AI: Intellectual Property Litigation Discovery Practical Law The Journal

Reuters

01-07-2025

As more patent, trade secret, and other IP-related cases involving AI make their way through US courts, counsel and the judges overseeing those cases need to consider highly nuanced discovery and other issues specific to AI technologies. Like other software IP cases, AI technology lawsuits present black box challenges, where internal operations are hidden. However, AI technology lawsuits differ from traditional software matters in that AI systems continuously learn as they receive new input data.
For example, in the hypothetical facial recognition authentication system discussed below, the AI system's ability to recognize faces in uncontrolled environments and authenticate users improves as it observes more faces. Other AI systems may make decisions or take actions that are not always foreseeable and may not be explainable by their developers, presenting additional challenges for litigants, their counsel, and fact finders.
As used in the hypothetical and throughout this article, an algorithm refers to a series of steps, written in a programming language such as Python, that when executed by a computer performs a specific function. An AI model refers to a set of algorithms used together to replicate a real-world process. For example, the facial recognition authentication system could be considered an AI model because it replicates the human cognitive process of identifying a person by recognizing their facial features.
A well-implemented discovery strategy that is aimed at gaining insight into the nature of an AI model can help litigants understand how a disputed AI system works, which is crucial to counsel seeking to:
Present an accurate and compelling story to a judge or jury.
Mount a technically accurate defense for a client.
Negotiate a settlement agreement while armed with all the facts.
This article provides an overview of key discovery issues and practical strategies companies and their counsel can use before and during an IP lawsuit involving an AI technology using a hypothetical HAV facial recognition technology as a guide, including:
The factual elements required to prove patent infringement and trade secret claims.
Pre-suit considerations, such as:
important investigative first steps; and
litigation (or preservation) holds.
Filing or opposing a motion to dismiss.
Key AI-related discovery issues, such as:
initial disclosures;
written discovery; and
party, fact, and expert witness depositions.
(For the complete version of this resource, which includes guidance on source code review and third-party discovery, see Artificial Intelligence: Intellectual Property Litigation Discovery Considerations: Overview on Practical Law; for a collection of resources on potential legal issues surrounding AI, see AI Toolkit (US) on Practical Law.)
Hypothetical AI Technology Lawsuit: HAV Facial Recognition Technology
Company XYZ has patented a new facial recognition system for authenticating HAV users, which uses an in-cabin camera and software to:
Interrogate passenger faces when they enter the vehicle.
Authenticate the person who takes the driving position in the vehicle.
Disengage safeguards, allowing the driver to operate the vehicle.
Schematically, the authentication process might look like the following:
XYZ has invested its resources in the future of the HAV market, forgoing the existing semi-autonomous passenger vehicle market. Company ABC has taken a different approach in the market by selling its facial recognition authentication system to vehicle original equipment manufacturers (OEMs) and Tier 1 suppliers for integration into semi-autonomous vehicles. Instead of patenting its system, ABC has chosen to maintain its software, which it calls IVIEWDRIVE®, and certain of its algorithms as trade secrets.
After XYZ hires some of ABC's machine learning engineers, ABC sues XYZ for misappropriation of its IVIEWDRIVE® software and algorithms in violation of the Defend Trade Secrets Act (DTSA) (18 U.S.C. §§ 1831 to 1839) and analogous state trade secret law. The lawsuit requires ABC to establish the existence and misappropriation of a trade secret. XYZ denies the allegations and counters with claims for patent infringement under 35 U.S.C. § 271(a). The counterclaims require XYZ to establish that ABC made, sold, offered for sale, or used XYZ's patented facial recognition authentication system.
The parties could potentially assert other legal claims, including copyright infringement of source code and non-IP-related claims, such as unfair competition, inducement, and tortious interference (for information on copyright issues presented by the use and training of generative AI, see Generative AI and Copyright in the October 2024 issue of Practical Law The Journal; for more on legal issues counsel may face concerning AI-related technology, see AI Key Legal Issues: Overview (US) on Practical Law). However, this article focuses on patent infringement and trade secret misappropriation.
The figure below summarizes how facial recognition authentication systems generally operate:
As shown above, the system:
Extracts data from the camera's output signal (conducting a frame-by-frame analysis of streaming video data).
Locates faces in the image data through object detection (typically using a machine learning algorithm for image processing called a convolutional neural network).
Identifies unique features (for example, facial landmarks) from the faces that can be used to differentiate them.
