle ways to
- 30:42inject uncertainty into a situation where there's
- 30:44no really articulable basis for uncertainty.
- 30:48And this is something I think all of us know.
- 30:50We can cast out by raising our eyebrows, maybe
- 30:54a long pause in an exaggerated situation, some
- 30:58sort of gesture in front of the jury. Yeah, there'll
- 31:01be an objection to it, but what's done is done.
- 31:04The point now, I think, is that the general population
- 31:07has a built -in unease about Photoshopping and
- 31:11other easily done manipulations. There are just
- 31:14too many stories about fake videos. So even when
- 31:18there is no reason at all to attack or to undermine
- 31:22evidence, even when no witness actually undermines
- 31:27it, the jury might still be suspicious. This
- 31:30is called the liar's dividend. The dashcam caught
- 31:35the accident? I don't know. The cell phone caught
- 31:38the assault? Really? The earth orbits the sun?
- 31:41Who's to know? Who's to know what the truth is?
- 31:43And injecting that kind of doubt, in a sense,
- 31:46out of nothing is the liar's dividend. Then let's
- 31:51move to our work in trials and what we can do
- 31:53about the situation. What is it now to authenticate
- 31:58when it's a digital item, when there's no original,
- 32:03as there might be with a classic object? And
- 32:06here's where authentication and secondary evidence
- 32:10merge, I suggest. With classical things like
- 32:12a handwritten note, We first establish the authenticity
- 32:16of the original, and then under the secondary
- 32:18evidence rule, show that the copy in court is
- 32:21a fair copy. With digital files, we can't technically
- 32:25do that because we don't have the touchstone
- 32:28of the original. We have one of perhaps an infinite
- 32:33set of copies and we establish that this copy
- 32:36fairly represents What I'll call some set of
- 32:39legal facts facts such as that I agreed to arbitration
- 32:44with Yahoo facts That I have an agreement with
- 32:48my service provider of some other sort the facts
- 32:51that Alice sold her farm to Bob The fact that
- 32:55Charlie sent a threatening email or text to David
- 33:00in each case. I can't technically tell the jury
- 33:04that I'm showing them the very same agreement
- 33:07or the very same email or text that was sent
- 33:10in the past. I can't say that what I have here
- 33:14in court looks exactly like what was sent, the
- 33:18same line breaks, exactly the same hue of color,
- 33:22the same font, and so on and so forth. But what
- 33:25I can say is that I have, for purposes relevant
- 33:30to the issues in this trial, What I have is the
- 33:35same. You can call it an authenticity test. You
- 33:39can call it a secondary evidence test. The point
- 33:43is that the copy that I have in court is the
- 33:45same in certain pertinent respects. In the key
- 33:50respects that matter to this case, it's the same
- 33:54as the past historical version of the electronic
- 33:57file or the assemblage of electronic files. That
- 34:02assemblage of electronic files constitutes a
- 34:06legal event, an agreement, a threat and so on
- 34:08and so forth. And we are going to need oral testimony
- 34:12for this. People on the stand who say, yes, these
- 34:17are the same words. This is the same text. This
- 34:21is what my computer recorded. This is pretty
- 34:24much enough. What I saw. As my article on self
- 34:29-authentication, which is in your materials notes,
- 34:34you also need to have circumstantial evidence
- 34:37in that context as well. And this is true even
- 34:40though the phrase self -authentication suggests
- 34:42that just holding up the document may be enough.
- 34:45But it is never enough. You always need some
- 34:48background information, some background evidence
- 34:50to pass any of the tests, either federal or state
- 34:54law, for self -authentication. The classic test
- 34:57for self -authentication in state courts looks
- 35:00to, for example, whether the facts that are recited
- 35:03in the document are only known to the purported
- 35:07author and unlikely to be known by anybody else,
- 35:11thus showing that the purported author really
- 35:15did write it. But under that test, no one can
- 35:21possibly evaluate what the author might know
- 35:23and what other people are not likely to know,
- 35:28unless one has an enormous amount of understanding
- 35:31and evidence about the context. You can't tell
- 35:35any of that. When anybody knows about the facts
- 35:37of the document just by looking at it, you need
- 35:40people on the stand to fill all that out, witnesses
- 35:43who are or aren't telling the truth. And when
- 35:48it comes to admissibility, as I have said, judges
- 35:51are almost never ever going to judge credibility.
