ChatGPT: Forensic,legal,and ethical issues

Introduction and background

ChatGPT is a large language model developed using reinforcement modelling and launched by OpenAI on 30 November 2022. ¹ This new AI chatbot is being programmed to respond to queries with specific details as well as execute prompts, and it is capable of producing conversations that seem similar to those among actual human beings. Within 5 days of its inception, ChatGPT had amassed one million users, owing to how simple its interface was to use and how quickly it could respond with comprehensive responses. ² Users have employed this chatbot for writing emails, letters, essays, business proposals, academics, research and other type of writing. The ChatGPT is capable of generating detailed answers and replies within a matter of seconds. However, the information provided by the chatbot may not always be authentic and accurate, and so it cannot be relied upon. Moreover, the ChatGPT is a potential cyber security threat, and threat to national security. Recently, the European Union Agency of law enforcement, Europol warned about the rising concern of ChatGPT in cyber frauds, phishing attempts and criminal abuse.^3,4 Europol also mentioned that ChatGPT can be used as a weapon by cybercriminals to breach the security of the people and the nation. The report highlighted that the phishing scams are frequently used to acquire sensitive information, including login credentials, financial information or personal details. With ChatGPT-powered bots, large-scale spam operations may be carried out or malware that compromises victims’ networks and security can be disseminated.^4,5 Fake ChatGPT apps are uploaded by the hackers on google play store and third-party app stores can steal the passwords and personal information of people from smartphones and spread the malware. ⁶

ChatGPT itself suffered a data breach, exposing the personal information of ChatGPT plus members who were active during that nine-hour window on 20 March 2023. The sources claimed that the breached data embraced the first and last names, email addresses, payment addresses, credit card types, credit card expiration dates, last four digits, etc., of the active users. OpenAI estimated that the Redis client open-source library bug (bug means a coding error in a computer programme) was responsible for the data leak of about 1.2% of the active ChatGPT plus subscribers. ⁷ The Italian Data Protection Authority has taken serious note of escalating privacy concerns and illegal data collection, and consequently, Italy has become the first country to outlaw the ChatGPT. ⁸

Regardless of these privacy concerns, certain internet-based resources nowadays encourage and specify the implementation of ChatGPT in the arena of digital forensics.^9,10 As a consequence, the authors have decided to examine the authenticity of ChatGPT in the existing forensic investigation methodologies. Keeping this in consideration, the primary objective of the current communication was to investigate how ChatGPT addresses questions and concerns about forensic science investigations to figure out whether or not it can be useful to law enforcement agencies, medical professionals, forensic experts, etc. To get at the final inferences, a suitable strategy was decided upon, and a number of attempts were executed. In order to help forensic investigators and others, this communication would authenticate the reliability and preciseness of the responses offered by ChatGPT.

Methodology

The authors executed certain experiments to evaluate the competency of ChatGPT in multiple scenarios of crime scene investigation. A series of experiments were carried out about hacking, phishing, bloodstain evidence, fingerprints, drowning, arson and other aspects. Further, multiple questions and searches have been raised both directly and indirectly to analyse the ChatGPT response pattern, such as how to remove blood stains, fingerprints, etc. After examining the pattern of response from ChatGPT, two separate experiments are put forth in this communication. In the first experiment, the ChatGPT was examined in relation to conducting an investigation into a road accident. Moreover, in the second experiment, an issue of criminal psychology – specifically, how someone or a criminal may use this technology to dodge the law and punishment – was presented. In the subsequent section, there will be a discussion about the findings and recommendations that emerged from these experiments.