Compares those features to the faces of people already known to identify a match.
Identification is possible if stored (known) facial data is matched to incoming feature data. The stored facial data comes from previously provided template face scans, which may be stored locally in the vehicle or possibly in an app carried by the driver into the vehicle and transmitted to the local computing platform.
Pre-Suit Investigation Considerations
Before filing a complaint or a responsive pleading that includes defenses and counterclaims, counsel should:
Perform a thorough pre-suit investigation.
Ensure that all pleadings (and all other court filings) comply with:
Federal Rule of Civil Procedure (FRCP) 11; and
relevant local court rules and case law.
Pre-Suit Investigation Legal Standards
FRCP 11 sets the standard for all representations made to the court, including those in pleadings. Specifically, FRCP 11(b)(3) requires that, before presenting any document to the court, counsel perform an 'inquiry reasonable under the circumstances' to ensure that any 'factual contentions have evidentiary support or, if specifically so identified, will likely have evidentiary support after a reasonable opportunity for further investigation or discovery.' Courts' local civil rules also often include an FRCP 11 analogue, such as the US District Court for the Eastern District of Texas's Local Rule 11(a), which requires designation of a lead attorney who is deemed responsible for the party's actions. In some respects, these rules impose on counsel a duty to perform some adequate amount of pre-suit investigation before filing a lawsuit.
Counsel or a party who fails to comply with FRCP 11(b) may face sanctions under FRCP 11(c). In patent cases, a court may also award attorneys' fees under 35 U.S.C. § 285 if it finds the plaintiff's pre-suit investigation to be inadequate (see Lumen View Tech. LLC v. Findthebest.com, Inc., 811 F.3d 479, 483 (Fed. Cir. 2016) (quoting Octane Fitness LLC v. ICON Health & Fitness, Inc., 572 U.S. 545, 555 (2014), which stated that 'a district court may award fees in the rare case in which a party's unreasonable conduct — while not necessarily independently sanctionable — is nonetheless so 'exceptional' as to justify an award of fees'); for more information, see Patent Litigation: Section 285 Attorney Fees Awards on Practical Law).
In trade secret cases, some state laws authorize courts to award attorneys' fees if a plaintiff brings a trade secret claim in bad faith (see, for example, MicroStrategy, Inc. v. Bus. Objects, S.A., 331 F. Supp. 2d 396, 430 (E.D. Va. 2004) (considering the Virginia Uniform Trade Secrets Act, Va. Code. Ann. § 59.1-338.1)). Counsel should therefore carefully consider the applicable local court rules and case law when preparing to file a lawsuit or responsive pleading.
Hypothetical Lawsuit: Pre-Suit Investigation
The adequacy and reasonableness of a pre-suit investigation in a typical IP-related lawsuit is assessed on a case-by-case basis. However, given the complexity, opacity, and learned nature of AI technologies, counsel involved in a lawsuit concerning AI technology should consider the pre-suit investigation a critical part of developing and drafting claims, defenses, and counterclaims. In the HAV facial recognition authentication system hypothetical, counsel's pre-suit investigation should include at least the following:
Before filing its complaint for trade secret misappropriation, ABC's counsel should:
evaluate whether its client has taken reasonable measures to keep its software and unique algorithms secret (18 U.S.C. § 1839(3));
consult with an economic expert to assess whether ABC's software and algorithms derive independent economic value, actual or potential, from not being generally known to, and not being readily ascertainable through proper means by, another person who can obtain economic value from the disclosure or use of the information (18 U.S.C. § 1839(3)); and
conduct a forensic analysis of ABC's trade secret information to understand the circumstances surrounding its possible theft or disclosure by the former engineers who now work at XYZ, including an in-depth forensic analysis of their emails, desktop and laptop computers, handheld electronic devices, and office files.
Before filing its answer to ABC's complaint, XYZ's counsel should:
determine the circumstances surrounding its hiring of ABC's former engineers, including evaluating software configuration changes commensurate with those hirings; and
assess whether the former ABC engineers 'knew or had reason to know that the trade secret was acquired through improper means, under circumstances giving rise to a duty to maintain the secrecy of the trade secret' (Ford Motor Co. v. Launch Tech Co., 2018 WL 1089276, at *16 (E.D. Mich. Feb. 26, 2018) (citing 18 U.S.C. § 1839(5))).