- 35:55If there's a minimal foundation, judges will
- 35:58let it in and have the jury decide the ultimate
- 36:01issue of whether or not it is what it purports
- 36:03to be. So even the best guilt -edged killer document
- 36:09or video or photo will fall by the wayside if
- 36:13the jury doesn't believe the sponsoring witnesses.
- 36:18If the word of a single witness is enough to
- 36:22get a text or photo or video and evidence, even
- 36:25if that witness is undermined or contradicted
- 36:27by other evidence. Even if it's clear that the
- 36:30thing offered at trial doesn't in some respects
- 36:33even look like an older version of the digital
- 36:36file, what can one do to head off the admission
- 36:39when you think the evidence is actually a fake?
- 36:43What are the tools that lawyers have to challenge
- 36:48these sorts of exhibits, digital files? or contrary
- 36:53-wise, what are the tools to avoid a fight, a
- 36:57trial over digital documents which are credible
- 37:00and which ought to be admitted because they do
- 37:03honestly provide something real to the fact -finder
- 37:07that matters to the issues in the case. The good
- 37:10news is that unlike what happens on the internet
- 37:13and social media, courtrooms have standards.
- 37:18So let me provide you now the last part of my
- 37:21remarks, an arsenal of tools. Underlying most
- 37:25of these things is a single concept, reliability.
- 37:30That's the crowbar I suggest to open up this
- 37:34locker of tools. And here's why. First of all,
- 37:38some expertise may be needed to evaluate suspect
- 37:42digital files. Number two, judges are familiar
- 37:46with the rules on expert testimony, we hope.
- 37:50And number three, reliability is the key test
- 37:56to the admission of expert testimony. First though,
- 38:00I'd like to think that foremost, and something
- 38:02completely different from what we've been talking
- 38:04about, is the ethical role of the lawyer. Lawyers
- 38:09have a duty of candor. They cannot present evidence
- 38:12known to be fake. Sure, it's okay to rely on
- 38:15a client's view of the facts, but there is an
- 38:18ethical limit. When you know that the client
- 38:21is lying, when you've got good reason to know
- 38:24that a digital file has been fabricated or modified
- 38:28in a material way, you simply cannot proceed
- 38:33with that. You've got to dissuade the client
- 38:36or you need to resign from the case. Your obligations
- 38:40as an officer of the court actually supersede
- 38:43the obligations to the client. Bad evidence should
- 38:48never find its way into a courtroom in the first
- 38:50place. The duty of candor to the court, as I
- 38:54say, supersedes. Next, in what could be seen
- 38:57perhaps as an extrapolation of the current view
- 39:00of duty of candor, I suggest that parties always
- 39:04disclose post -production to the other side,
- 39:08any enhancement, as a matter of course. There's
- 39:11a professional organization, the Coalition for
- 39:14Content Provenance and Authenticity, Cite in
- 39:17Your Materials, which has set forth standards
- 39:19on how digital enhancements and AI -generated
- 39:22materials could be marked so as to avoid misunderstandings
- 39:26or, let me be blunt, fraud. And if this isn't
- 39:31an ongoing duty or it's not accepted by the lawyers
- 39:34on both sides as an duty, you can actually make
- 39:37it one. by issuing discovery demands or getting
- 39:41the judge in a case management order to make
- 39:44it a duty. And as part of that process, there
- 39:48should be a pretrial exchange of exhibits and
- 39:51a meeting between the lawyers to identify any
- 39:53disputes on admissibility. This is routine in
- 39:56many federal courts, but I'm afraid it is not
- 39:59routine in state courts, and it should be. Next,
- 40:04you should issue requests for admissions, RFAs.
- 40:07Request that a document is authentic. Get the
- 40:11other side to agree to that, long before trial,
- 40:14and pinpoint disagreements. In state court, you
- 40:17should also issue, at the same time, form Inauguratory
- 40:2117 .1, which demands the basis for any refusal
- 40:25to admit an RFA. Next, you may need to examine
- 40:30another side's computer, computers that are owned
- 40:34or operated by parties in the case. This can
- 40:38be hard to get, but the results can be decisive.
