"This book is in the AAFS Reference Series Library and serves a guide on using AI in forensic science. In Zeno's year of Presidency of the AAFS, one of the topics was Artificial Intelligence and the impact it can have on forensic science. In many forensic products Artificial Intelligence and Deep Learning is already included without the user being aware of it. Examples are software for facial and speaker comparison, many digital forensic packages for searching on for instance firearms. Furthermore, it isupcoming in chemometrics and many other fields. Also, the use of AI can impact forensic science, for instance on easier to make deepfakes, so spoofing evidence becomes easier. Different jurisdictions will handle the use of AI differently, depending on the laws. Examples are provided of good use of artificial intelligence, where the expert should be in the loop. The expert as well as the courts also needs to know the limitations of the approach. The book is composed out of chapters which can be used in a course, and will finalize with the newest research in developing approaches with graph neural networks. The book first does an introduction in the field as such, then it will go deeper in the legal issues with AI and the need for developing standard. Manyexamples of using AI are discussed and presented, such as smart cities, IoT in Hansken and at the end Also, the use of AI for making deepfakes will be discussed as well as how to detect it. The field develops rapidly, and much awareness is also being made by the European Commission on regulation of AI in practical use. In the United States NIST is working on standards on using Artificial Intelligence. New legislation is expected in many states, for example Colorado has legislation on use of facial recognition services and Vermont has legislation enacted on ethical use of artificial intelligence. One of the most significant applications of AI in forensic science is in the analysis of DNA evidence. DNA sequencing technology has advanced significantly in recent years, allowing scientists to analyze large amounts of genetic data quickly and accurately. However, the sheer volume of data generated by these techniques can be overwhelming for human analysts. AI algorithms can be used to quickly and accurately identify genetic markers that are associated with specific individuals or groups, making it easier for forensic scientists to identify suspects or eliminate suspects from an investigation"--
ARTIFICIAL INTELLIGENCE (AI) IN FORENSIC SCIENCES
Foundational text for teaching and learning within the field of Artificial Intelligence (AI) as it applies to forensic science
Artificial Intelligence (AI) in Forensic Sciences presents an overview of the state-of-the-art applications of Artificial Intelligence within Forensic Science, covering issues with validation and new crimes that use AI; issues with triage, preselection, identification, argumentation and explain ability; demonstrating uses of AI in forensic science; and providing discussions on bias when using AI.
The text discusses the challenges for the legal presentation of AI data and interpretation and offers solutions to this problem while addressing broader practical and emerging issues in a growing area of interest in forensics. It builds on key developing areas of focus in academic and government research, providing an authoritative and well-researched perspective.
Compiled by two highly qualified editors with significant experience in the field, and part of the Wiley — AAFS series ‘Forensic Science in Focus’, Artificial Intelligence (AI) in Forensic Sciences includes information on:
- Cyber IoT, fundamentals on AI in forensic science, speaker and facial comparison, and deepfake detection
- Digital-based evidence creation, 3D and AI, interoperability of standards, and forensic audio and speech analysis
- Text analysis, video and multimedia analytics, reliability, privacy, network forensics, intelligence operations, argumentation support in court, and case applications
- Identification of genetic markers, current state and federal legislation with regards to AI, and forensics and fingerprint analysis
Providing comprehensive coverage of the subject, Artificial Intelligence (AI) in Forensic Sciences is an essential advanced text for final year undergraduates and master’s students in forensic science, as well as universities teaching forensics (police, IT security, digital science and engineering), forensic product vendors and governmental and cyber security agencies.