Before filing its counterclaims for patent infringement, XYZ's counsel should:
engage a technical expert to test ABC's AI-based facial recognition authentication system in a vehicle (or have the expert test it as a separate system, if available); and
determine if the AI system's operation can shed light on how the system works and if it can be reverse engineered.
Before answering XYZ's patent infringement counterclaims, ABC's counsel should:
review the known facts and key documents concerning its client's AI system; and
compare them to the asserted patents, prosecution histories, and prior art to assess the strength of XYZ's patent infringement counterclaims.
Counsel for both parties should engage independent consultants to help them identify evidentiary support for any factual contention asserted in their respective pleadings.
(For more on pre-suit considerations generally, see Patent Litigation: Pre-Suit Considerations and Trade Secrets Litigation on Practical Law.)
Litigation Hold Considerations
In addition to a pre-suit investigation, counsel must advise their client about taking steps to avoid spoliation of evidence as soon as the client either:
Is aware of the possibility of a lawsuit or legal action against it.
Intends to begin a lawsuit or take legal action.
A litigation hold generally applies to electronically stored information (ESI), but it also implicates hard-copy documents, equipment, and other things. In any IP lawsuit involving AI technology, a prompt litigation hold is critical because these cases typically involve source code versions and data that may be routinely updated or modified.
(For model patent litigation hold notices, with explanatory notes and drafting tips, see Patent Litigation: Litigation Hold Notice (Patent Owner) and Patent Litigation: Litigation Hold Notice (Accused Infringer) on Practical Law; for a collection of resources to help counsel implement a litigation hold generally, see Litigation Hold Toolkit on Practical Law.)
Litigation Hold Legal Standards
A party's duty to preserve evidence 'comes into being when the party has notice that the evidence is relevant to the litigation or should have known that the evidence may be relevant' (Guzman v. Jones, 804 F.3d 707, 713 (5th Cir. 2015)).
FRCP 37(e) provides for discretionary sanctions, including dismissal, if ESI that should have been preserved is lost 'because a party failed to take reasonable steps to preserve it, and it cannot be restored or replaced through additional discovery.' Spoliating evidence with the intent to deprive an adversary of use of the evidence in litigation constitutes bad faith conduct and may be sanctionable. Depending on the jurisdiction, a challenging party may need to demonstrate clear and convincing evidence of spoliation. (CAT3, LLC v. Black Lineage, Inc., 164 F. Supp. 3d 488, 499, 501 (S.D.N.Y. 2016).)
Hypothetical Lawsuit: Litigation Hold
ABC's counsel should advise ABC to begin a litigation hold as soon as it starts to evaluate the possibility of taking legal action against XYZ for trade secret misappropriation. ABC's counsel should also consider serving a litigation hold letter on ABC's former engineers who joined XYZ, in case ABC believes taking action against them individually is warranted.
If ABC and XYZ are in discussions about the merits of ABC's trade secret allegations before commencement of the lawsuit, XYZ's counsel should advise its client to institute a litigation hold even before receiving service of ABC's complaint.
Regarding the disputed AI technology, counsel for both parties should advise their respective clients to:
Freeze training and testing data sets that were and are being used for AI model development during the period of time at issue in the dispute.
Identify and preserve any new data points that have been added to training data sets (for example, where data was added to re-train AI algorithms to improve models).
Freeze all source code versions created during the period of time at issue in the dispute.
ABC's counsel should similarly implement a litigation hold concerning XYZ's patent infringement counterclaims as soon as it anticipates the counterclaims or promptly after receiving service of XYZ's answer.
(For more on implementing a litigation hold generally, see Implementing a Litigation Hold on Practical Law.)
Motion to Dismiss Considerations
At the early stages of a lawsuit, a party's pre-suit investigation and analysis of the opposing party's pleadings may help counsel assess the merits of and likelihood of prevailing on a motion to dismiss. FRCP 12, local court rules, and relevant case law provide the procedure for and standards by which courts address motions to dismiss.
(For more on motions to dismiss claims and counterclaims in patent litigation, see Patent Litigation: Motions to Dismiss on Practical Law.)
Motion to Dismiss Legal Standards
FRCP 12 sets out the procedures for moving to dismiss a party's complaint, either before or after the defendant has filed a responsive pleading. For example:
FRCP 12(b)(6) is often the basis for a motion before the close of pleadings. It requires the movant (typically the defendant) to establish that the respondent (typically the plaintiff) failed to state a claim on which relief can be granted.