Observations

Experiment 1

The first experiment was carried out to examine whether ChatGPT responds to queries involving the investigation process at various crime scene locations. The question, ‘How to conduct a crime scene investigation in a road accident?’ was asked to ChatGPT in this experiment. However, the response from ChatGPT was rather brief, universal and insufficient (Figure 1). The ChatGPT mentioned incomplete evidentiary details that might be present on the vehicle and on the victim's body. The earliest opportunity must be deployed to cautiously and systematically acquire every bit of information. Furthermore, the steps outlined by the ChatGPT lacked numerous details that are crucial in a thorough investigation of a road accident, including speed and its estimation, skid marks, visibility, curved scuffmarks and so forth. Therefore, the responses provided by the ChatGPT cannot be considered accurate and authentic in investigation of a road accident. Furthermore, when the ChatGPT was prompted for the list of references for the content generated on crime scene investigation in a road accident, five references were provided (Figure 2). All the references generated by the ChatGPT are either incorrect or do not exist in the literature. The weblinks available in the references were not found. Then the search was carried out by the source titles mentioned for each reference generated. In the first, third, fourth and fifth references, the publishing year was not found as well as no such source/article/pdf was found at the official website listed by the ChatGPT. While ChatGPT incorrectly indicated that the book in the second reference could be retrieved from the Federal Bureau of Investigation (FBI), the book in the reference is actually available from the US Department of Justice. Therefore, it can be concluded that the information and the sources provided by the ChatGPT are not reliable, authentic, accurate or comprehensive. Meanwhile, the information provided by the ChatGPT depends on the selection of words and terms used by the users.

OPEN IN VIEWER

Figure 1.

Steps of crime scene investigation listed by ChatGPT for a road accident in a very brief manner.

OPEN IN VIEWER

Figure 2.

List of references generated by ChatGPT for crime scene investigation for a road accident.

Experiment 2

The second experiment addressed the issue of criminal psychology, in particular, whether a criminal or someone else can leverage ChatGPT to assist them in executing a criminal event. In this experiment, questions were posed to determine how the ChatGPT would react if a person with criminal intent or a criminal seeking to commit an offence was in search of a way to avoid being apprehended by the police. Moreover, an individual with such a criminal psychological mind may probably seek help through online sources rather than directly asking from someone. As a consequence of this, the authors sent a number of queries on crime, evidence and investigation to ChatGPT. But we only discussed one experiment here. It has been observed that ChatGPT does not answer when the questions were asked directly related to a crime, law and offence. We asked ChatGPT ‘how to clear fingerprints from a dagger used as a murder weapon?’ (Figure 3). But the same question when modified and asked indirectly, then the ChatGPT listed various suggestions; the ChatGPT was asked ‘how to clear fingerprints from a dagger?’ (Figure 4). We really appreciate the intelligence of ChatGPT that it did not provide any solution when the questions were asked directly related to the words ‘murder’, ‘crime’, ‘illegal activities’, ‘offence’, etc. The chatbot is very well trained, however, sometimes, the ChatGPT even answered some related questions directly also. It means if we consider from a criminal's point of view, a criminal may try any possible means to escape from the law and punishment. It could be anything whether seeking help from any source, person or internet. But a culprit who is looking for solutions before and after committing a crime may search internet and even may be more comfortable asking questions to an advance AI technology such as the ChatGPT. We observed that the reverse psychology works with the ChatGPT as well as its pattern of answering the offensive questions.

OPEN IN VIEWER

Figure 3.

No suggestions were obtained when the ChatGPT was asked directly using offensive words such as ‘murder’, ‘crime’, ‘offence’ or other illegal activities’, etc.

OPEN IN VIEWER

Figure 4.

Response of the ChatGPT on the same query without use of offensive words; here, the ChatGPT explains several methods to clear the fingerprints from a dagger.

Criminals may find it easier to talk to chatbots than to study dedicated literature since, with the right questions, they can quickly be provided with information on every facet of a crime scene investigation. In other words, depending on the requirements and inquiries of the criminal, such chatbots may both help a person commit a crime and may also offer a way to elude the police.

Discussion

We are all surrounded by technologies in our routine activities. We have become so much dependent on technologies that instead of thinking, analysing through our brains, we want solutions through search engines. The present era is witnessing the advent of advanced tools with more potential and intelligence. The launch of ChatGPT has brought numerous opportunities for humans; the chatbots give answer to every possible question. People stopped thinking and reading the books. Moreover, after the launch of ChatGPT, GPT-4 has also been launched within 2–3 months on 14 March 2023. GPT-4 accepts the input in the form of texts as well as images. This chatbot is well-trained in different languages including Hindi, Bengali, Telugu, Marathi and Gujarati. Recently, till April 2023, more than 200 AI tools have been launched.

