In April of 2021, Governor Whitmer created a task force to recommend improvements to the use of forensic science. Her executive order said:
Wrongful convictions deprive innocent people of their freedom, deny closure to victims of crime, and undermine faith in the criminal justice system. Studies have demonstrated that the misapplication of forensic science is the second most common contributing factor in wrongful convictions in the United States. . . . It is vital that the State of Michigan and its courts rigorously adhere to best practices for the use of forensic science within the criminal justice system.1
The task force was co-chaired by the chief justice of the Michigan Supreme Court and the director of the Michigan State Police. After a year and a half of study, the task force delivered its report to the governor in December of 2022.2
The task force made a series of recommendations to both the Supreme Court and the legislature to improve the quality of forensic science in Michigan. The legislative recommendations include formation of a statewide forensic science commission, accreditation and registration of agencies and experts, steps to improve the independence of laboratory personnel, and others.3 Recommendations to the Supreme Court focused on forensic science education for lawyers and judges and changes to Court Rules and the Rules of Evidence.4 One of the most important of those Supreme Court recommendations was the amendment of MRE 702 to strengthen the requirement for judges to conduct meaningful hearings on the admissibility of forensic science evidence.
The governor is right about the relationship of faulty forensic science to wrongful convictions of the innocent. Nationally, 25% of wrongful convictions are based on false or faulty forensic science evidence, and 32% of sexual assault wrongful convictions are based on bad forensic science evidence.5 In 2021, there were 11 new exonerations in Michigan, and eight of them involved false or misleading forensic science evidence.6 In 2022, Michigan was second in the nation with 16 exonerations, 10 of those wrongful convictions involved false or misleading forensic science evidence, and nine of those were murder convictions.7 It is important to remember that wrongful convictions not only incarcerate innocent persons but also mean that the actual perpetrators of the crime remain free to possibly commit more crimes.
How and why is such faulty evidence admitted in criminal trials? A basic premise of our trial process is that the judge must decide what evidence is admissible for the jury to hear. Judges are truly the gatekeepers of all forms of evidence. In the 1993 Daubert case,8 the U.S. Supreme Court made it very clear that judges are gatekeepers of forensic science evidence and that judges are required to determine if offered expert evidence is “scientifically valid and . . . whether that reasoning or methodology properly can be applied to the facts in issue.”9 Federal Rule of Evidence 702 was subsequently amended to incorporate the Daubert requirements.
Michigan is a “Daubert state.”10 MRE 702 was amended to correspond to the Daubert modified FRE 702. In Gilbert v DaimlerChrysler Corp, the Michigan Supreme Court said: “We now clarify that MRE 702 requires the trial court to ensure that each aspect of an expert witness’s proffered testimony — including the data underlying the expert’s theories and the methodology by which the expert draws conclusions from that data — is reliable.”11 The Court went on to say that “MRE 702 has since been amended explicitly to incorporate Daubert’s standards of reliability. . . . It has not altered the court’s fundamental duty of ensuring that all expert opinion testimony — regardless of whether the testimony is based on ‘novel’ science — is reliable.”12
In Daubert, Justice Blackmun said that the Court was confident that judges “possess the capacity to undertake this review.”13 Unfortunately, whether judges are able to do so depends on whether they are willing to do so. The reality of the wrongful conviction data is that many judges have not effectively performed their gatekeeping responsibility. Many judges avoid holding a Daubert hearing, even when requested by the defense, and routinely admit almost all prosecution-proffered forensic science evidence. Appellate courts rarely reverse a conviction based on faulty forensic science evidence, even when it is challenged on appeal. In many criminal cases, judges have overwhelmingly failed to conduct hearings to determine the reliability of most traditional forms of forensic science. Many have abdicated their gatekeeping duty and continued to admit whatever evidence the prosecution offers.14 Even if a hearing is conducted, many judges continue to routinely admit forensic science evidence just because it has “always” been admitted. Some even reverse the burden of proof and require the defense to show why traditionally admitted evidence is now inadmissible.15
The failure of judges to conduct meaningful Daubert hearings was well documented by scholars.16 Many judges simply refused to consider or weigh challenges to types of evidence that has been routinely admitted in the past.17 Civil lawyers urged an amendment to FRE 702 to try to get judges to hold meaningful hearings, stating, “The evidence demonstrates the need for an amendment clarifying that Rule 702 requires courts to find that the rule’s admissibility requirements are established by a preponderance of the evidence prior to admitting expert evidence.”18 Eventually, these problems were examined by the United States Judicial Conference, the organization that advises the U.S. Supreme Court about federal rules.19 They recommended, and the Supreme Court and Congress adopted, the following amendments to FRE 702:
A witness who is qualified as an expert by knowledge, skill, experience, training, or education may testify in the form of an opinion or otherwise if the proponent has demonstrated to the court that it is more likely than not that:
- a) the expert’s scientific, technical, or other specialized knowledge will help the trier of fact to understand the evidence or to determine a fact in issue;
- b) the testimony is based on sufficient facts or data;
- c) the testimony is the product of reliable principles and methods; and
- d) opinion reflects a reliable application of the principles and methods to the facts of the case.