FRCP 12(c) is used after the pleadings are closed. It requires the movant to demonstrate that it is entitled to judgment on the facts and evidence set out in the pleadings.
In patent litigation involving patents directed to computer-implemented technology, including software used in AI algorithms, defendants often move to dismiss on the basis that the asserted patent claims are abstract and therefore ineligible for patent protection under 35 U.S.C. § 101. The US Supreme Court recognized in Alice Corp. Pty. Ltd. v. CLS Bank International that a patent claim is ineligible under Section 101 when it is directed to one of three general patent-ineligible categories, that is, laws of nature, natural phenomena, and abstract ideas (573 U.S. 208, 216-17 (2014) (providing a framework for assessing abstract concepts under Section 101)). If the claim is ineligible under Section 101, Alice requires consideration of whether the particular elements of the claim, evaluated both individually and as an ordered combination, add enough to transform the nature of the claim into a patent-eligible invention (Elec. Power Grp., LLC v. Alstom S.A., 830 F.3d 1350, 1353 (Fed. Cir. 2016) (citing Alice, 573 U.S. at 217)).
(For a collection of resources on patent subject matter eligibility under 35 U.S.C. § 101, see Section 101 Patent Eligibility Toolkit on Practical Law; for a model motion to dismiss computer-implemented claims as directed to patent-ineligible subject matter, with explanatory notes and drafting tips, see Patent Litigation: Motion to Dismiss Computer-Implemented Claims as Invalid Under 35 U.S.C. § 101 (Patent-Ineligible Subject Matter) on Practical Law.)
In the trade secret misappropriation context, a defendant may move to dismiss on the basis that the trade secret holder has not described the subject matter of its trade secrets with sufficient particularity 'to separate it from matters of general knowledge in the trade or of special knowledge of those persons who are skilled in the trade, and to permit the defendant to ascertain at least the boundaries within which the secret lies' (Pellerin v. Honeywell Int'l, Inc., 877 F. Supp. 2d 983, 988 (S.D. Cal. 2012) (quoting Diodes, Inc. v. Franzen, 260 Cal. App. 2d 244, 253 (1968))). However, while some degree of particularity is required, at the pleading stage, plaintiffs can describe trade secret information in general terms so as not to publicize the secret itself (see Covenant Aviation Sec., LLC v. Berry, 15 F. Supp. 3d 813, 818 (N.D. Ill. 2014)).
(For more on trade secret litigation claims and defenses, see Trade Secrets Litigation on Practical Law.)
Hypothetical Lawsuit: Motion to Dismiss
ABC's counsel may consider moving to dismiss XYZ's patent infringement counterclaims under FRCP 12(b)(6) based on arguments that accused infringers have used in successful Alice -type Section 101 challenges, including that the patent claims asserted in XYZ's counterclaims:
Are directed to a mental process performed by a computer.
Are directed to the abstract concept of using mathematical algorithms to perform predictive analytics through data collection and analysis.
Are result-focused and so functional as to effectively cover any solution to an identified problem (that is, detecting the face of a person inside a vehicle).
Lack the specificity necessary to show how the claimed computer processor's operations differ from prior human methods and, therefore, the counterclaims are directed to an abstract idea rather than to a technological improvement.
Are directed to unpatentable mathematical formulas and algorithms divorced from any particular solution to a technological problem. (See Alice, 573 U.S. at 222-24 (citations omitted).)
In a DTSA or state trade secret matter, a plaintiff's pleading must avoid conclusory and generalized allegations of trade secret misappropriation, including vague timelines of alleged acts of misappropriation and uses of purported trade secrets. Therefore, XYZ's counsel should consider moving to dismiss where ABC's DTSA or state trade secret misappropriation claims:
Do not set out with sufficient particularity the metes and bounds of the allegedly misappropriated AI-based facial recognition authentication system software but instead only:
refer to the software by a trademark name (IVIEWDRIVE®); and
describe its general functionality.
Do not set out with sufficient particularity the specific algorithms allegedly used in XYZ's AI-based facial recognition authentication system, such as the type and characteristics of those algorithms (for example, a specific type of neural network).
Are factually baseless regarding XYZ's alleged use of ABC's trade secret algorithms because XYZ's facial recognition authentication system algorithms are different than ABC's alleged trade secret algorithms.
In the hypothetical lawsuit, ABC's counsel may also allege future misappropriation based on the inevitable disclosure of trade secrets doctrine regarding its former engineers (for more on the inevitable disclosure of trade secrets doctrine, see Trade Secrets Litigation on Practical Law).