About the editors, ix
List of Contributors, x
Series Preface, xi
Preface Book, xii
Acknowledgements, xiii
1 Introduction, 1
Zeno Geradts and Katrin Franke
2 AI-based Forensic Evaluation in Court: The Desirability of Explanation and
the Necessity of Validation, 3
Rolf J.F. Ypma, Daniel Ramos, and Didier Meuwly
2.1 Introduction, 3
2.1.1 AI for Forensic Evaluation, 6
2.2 The Desirability for Explanation and the Necessity of Validation, 7
2.3 Explainability (and its Validity), 8
2.3.1 Reasons to Pursue Explanations, 9
2.3.2 Types of Explanations, 9
2.3.3 Limitations of Explanations, 11
2.4 Validation (and its Explanation), 11
2.4.1 Measure the Method's Performance, 12
2.4.2 Approach in Four Steps, 12
2.4.3 Accountability, 16
2.5 Conclusion, 17
3 Machine Learning for Evidence in Criminal Proceedings: Techno-legal
Challenges for Reliability Assurance, 21
Radina Stoykova, Jeanne Mifsud Bonnici, and Katrin Franke
3.1 Introduction: AI in the Intersection of Criminal Procedure and
Forensics, 21
3.1.1 Technical Fragmentation in Digital Investigations, 21
3.1.2 Legal and Methodological Fragmentation in Digital Investigations, 22
3.1.3 Specifics of ML-based Investigative Approach, 23
3.1.4 Scope and Definitions, 25
3.2 Legal Framework, 27
3.2.1 The Fair Trial Principle, 28
3.2.2 Necessity and Proportionality of Investigative Measures, 32
3.2.3 The AIA Proposal, 33
3.2.4 AI System Development and Legislative Contradictions, 35
3.3 Machine Learning Pipelines: Techno-legal Challenges, 44
3.3.1 Task + Purpose Limitation and Data Minimization, 44
3.3.2 Dataset Engineering and Data Governance, 48
3.3.3 Pre-processing for Input: Trade-offs between Accuracy and
Computational Costs, 53
3.3.4 Modelling, 56
3.4 AI Use in Investigations: AI System Design + Data Protection = Fair
Trial?, 63
3.5 Conclusion, 66
4 Formalising Representation and Interpretation of Digital Evidence to
Reinforce Reasoning and Automated Analysis, 74
Eoghan Casey and Timothy Bollé
4.1 Introduction, 74
4.2 Background and Related Work, 76
4.3 Method, 77
4.4 Representing Digital Traces, 79
4.5 Representing Computed Similarity, 86
4.6 Representing ML Classification, 89
4.7 Representing Hypothesis Test Results (a.k.a. Inferences), 91
4.7.1 Location Example, 93
4.7.2 Identification Example, 95
4.8 Effective/Reliable/Responsible Automated Analysis, 99
4.9 Conclusion, 101
5 Servicing Digital Investigations with Artificial Intelligence, 103
Harm van Beek and Hans Henseler
5.1 Introduction, 103
5.2 Introduction To Hansken, 104
5.2.1 Normalized Trace Model, 105
5.2.2 Forensic Tool Application, 106
5.2.3 Hansken's Application Programming Interfaces, 108
5.3 Large Scale Application of AI Techniques, 109
5.3.1 Rule-based AI Techniques Implemented in Hansken, 109
5.3.2 Deep-learning AI Techniques Currently Implemented in Hansken, 111
5.3.3 Deep-learning AI Techniques to be Implemented in Hansken, 115
5.3.4 The application of large language models in digital forensics, 118
5.4 Conclusions and Further Reading, 120
6 On the Feasibility of Social Network Analysis Methods for Investigating
Large-scale Criminal Networks, 123
Jan William Johnsen and Katrin Franke
6.1 Introduction, 123
6.2 Previous Work, 125
6.3 Material and Methods, 127
6.3.1 Real-world Underground Forum Database Dumps, 127
6.3.2 Network Centrality Measures, 129
6.3.3 Measuring Association Using Bi-variate Analysis, 129