- 40:43I'm reminded of a 2018 case cited in your materials,
- 40:47People versus Jenelnik. Let me quote you something
- 40:51from that case, from the Jenelnik case. One of
- 40:55the claims in the lawsuit was for age discrimination,
- 40:57and one of the key documents was an email produced
- 41:00in discovery by Jenelnik purportedly sent to
- 41:03him. by the employer. The email concerned Janelnik's
- 41:07failure to file sales reports and contained language
- 41:11concerning Janelnik's quote aging memory and
- 41:15quote senility. A forensic examination of Janelnik's
- 41:19external hard drive in October of 2013, however,
- 41:24disclosed a version of this email without any
- 41:27reference to Janelnik's age or senility. And
- 41:33then there's a case I had. A demand for documents
- 41:36had gone out early in the case. And as to a certain
- 41:39category, there was nothing. And the written
- 41:41response to the demand was, there isn't anything.
- 41:44We don't have it. No such document. Fast forward.
- 41:47It's a few days before trial. And the document
- 41:50was just found. It just showed up. It was a miracle.
- 41:56Suspicious because the newly found document had
- 42:00just the words in it. that the producing side
- 42:04needed to help win the case. Should I have barred
- 42:08this item from trial? In state court, unlike
- 42:12federal court, there is no continuing duty to
- 42:16update discovery responses in order to make them
- 42:19continuously correct. So unless there's a finding
- 42:24that a prior response was deliberately false,
- 42:27it's pretty tough to block admission of this
- 42:29newly found document. And in this case, number
- 42:33one, there hadn't been a final supplemental discovery
- 42:36demand. There should have been, I guess. And
- 42:38in this case, the issue wasn't that the previous
- 42:41response was false. The suggestion was that that
- 42:45previous response had actually been true and
- 42:48the new document was false. So the document came
- 42:52into evidence. But the circumstances of its new
- 42:57finding sunk the case for the proponent. The
- 43:01fact finder didn't trust it. It backfired. So
- 43:05assume now discovery is pretty much done. Maybe
- 43:09there's a disagreement that you've pinpointed
- 43:11on admissibility. This can be handled perhaps
- 43:14in emotion in Liminate. We've now set the groundwork
- 43:18for a pre -trial hearing, so -called 402 hearing
- 43:21on admissibility. These hearings, I suggest,
- 43:24should be done as early as possible because the
- 43:26ruling can affect trial strategy, including which
- 43:29witnesses need to be subpoenaed to trial. This
- 43:33unfortunately is contrary to what often happens
- 43:36in state court certainly in cases where you don't
- 43:39find out who your judge is until the morning
- 43:41of trial and in other cases It's the judge may
- 43:45not be known until it's really too late. It's
- 43:47it's awkward But to the extent possible these
- 43:51are hearings that occur Weeks and in cases that
- 43:54I had as a single assignment judge including
- 43:56in complex litigation sometimes months before
- 43:58the actual trial date so We're at the pre -trial
- 44:02conference, whenever it may be. It may be that
- 44:05the reliability of a proposed electronic item
- 44:07needs to be tested. It needs to be shown by one
- 44:10party or another fake, or it has to be shown
- 44:12to be a fake. Or, and I'll get to this in more
- 44:15detail, it's sufficiently misleading so it shouldn't
- 44:20come into evidence. And this may need experts.
- 44:24Which is also unfortunate, because now we're
- 44:27talking about sometimes great expense. But experts
- 44:31can testify on a dysia of manipulation and trickery.
- 44:36The eye movement, the way jaw lines are carried
- 44:39from frame to frame, signs that a watermark,
- 44:43which are imposed automatically by Sora and other
- 44:46AI apps, has been covertly removed. The point
- 44:50here is that setting aside the expense issue,
- 44:53that making the problem one of expertise is very
- 44:56good for the party objecting to the document,
- 44:59because then that allows the judge to use traditional
- 45:03rules about reliability and bar the evidence
- 45:07at trial. Judges are far, far, far better placed
- 45:11than juries to evaluate expert testimony. of
- 45:16almost any sort, I suggest. And I cover that
- 45:18in the expert manual that is cited in your materials.
- 45:22Judges can take the time. They can read all of
- 45:25the reports, some of which are technically admissible
- 45:27and some of which may not be technically admissible.