Initial Disclosures Considerations and Hypothetical Lawsuit
If a lawsuit survives an early motion to dismiss, the parties will need to exchange their initial disclosures under FRCP 26, local court rules, and relevant case law, which provide the scope, content, and procedures for making initial disclosures. In a lawsuit involving an AI technology, counsel should carefully consider the possibility of having to make its source code available for inspection early in the case.
FRCP 26(a)(1)(A)(ii) requires production of a copy or a description by category and location of all documents, ESI, and tangible things that the disclosing party has in its possession, custody, or control and may use to support its claims or defenses, unless the use would be solely for impeachment. Depending on the jurisdiction in which the hypothetical lawsuit discussed above is pending, a party's initial core document disclosure may involve identifying or making available:
(For the complete version of this resource, which includes more on initial disclosures, including initial disclosures of individuals under FRCP 26(a)(1)(A)(i), see Artificial Intelligence: Intellectual Property Litigation Discovery Considerations: Overview on Practical Law.)
Document Production Considerations
Once fact discovery begins, counsel may exchange requests for the production of documents, interrogatories, and requests for admission.
(For a collection of resources to help counsel draft and respond to written discovery requests in patent litigation, see Patent Litigation Written Discovery Toolkit on Practical Law.)
Document Production Legal Standards
FRCP 26(b) governs the procedures and standards by which parties in civil litigation may obtain discovery. Unless limited by a court, parties may seek discovery of any non-privileged matter that is relevant to a party's claim or defense and is proportional to the needs of the case, taking into consideration:
The importance of the issues at stake in the action.
The amount in controversy.
The parties' relative access to relevant information.
The parties' resources.
The importance of the discovery in resolving the issues.
Whether the burden or expense of the proposed discovery outweighs its likely benefit.
In a lawsuit involving an AI technology, knowing how the AI system makes decisions and takes actions may be highly relevant to a party's case.
Evidence is relevant to a party's claim or defense if it tends to make the existence of any fact that is of consequence to the determination of the action more probable or less probable than it would be without the evidence (Fed. R. Evid. 401). Even if the information sought in discovery is relevant and proportional, discovery is not permitted where no need for the information is shown (see Am. Standard Inc. v. Pfizer Inc., 828 F.2d 734, 743 (Fed. Cir. 1987)).
Additionally, parties have an obligation to supplement their responses to document requests as relevant information is identified during the lawsuit.
Hypothetical Lawsuit: Document Requests
In a lawsuit involving an AI technology, knowing how the AI system makes decisions and takes actions may be highly relevant to a party's case. Assuming counsel for ABC and XYZ can justify a need for the information under FRCP 26(b), they should serve targeted discovery requests seeking the following information regarding the disputed AI system:
The data sets considered and used (both raw and processed).
Software information, including earlier and later versions of the contested version.
The software development processes used.
Sensors for collecting real-time observational data for use by the AI system.
Source code files.
Specifications describing what the system is and is supposed to do.
Schematics illustrating how components of the system work together.
Flow charts describing how percepts (data) are inputted, processed, and used.
Formulas, including those forming the basis for algorithms written in code.
Drawings, especially those that help elucidate how the system works.
Other AI system documents.
Interrogatories Considerations
Interrogatories are useful for a party to develop facts, discover the identity of additional witnesses, and identify the location of relevant documents (see Harris v. Escamilla, 2016 WL 1224057, at *6 (E.D. Cal. Mar. 29, 2016) (stating that 'the purpose of interrogatories is to expedite trial by narrowing and clarifying the issues and identifying potential witnesses or documents'); Alston v. Sharpe, 2015 WL 6395937, at *2 (D. Conn. Oct. 22, 2015)). In an AI lawsuit, counsel should try to use at least some interrogatories to focus on the data, the algorithms, model development, and model deployment (in both software and hardware) to understand the disputed AI system.
Interrogatories Legal Standards
FRCP 33 governs interrogatory practice and specifies:
Limits on the number of interrogatories (typically 25, including discrete subparts, and possibly more when sought by leave of court or agreed on by the parties).
The scope of interrogatories (that is, interrogatories must relate to a matter that may be inquired into under FRCP 26(b)).
How a party answers and raises objections to interrogatories.
How much time is permitted to respond to interrogatories (30 days by default).
The use of answers to interrogatories as evidence.