6.3.4 Topic Modelling Algorithms, 130
6.4 Experimental Setup, 130
6.4.1 Evaluating Network Centrality Measures for Forensics, 130
6.4.2 Our Novel Approach for Analysing Cybercriminal's Technical Skills,
133
6.5 Experimental Results and Discussion, 137
6.5.1 Correlation Testing, 137
6.5.2 Our Newly Proposed Method, 142
6.6 Conclusion, 145
7 Mapping NLP Techniques to Investigations and Investigative Interviews,
149
Kyle Porter and Bente Skattør
7.1 Introduction, 149
7.2 Criminal Investigation, 150
7.2.1 Investigative Interviews, 150
7.3 Assessing the Needs of Investigators in an NLP Context, 151
7.3.1 Mapping Interviewer Needs to Existing NLP Tasks, 151
7.4 Automatic Speech Recognition, 152
7.4.1 ASR Basics, 152
7.4.2 ASR, Digital Investigation, and the State of the Art, 153
7.5 NLP Basics, 154
7.5.1 Common Terminology, 154
7.5.2 Vector Space Models and Embeddings, 156
7.5.3 Modern NLP Models, 157
7.6 Text Extraction, 157
7.6.1 Entity Identification and Named Entity Recognition, 157
7.6.2 Named Entity Recognition Metrics, 158
7.6.3 NER Applied to Investigations, 159
7.6.4 Entity Linking, 159
7.6.5 Limitations of Using NER, 160
7.6.6 Extraction Methods outside NER, 161
7.7 Text Classification, 161
7.7.1 Classification Evaluation Metrics, 162
7.7.2 Text Classification and Digital Investigation, 162
7.7.3 Classification Limitations, 163
7.8 Text Reduction, 164
7.8.1 Thematic Extraction and Topic Modelling, 164
7.8.2 Topic Modelling and Digital Investigations, 165
7.8.3 Limitations of Topic Modelling, 166
7.8.4 Text Summarization, 166
7.8.5 Text Summarization and Digital Investigations, 167
7.8.6 Summarization Limitations, 167
7.9 Discussion and Conclusion, 167
7.9.1 Future Work, 169
8 The Influence of Compression on the Detection of Deepfake Videos, 174
Meike Kombrink and Zeno Geradts
8.1 Introduction, 174
8.2 Method, 178
8.2.1 Dataset, 178
8.2.2 Deepfake Detection, 180
8.3 Results, 183
8.3.1 Compressed Dataset, 183
8.3.2 Algorithms, 184
8.4 Discussion, 190
8.4.1 Deepfake Detection, 190
8.4.2 Compression, 191
8.4.3 Future Work, 193
8.5 Conclusion, 193
9 Event Log Analysis and Correlation: A Digital Forensic Perspective, 195
Neminath Hubballi and Pratibha Khandait
9.1 Introduction, 195
9.2 Sources of Logs, 197
9.2.1 End Host System Logs, 198
9.2.2 Networking Devices and Security Applications, 203
9.2.3 Application Logs, 207
9.3 Need for Correlation, 208
9.4 Correlation Techniques, 210
9.5 Conclusions, 214
10 (Hyper-)graph Analysis and its Application in Forensics, 216
Marcel Worring
10.1 Introduction, 216
10.2 Survey of Methods, 218
10.2.1 Preliminaries, 218
10.2.2 Tasks, 219
10.2.3 Graph Neural Networks, 220
10.3 Explainability and Visualization, 224
10.4 Conclusion, 227
11 Conclusion, 230
Zeno Geradts and Katrin Franke
Index, 232
Edited by
Zeno Geradts, is a senior forensic scientist at the Forensic Digital Biometrics Traces Department at the Netherlands Forensic Institute, Ministry of Justice and Security, The Hague, The Netherlands.
Katrin Franke is Professor of Computer Science at the Department of Information Security and Communication Technology at NTNU in Gjøvik, Norway. She has over 25 years experience in basic and applied research for financial services and law enforcement agencies (LEAs), working closely with banks and LEAs in Europe, North America and Asia.
