Meuwly is not the ‘CSI’ examiner who searches crime scenes, but his work is certainly driven by real-life criminal cases. At NFI, he is responsible for the scientific quality of casework, research and education of forensic practitioners. He also serves as an advisor in complex and serious forensic cases, including several international terrorist cases. His research at the UT focuses on the use of biometric traces such as fingerprints, speech and face recognition.
No magic numbers
With this expertise and a degree in criminalistics and criminology, the scientist explains that the world of forensic science is more intricate than people might think. ‘As forensic scientists we cannot provide ‘yes’ or ‘no’ answers. There are no magic numbers or results. We also don’t make any decisions. With automated methods, we provide quantitative data, but it always has to be combined with the human-based method of the forensic examiner. Forensic science offers a probabilistic evaluation of evidence.’
‘Let’s say someone left a fingerprint on a bottle. But did they use the bottle to drink from it or to beat someone?’
What should we imagine under ‘probabilistic evaluation’? ‘In a criminal case, the prosecution and the defense have opposite explanations about different aspects of the case,’ replies Meuwly. ‘For example, who is the speaker of an audio recording can be disputed or the activity that led to a fingermark can be disputed. Let’s say someone left a fingerprint on a bottle. But did they use the bottle to drink from it or to beat someone?’ That is where the scientist and his colleagues come in, providing evaluation of evidence in terms of a ‘likelihood ratio’. ‘We are not paid for provide the strongest results, but the most correct strength of the evidence.’
Didier Meuwly’s field of expertise is digitized material. ‘In forensic science we differentiate between trace material collected physically, like DNA with a swab, and trace material that is digitized, like the face or the voice captured by a camera or a microphone respectively. This includes mainly fingermarks, speech recordings and facial recognition. However, we can also work with information like gait – how people walk -, which you can often see from CCTV footage. What we work with depends on the type of traces collected at the crime scene.’
‘Criminals don’t tend to look and smile at CCTV cameras’
At the UT, the scientist conducts research into automated fingermark, face and speaker recognition. More specifically, he and his PhD students look into the validation and calibration of casework methods. ‘We test and validate the developed methods both generically and case by case. For example, the facial recognition technology used to examine crime scene recordings is completely different than the one embedded in our phones. Criminals don’t tend to look and smile at CCTV cameras. Trace material can be of very bad quality and our aim is to assign them the most correct probative value, which can be low. On average, the strength of evidence in speech and face recognition are still rather low compared to fingermarks or DNA traces, but methods develop all the time. When I started in the field, we couldn’t use landline recordings for speech recognition. Now we can use low quality mobile phone recordings thanks to the development of artificial intelligence and machine learning.’
Didier Meuwly studied criminalistics and criminology at the University of Lausanne in Switzerland. From 2002 to 2004 he worked as a senior forensic scientist at the R&D department of the Forensic Science Service, then an executive agency of the British Ministry of the Interior. Now he shares his time between The Netherlands Forensic Institute (NFI), an agency of the Dutch Ministry of Justice and Security, and the University of Twente, where he holds the chair of Forensic Biometrics. As one of the principal scientists of the NFI, his activities focus on the scientific quality of the forensic casework and research. He advises the NFI about forensic research strategy, programs and projects. He is also a member of the R&D standing Committee of the European Network of Forensic Science Institutes.
He has been the chair of Forensic Biometrics at the UT since 2013. His research focuses on the use of physical and digital biometric traces, in particular the fingermark, speech, face, alone and combined. He specializes in the automation and validation of the probabilistic investigation and evaluation of these traces. He also contributes to the forensic education in the Netherlands and Belgium, at the Dutch Police Academy and the University of Brussels.
Latest technological developments make the job easier and more complex at the same time. Take wearables, for example. ‘Wearable sensors like a smart watch capture digital information about the activity of the wearer. They can inform about what the wearer was doing at a specific time, but their data is often encrypted and can only be accessed in particular conditions specified in the GDPR.’ If obtained, they are very useful, though. ‘There was a recent case in Greece when a man was found tied at home with his wife killed. He said three men tied him up and killed his wife, but both his mobile phone and the smartwatch of his wife were contradicting his testimony: he was walking at the time he said he was tied up and his wife was dead when he pretended she was still alive.’
‘How do you explain results from a complex machine to nonscientists in court?’
Artificial intelligence (AI) is another method that could make the work of forensic examiners more accurate and efficient. Yet, it could also make criminal cases difficult to resolve. ‘I’m currently working on a research article about explaining and validating AI for its forensic use,’ says Meuwly. ‘People argue that AI is a sort of black box that nobody truly understands, and so it cannot be used for evaluating evidence at court. But all sophisticated modern technology is a black box at first! You need to validate it, show that it is reproducible and statistically valid; test-drive it and then pass it on to the examiner who can validate it in practice. Still, the question remains: how do you explain results from a complex machine to nonscientists in court? These methods have a great potential to provide better answers, but also carry the risk to mislead if used improperly.’
Science and logic
The scientist’s main goal is improving the practice and developing methods that are accepted by forensic examiners and the courts of justice. ‘My ultimate hope is that the following generation of forensic scientists will be able to use the results of this research as a trustworthy step to further improve the forensic practice,’ says Meuwly. ‘When performed properly, the forensic practice is an important auxiliary of criminal justice, providing independent material evidence based on science and logic.’