- 45:30They can ask questions. They can target questions
- 45:34to the witness on crucial issues. They can listen
- 45:37to conflicting evidence from experts without
- 45:40having to decide what is formally admissible
- 45:43or not admissible at trial, and thereby they
- 45:46can get the background that they need to really
- 45:48decide the reliability issue. These are all things
- 45:52that are sometimes very difficult to do in front
- 45:55of a jury. This idea of creating digital files,
- 45:58and in particular those files which may have
- 46:01been devised by or manipulated by AI, doing it
- 46:05as a matter of expert testimony, has been taken
- 46:09up by the committee on the Federal Rule of Evidence,
- 46:13which is proposing now FRE 707. The proposed
- 46:17rules contain this notion of cross -examination,
- 46:21cross -examining the evidence in the same way
- 46:24as experts, live human experts are cross -examined
- 46:27on the stand. It expressly calls for the same
- 46:30standards of admissibility of machine -generated
- 46:33files as we have by experts. It tracks my comments.
- 46:38I think, pretty closely. And as always with experts,
- 46:42as I've said, reliability is the heart of where
- 46:45a judge needs to go in deciding admissibility.
- 46:49If you'd like some further ideas about how that
- 46:52kind of crossing the amination of a machine -generated
- 46:55document or opinion could work, have a look at
- 46:59my 2018 article, which is called The Opinion
- 47:02of Machines, which is again cited in your work.
- 47:05Now, I have to say that FRE 707 probably won't
- 47:09help alone detect deep fakes passed off as real
- 47:12pictures, because it only applies when the item
- 47:15is concededly machine generated. So it may or
- 47:17may not be useful in detecting a forgery. So
- 47:20back to this judge's review involving experts
- 47:23or not. Let me now provide another tool, this
- 47:26notion of provenance. Museums and buyers of artwork
- 47:30insist on knowing what the provenance of an artwork
- 47:34is going to be. It's history. It's record of
- 47:37ownership, which is used as a guide to authenticity,
- 47:40just the way Walter Benjamin talked about. It's
- 47:43what I think of as a chain of custody. And those
- 47:45of you with experience in criminal law will know
- 47:47exactly what I'm talking about. We want evidence
- 47:50of every step that that file took when it was
- 47:56made. who made it, and the list of every consecutive
- 47:59owners or possessors of that digital file, all
- 48:05the way up to the moment that it comes into court.
- 48:08What is its story, we might ask? And it's true,
- 48:13this notion of provenance was created for classical
- 48:15objects, but I think we can apply it to digital
- 48:18objects too. Each step in the creation of a digital
- 48:20file can be explored. Which computers, which
- 48:23applications were used, what transformations
- 48:25were performed, who did it, who controlled that
- 48:29digital file up to the moment that it was brought
- 48:31into court. And if there's a gap in this provenance,
- 48:34in this chain of custody, if you will, then that's
- 48:37your red flag, which is explored perhaps ultimately
- 48:41to have the item not admitted at trial. Now as
- 48:45one of my very last points, let me remind you
- 48:48of evidence code section 352 in state court.
- 48:53There's an equivalent, of course, in federal
- 48:55court, FRE 403, and both of these allow a judge
- 48:58to exclude evidence to avoid unfair prejudice.
- 49:02confusion of issues. That is, evidence that could
- 49:05mislead the jury or distract them from the main
- 49:07issues. 352 is a very, very powerful tool. Think
- 49:16of this. Otherwise, admissible evidence. It's
- 49:18relevant. It has enough to show authenticity.
- 49:23The secondary evidence rule has been met. There's
- 49:26no hearsay problem. And still, the evidence could
- 49:29be excluded. under 352. Not only that, the issue
- 49:33is a matter of the trial judge's discretion.
- 49:36And so theoretically, at least, it should be
- 49:38very difficult to overturn on appeal. Now, 352
- 49:42deserves its own half -day course. But let me
- 49:44just make a few notes. How might it work with
- 49:49digital files? Take the Morgan Freeman video,
- 49:51for example, that I showed you at the beginning
- 49:52of this session. It's a fake. Now, does that
- 49:55mean that by definition it's not admissible?