The ChatGPT can be used for multiple purposes depending upon the user's need. It can be used to get business ideas, streamline the writing process, compose poetry, lectures, project proposals, academic and school assignments, etc. This has become actually a boon for the people. But it can be used in a negative sense as well, that is, as a support tool to aid the criminal activity. After conducting some experiments, we decoded that reverse psychology works with the ChatGPT for asking certain types of questions. Additionally, the ChatGPT provides very general responses to questions about crime scene investigation. Because of this, it cannot be used in the field of forensic science for investigating purposes, evidence collection, analysis, etc.

The chatbot raised ethical concerns when a New York-based lawyer was caught for using the ChatGPT for legal research. Non-existence of the several cases were found by the court, which had been cited as references by the lawyer in an ongoing case. So, several ethical concerns are associated with this chatbot of artificial intelligence (AI) leading to the spread of misinformation and bias. It has already been mentioned that ChatGPT provides the information based on the data that it has been trained. If the trained data includes misinformation and a biased dataset, the output obtained from this chatbot will also be biased and inaccurate. ¹¹ Apart from this, ChatGPT has also falsely reported a fake news article that a US-based Law professor sexually harassed a student. Such a news article was never published in real, but it was only generated by the ChatGPT when it was asked to list the five instances of sexual harassment at American Law Schools by professors. This is a negative aspect of AI chatbot for providing fake replies in a convincing manner, that is, ‘hallucinations’. ¹²

With this communication, the authors want to draw the attention of the authorities to setting up a strict regulatory framework and guidelines on AI-assisted tools. It also needs to be considered that such a technology cannot take the place of human intellect. There are some disciplines that can only work with human intelligence such as the criminal investigation process, law proceedings and various fields of forensics and medical sciences. The expertise needed in these domains can only be acquired by someone who has received the proper training, is highly competent, and has amassed the necessary experience to manage the practicality of the field, as opposed to chatbots, which are algorithm-based and trained on the data for a specific period of time.

Although, AI is doing wonders in the advancement of forensic science, such as with the advent of advanced software for facial reconstruction and recognition, drone forensics, robotic spies, image and video detection, audio-video forensic analysis, etc. But the use of AI-based technologies is restricted to some extent and there is a need to take a pause on AI so that authenticity, correctness and comprehensiveness may not be questioned. Recently, A book named ‘ChatGPT and Legalities’ ¹³ was released by Dr Pavan Duggal, an Indian Supreme Court advocate and an authority on cyber law, cyber security law, and emerging technology law. He advised suspending AI-based research, saying that right now is the ideal moment to establish legislative frameworks to govern AI. His book's primary goal is to advance legal jurisprudence on AI.

There is an urgent need to think about framing guidelines for the rational use and practices of AI technology. For this, some international organisations such as INTERPOL (the International Criminal Police Organization), ¹⁴ the British Academy of Forensic Sciences, ¹⁵ the American Academy of Forensic Sciences ¹⁶ and others should step up and look into the positive and negative aspects of AI-based tools while framing some guiding principles regarding ethical and legal concerns. Any new AI technology must be subject to scrutiny before it is released, and it must also adhere to strict procedures and rules designed to ensure that these technologies are used thoughtfully.

Conclusion and recommendations

Through this communication, the authors advocate certain specific ethical and legal standards for coherent and rational use of the AI tools like the ChatGPT. Without a doubt, the ChatGPT has many advantages and very helpful in many circumstances where one might need immediate assistance. Even the most internet-savvy person may need this chatbot's assistance in a number of ways, so it seems only natural that anyone should also ask it frequent questions. However, the experiments conducted on the chatbot have shown that it cannot be relied upon as authoritative source of information. The ChatGPT has been trained on the data available till a specific period of time. Furthermore, people are well aware that the information found online might be false, misleading, or not always accurate. As a result, there is a risk involved in accepting the information supplied by the ChatGPT. The authors do not advocate putting all the AI-assisted technologies on hold. It is a fact that AI has contributed a lot to the advancement of forensic science and provided much help in crime scene investigation. The authors recommend framing the ethical and legal guidelines to prevent the misuse of AI technologies and spreading misinformation in the professional fields such as Law, Forensic science, Medical Sciences, etc. Therefore, the role that AI has played in enabling cutting-edge technologies to resolve various criminal cases can be valued. But to prevent cross-violation, there is also a need to set up a boundary between AI and ethical conduct.