A party's optional production of business records in response to interrogatories if:
the answer to an interrogatory may be determined by examining, auditing, compiling, abstracting, or summarizing a party's business records (including ESI); and
the burden of deriving or ascertaining the answer will be substantially the same for either party.
That parties have an obligation to supplement their interrogatory responses as relevant information is identified during the lawsuit.
FRCP 37(d)(1)(A)(ii) gives federal courts discretion to sanction a party if, after being properly served with interrogatories under FRCP 33, the party 'fails to serve its answers, objections, or written response.'
Hypothetical Lawsuit: Interrogatories
ABC's counsel should consider serving interrogatories to identify:
Documents and information describing XYZ's facial recognition authentication system and source code to see if they contain any of ABC's code.
Documents and information relating to XYZ's hiring of ABC's former engineers.
Witnesses knowledgeable about XYZ's facial recognition authentication system and source code and hiring practices.
Likewise, XYZ's counsel should consider serving interrogatories seeking the identification of:
Documents and information describing ABC's facial recognition authentication system so that XYZ may better evaluate ABC's patent infringement or trade secret misappropriation claims and damages calculations.
Witnesses knowledgeable about ABC's system.
Counsel for both parties should also consider serving interrogatories to gather information concerning documents and witnesses who may be able to testify about:
The nature and extent of the testing conducted on the contested AI system before deployment.
How and why specific algorithms were selected for the facial recognition authentication system.
The modeled feature space used in developing the AI model and its relationship to the primary decision variables at issue.
The relevant scientific community for the technology at issue.
The relevant industry or technical standards applicable to the disputed AI system and its outputs (including accuracy, precision, and recall).
Because source code may be highly relevant in an IP lawsuit involving an AI system, counsel may also consider using interrogatories to identify:
Relevant software versions so that the applicable versions may be inspected.
Software development documents, which counsel may use in depositions to understand how the software was developed.
Documents regarding the various model inputs used, algorithm architectures selected and de-selected, parameters chosen, and hyperparameters derived for the various algorithms (for example, anything related to system development).
Discovering how an AI system is deployed in hardware may help reveal the underlying nature of the software. Therefore, counsel should consider serving interrogatories to identify:
All hardware used in a disputed AI system, such as in-cabin cameras for providing image data to the parties' respective facial recognition systems.
Documents describing hardware performance, capabilities, and limitations.
Witnesses with knowledge of why certain hardware was selected and its chosen placement, which could demonstrate copying.
Deposition Considerations
After counsel have obtained the AI-specific written discovery from the adverse party and generally understand the disputed AI system's source code, counsel should prepare to take depositions to fill remaining gaps in their understanding of the facts relevant to the parties' contentions. In a lawsuit involving a disputed AI technology, counsel should expect to depose general fact and expert witnesses as well as one or more source code custodians and other individuals involved in developing the disputed technology.
Deposition Legal Standards
Depositions by oral examination are governed by FRCP 30, local court rules, case management orders, judges' individual practices, and generally accepted norms in a particular jurisdiction. These rules of practice provide the procedures counsel must follow at a deposition, including:
Giving reasonable written notice of a deposition to opposing counsel.
Determining whether and which documents a witness should produce.
Recording the deposition.
Handling examination, cross-examination, and other matters (including raising objections on the record).
Party Witnesses
In an FRCP 30(b)(6) deposition, 'there is no distinction between the corporate representative and the corporation.' The corporate representative does not give their personal opinion during the deposition but instead 'testifies on behalf of the corporation and thus holds it accountable.' Companies must make a good faith effort to designate a knowledgeable person for an FRCP 30(b)(6) deposition and 'to prepare them to fully and unevasively answer questions about the designated subject matter.' (Sprint Commc'ns Co. v. Theglobe.com, Inc., 236 F.R.D. 524, 527 (D. Kan. 2006) (quotations and footnotes omitted); for more information, see Patent Litigation: Deposing a Rule 30(b)(6) Witness on Practical Law.)
Fact Witnesses
In a technical deposition of a fact witness, such as a data scientist, machine learning engineer, software engineer, or stack developer, investigating the algorithm behind an AI model will help answer questions about how and why a particular system caused a particular result that is material to the litigation. Counsel should therefore ask fact witnesses:
Which algorithms were considered.
Whether the algorithms were separately tested.
How the algorithms were tested.
Why the final algorithm was chosen.
Whether an independent third party reviewed the algorithm and model output.