The amendments became effective in federal court on December 1, 2023.20 The State Bar of Michigan’s Michigan Rules of Evidence 702/703 Workgroup recommended that MRE 702 be amended to substantially incorporate the federal amendments.21 The amended MRE 702 was adopted by the Michigan Supreme Court, effective on May 1, 2024.22
The goal of these changes is intended to be a clear message to judges. First, they must fulfill their duty to conduct hearings to determine the reliability of the science underlying any proposed expert testimony. That reliability must be demonstrated “to the court,” and admissibility is not a matter of “weight” to be left to the jury.23 Second, the burden of proving the reliability of that science and its application to the facts of the case is on “the proponent.”24 Prior admission of similar testimony by courts does not somehow reverse the burden of proof. Third, the burden on the proponent of proving reliability and applicability is the traditional preponderance standard of “more likely than not.”25
It may be correct that, as the SBM work group said, these are not substantive changes to the law and they are just reinforcing the mandates of Daubert. But, as the work group went on to say, “To the extent one purpose of the rule amendment is to prod courts to remember their gatekeeping functions, the salutary function of the rule is unobjectionable.”26 Nevertheless, it is a prod that is clearly needed. Judges have an affirmative duty to close the gate on the admission of types of evidence that have led to the wrongful conviction of innocent people.
And now, the use of artificial intelligence (AI) to generate forensic science evidence has complicated the gatekeeping duty of judges even further. Many judges are justifiably skeptical of the legal uses of AI because of its well-documented proclivity to manufacture, or “hallucinate,” case citations. There is even a database documenting AI hallucination in legal cases, and the number of examples has already exceeded 400.27 Concern over such inventions and the problem of “deep fakes,” however, is not limited to legal applications. AI machine learning is transforming the production of forensic science evidence.28 AI is based on programmed algorithms that can process and interpret vast amounts of data. That makes it ideal for forensic science applications. It is being used for digital forensics analysis, biometric identification such as facial recognition, DNA analysis, crime scene reconstruction, and many other forensic science functions.29
But how are judges supposed to decide if the results of such machine-produced conclusions should be admitted into evidence? The existing rules of evidence did not seem to address this new form of forensic science information. In response to this issue, the federal court rules are being amended again. The Judicial Conference of the U.S. has recommended adoption of a new Federal Rule of Evidence 707.30 The new rule states:
Rule 707. Machine-Generated Evidence
When machine-generated evidence is offered without an expert witness and would be subject to Rule 702 if testified to by a witness, the court may admit the evidence only if it satisfies the requirements of Rule 702 (a)-(d). This rule does not apply to the output of basic scientific instruments.
The stated rationale for the new rule by the committee explains that “This newly adopted rule is designed to address growing concerns around the use of AI-generated evidence, particularly when such evidence functions similarly to expert testimony and raises parallel questions of reliability, bias, error, and interpretability.”31
The result of this new Rule 707, together with the amended Rule 702, is that courts must now apply the Daubert criteria to AI-generated evidence in the same way as any other expert evidence.32 The proponent of the AI-generated evidence must establish that the AI evidence meets the Daubert requirement of testing, error rates, peer review, standards, and general acceptance. This may prove difficult for several reasons. The courts may require, especially in criminal cases, disclosure of the source code used to produce the algorithms for the AI evidence, and AI commercial providers may be reluctant to comply. Most problematic is the lack of validation for many of the AI tools being used in laboratories.33
FRE 707 is open for public comment until February 16, 2026. If it is finally adopted, as expected, it would probably take effect later that year. The Michigan Supreme Court will then decide if it should be replicated in the Michigan Rules. It should be, and Michigan judges should begin to actively exercise their duty to ensure that the forensic science evidence the jury hears is truly valid and reliable.