- 49:58Not at all. Not at all as I said digital all
- 50:01digital files are faithless some get in and something
- 50:04something some don't A party might be offering
- 50:07that video to try to admit it because it's shown
- 50:10to display exactly what Morgan Freeman actually
- 50:13sounds like, with no pretense at all that he
- 50:16made the video. Maybe that's relevant to the
- 50:18case. Let's assume it is. And it's true. The
- 50:20video really does tell you that Morgan Freeman
- 50:22sounds like that. I can be your witness because
- 50:25I know what Morgan Freeman sounds like in all
- 50:27of his movies, and I'll testify that that sounds
- 50:29like Morgan Freeman. So maybe with the express
- 50:32caveat that this is fake, in the sense that it
- 50:36was created by Sora 2, we might still admit the
- 50:39video, but could it be excluded under 352? Maybe
- 50:45it's too convincing, and more importantly, it's
- 50:47convincing for the wrong reasons. Using the fictitious
- 50:51image and the lip syncing, to very strongly suggest
- 50:55that this must be Freeman. And if that's the
- 50:58disputed issue in the case, and it must be otherwise
- 51:00we wouldn't be showing this video, then the jury
- 51:03really may be misled by this video in the sense
- 51:06of believing it for the wrong reason. And that
- 51:10is a 352 problem. Remember digital files, unlike
- 51:14classical objects, are in effect mutated. copies
- 51:17of each other. It's very possible to print out
- 51:20an online version of an agreement, which also
- 51:23could be misleading, leaving out all the surrounding
- 51:26text on the website, for example, the experience
- 51:29of being online, the color of the font, what
- 51:31the links actually look like. Have a look at
- 51:34the case cited in your materials called Cruz
- 51:37versus Tapestry. It's not about this issue in
- 51:41particular, but you will see there the website
- 51:43appeared one way to the consumer. and critically
- 51:46was described in a profoundly different way by
- 51:49the appellant's brief. Now, most of these printouts,
- 51:53digital files and so on and so forth, really
- 51:56do go through some post -production processing.
- 51:59There are recreations that could be a 352 issue.
- 52:02Now, a final word on handling these issues when
- 52:05you don't have the luxury of a 402 hearing. and
- 52:09you have some very complicated situations. After
- 52:12all, most cases really don't go to trial. Settlement
- 52:15and pre -trial motions like demurs and motions
- 52:17to dismiss or summary judgment take care of most
- 52:19cases. And also posing a real problem here are
- 52:22short cause cases, like in some family law cases,
- 52:25small claims and so on and so forth. And there's
- 52:27no time for some of the tools that I've noted.
- 52:31Surprises are likely, and summary judgment fact
- 52:35issues can easily be contrived or eliminated
- 52:38with declarations that are based on fake evidence.
- 52:42There may be no time to find out, and unless
- 52:44you've done your homework in discovery with the
- 52:47discovery tools that I've outlined, the fakes
- 52:50are going to be hard to handle. It may take some
- 52:52time for us, I think, as a legal community to
- 52:55re - to re -examine the very low bar, the low
- 52:58threshold that we have for authenticity. The
- 53:01shift in 1999 was done in part because there
- 53:05was a consensus that the world of today is no
- 53:08longer one in which is so deeply distrust copies
- 53:11of documents. The days of manual transcription
- 53:13and its inevitable error -ridden products have
- 53:16long since passed, according to the De Silva
- 53:19article that's in your materials. It's true that
- 53:23we have these guardrails in court, ethical duties,
- 53:25extensive discovery rights, the availability
- 53:28of experts, the power and the duty of the judge
- 53:30to exclude unreliable evidence that make the
- 53:34use of what turns out to be fake evidence unlikely.
- 53:37But the confidence I think that we had almost
- 53:4030 years ago when we moved to the best evidence
- 53:43rule has been undermined by generative AI products.
- 53:47And a new look at this evidentiary rule, the
- 53:51way that secondary evidence works now in California
- 53:53courts would be wise. In the meantime, be careful
- 53:58out there. Thank you very much. It's been a pleasure
- 54:01talking with you. This concludes today's presentation.
- 54:05Thank you so much Judge Carno for an excellent
- 54:07presentation and thank you so much to our audience.
- 54:09Have a great day everybody.