Regarding the data set used to create the AI model, counsel should ask:
What data sets were used for training, validation, and testing of the algorithm.
How testing and validation were conducted and whether alternatives were considered.
What sort of exploratory data analysis was performed on the data set (or sets) to assess usability, quality, and implicit bias.
Whether the data was adequate for the domain that the developer was trying to model and whether other data has been used.
Regarding the final model, counsel may want to explore the following issues with fact witnesses:
How old the model is (that is, whether the model is based on old or incomplete data sets).
Whether the model represents a time series (for example, a model based on historical data that tends to increase over time) and, if it does, whether the underlying distribution shifted enough such that the model is now outdated.
Whether newer data was considered and, if not, why.
How accurate the model is and how its accuracy is measured.
If written discovery reveals that an independent third party reviewed the model before a party deployed it, counsel should explore the testing and its results. If sensors are used as the source for new observational data fed to an AI model, counsel should explore why those sensors were chosen, how they operate, their limitations, and what alternative sensors could have been used instead but were not selected.
Expert Witnesses
In an expert deposition, the goal of the deposition shifts to exploring the expert's assumptions, inputs, applications, outputs, and conclusions for weaknesses. If an expert prepared an adversarial or counterfactual model to dispute the contested AI system or an opposing expert's model, counsel should keep in mind the factors in Daubert v. Merrell Dow Pharmaceuticals, Inc. (509 U.S. 579 (1993)) and Federal Rule of Evidence 702 when deposing the expert. For example, counsel may need to explore the following issues during the expert deposition:
Whether an adversarial or counterfactual model was developed and why.
Whether the expert's analysis can be challenged objectively for reliability.
Whether the expert's technique was subject to peer review, publication, or both.
What the model's known or potential rate of error was when applied to facts relevant to the lawsuit.
What technical standards were applied to the analysis.
Whether the analysis was based on techniques or theories that have been generally accepted in the scientific community.
Hypothetical Lawsuit: Depositions
ABC's and XYZ's counsel should explore the above issues as a starting point during depositions of the parties' respective corporate representatives, other fact witnesses, and expert witnesses identified under FRCP 26 and 30. Similarly, if necessary, counsel should elicit deposition testimony on the above issues from third-party witnesses commanded to testify by subpoena.
For example, ABC's counsel should seek testimony with at least the following goals in mind:
Showing that XYZ's expert's testing of ABC's facial recognition authentication system for infringement purposes was not reliable.
Demonstrating that the patents asserted in the infringement counterclaims:
were disclosed in the prior art and therefore were anticipated or would have been obvious on a claim-element by claim-element basis; or
cover abstract ideas that are ineligible for patent protection under Section 101.
Establishing that XYZ's source code was modified around the time of its hiring of ABC's former engineers to incorporate specific algorithms that ABC has maintained as trade secrets.
Showing that XYZ moved its in-cabin camera to a location inside vehicles that would optimize its use of ABC's trade secret algorithms.
Demonstrating that ABC's former engineers who now work for XYZ will inevitably disclose to XYZ the trade secret algorithms in the IVIEWDRIVE® software that they misappropriated from ABC.
Establishing that XYZ could use different algorithms in its system and therefore would not be irreparably harmed by a permanent injunction.
XYZ's counsel should seek deposition testimony with at least the following goals in mind:
Demonstrating that ABC's facial recognition authentication system (both hardware and software), on a claim-element by claim-element basis:
performs every element of the asserted patent claims; and
works in the same way and produces the same result as the asserted patent claims.
Demonstrating that the asserted claimed inventions were not disclosed in the prior art and would not have been obvious to a person of ordinary skill in the art.
Establishing that ABC did not take reasonable measures to protect its trade secret algorithms and that the algorithms:
are in the public domain; or
could easily be reverse engineered using common knowledge.
Showing that ABC's trade secret algorithms are not present in XYZ's system (using a source code comparison).
Demonstrating that ABC's former engineers did not provide the IVIEWDRIVE® software to XYZ.
Establishing that ABC made and sold facial recognition authentication systems that infringed XYZ's patents to OEMs and Tier 1 manufacturers and therefore should pay a reasonable royalty.
Michael Burling joined Practical Law from Paul Hastings LLP, where he was a patent litigator in the life sciences group. Previously he was a patent litigator at Ropes & Gray LLP and at Fish & Neave LLP focusing on the pharmaceutical and biotechnology fields.