Developing a framework for the ethics of AI-based lie detection in government

Abstract.

There is no end in sight for the development of AI lie detectors as long as decision-making models remain effective and decision-makers continue to find innovative uses for technology advancements. The lying machine may aid the handler in investigating different crimes like rape, theft, murder, and so on as backup evidence, but it cannot serve as the primary evidence in certain cases. AI technology may not be able to prevent relevant issues; in fact, if it is not controlled, the use of a lie machine may exacerbate issues related to privacy, security, and ethics. While crimes and deceit may be able to be detected, AI technology may not be able to prevent relevant issues. The government may become more efficient with the use of lying machines. This article proposes numerous ethical frameworks for the use of artificial intelligence that may facilitate the movement of services to digital platforms while also improving the lives of ordinary people controlled in society. These frameworks are suggested as a result of this article.
Keywords: Artificial intelligence (AI), ethics, framework, governance, lie detector, polygraph.

Table of Contents

1 Introduction 3
2 Ethnic and AI in lying detector 4
2.1 Privacy 6
2.2 Security 6
2.3 Ownership 7
3 Trust and governance 7
3.1 Explicability of the artificial lie system 8
3.2 Degree of professional competence 9
4 AI-governed framework for lying machines 9
4.1 Integrate AI as a citizen-centric programmed 9
4.2 Participation of citizens 9
4.3 Leverage existing resources 10
4.4. Be data-ready and apply privacy prudence 10
4.5 Reduce ethical hazards and eliminate AI bias in decision-making 10
5 Conclusion and Recommendation 11
References 11

1 Introduction
Today in the world we live in, individuals are proficient at lying without being caught or noticed. According to (Carson, 2006), lying is the oldest kind of deceit, as ancient as humanity itself. Lying may be defined as the act of making a false statement with the purpose of convincing another person to believe it. (Fallis,2010). Lying is classed as deceit, and some people utilize it while talking, making it harder for individuals, law enforcement, and the government to determine the truth when interacting with them. The invention of the lying machine made it possible to get the truth from individuals while conversing. In the past, traditional lie detection methods such as polygraph, perspiration and respiratory heart measurement, heartbeat sensor, and blood pressure monitor have been used .The most significant was polygraph. Augustus Larson devised the first polygraph, which continuously measured a subject's blood pressure, pulse, and respiration rate.
The Larson machine was first employed to investigate a theft case that occurred in a Berkeley women's dorm, and a year later it was used to convict a guy in San Francisco. Later, it was discovered that the Larson machine was incorrect due to the complexity of lying, (Benjamin,2013) since various individuals respond differently when they lie. In addition, the Larson machine was examining for changes in heart rate, sweat pore size, and muscular contraction in response to a query about the presence of such characteristics in anxious individual. The modern polygraph test has the same basic structure as the Larson, in which the examiner asks a series of questions to measure the subject's normal psychological state, while observing the machine transcribe these measurements as waveform lines on a page or screen. However, unlike the Larson, the modern polygraph test also includes digital scoring scores. Even while newer polygraph machines have been used to recruit employees in certain government agencies, they are still too sluggish to be employed in busy areas such as airports and borders. Additionally, it is difficult and needs professional operation, and its accuracy towards anxious individuals was not yet resolved. As a consequence, a new generation of lie detector machines based on artificial intelligence has been developed, which are said to be quicker, more user-friendly, and more accurate than polygraphs.
According to IEEE -USA defines AI as “The theory and development of computer systems that are able to perform tasks. That normally require human intelligence, such as visual perception, speech Recognition, learning, decision – marking, and natural language processing.”. In order to determine if a person is speaking the truth or not, a system with artificial intelligence was created to analyses the facial traits or emotional state of a speaker. The market's first artificial machine was the silent talker. The team teamed with a health-care NGO in Tanzania to capture the facial expressions of eighty women while they completed online courses on HIV treatment and condom usage. The team's 2012 research was the first to demonstrate the Silent Talker technology in the field. The objective was to establish if patients understood the therapy they would get; as stated in the study's introduction, "the evaluation of participants' understanding throughout the informed consent process continues to be a major problem." When the researchers cross-referenced the AI's predictions about whether the ladies comprehended the lectures with their marks on quick post-lecture examinations, they discovered that the AI was 80% accurate in predicting who would pass and who would fail. I border ctrl also sponsored the trial. I border control is an automated European Union border security system. According to study conducted by silent walker on a group of Tanzanian women to determine who understood and did not understand the HIV treatment at the health care facility. Three essential concerns must be asked about the research:
• Were the Ethiopian women aware that they were utilized in a sociological experiment?
• What would be terrible about the twenty percent that Silent Walker failed to predict?
• Will I Border CTRL implement the precision of the quiet talker AT EU border?
Is the artificial model for detecting lying accurate?
There has been much debate about the dependability (also known as validity) of utilizing artificial intelligence in lie detector tests for a very long time. Theoretically, this circumstance offers a dilemma since there is no proof that a particular pattern of physiological responses is exclusively related with dishonesty. A person who is honest may experience anxiety when asked a question that requires an honest response, but a dishonest person would not feel this way.
Widespread opinion is that a person who is speaking the truth is more likely to avoid control questions than to answer critical ones. This is due to the fact that control questions are meant to make a subject feel anxious about their prior honesty, but relevant questions ask a subject about a crime they are aware they did not commit. Consequently, this consequence arises. Depending on the results of the inquiry, a diagnosis of "deception" may be made if there is a pattern in which a person's physiological reaction is stronger to important questions than to control questions. If the responses to the control questions are positive rather than negative, it will be determined that no dishonesty occurred. Test results are considered "inconclusive" if they demonstrate no discernible differences between the important questions as well as the control questions.
2 Ethic and AI in lying detector
What Since the advent of artificial intelligence, ethics of how to use A.I. and the possible inclination of employing A.I. have been a key worry that has been expressed in many ways? Ethics continues to guarantee doing the right thing. This leaves a specification of the proper data collecting and machine-learning data use. Also, A.I. is still restricted by the creator's intelligence, therefore the accuracy of the A.I. machine might be questioned as to whether it is acceptable for usage. Imagine if a wearable gadget could also capture continuous data on facial expressions. Imagine if you could use this data to analyses your everyday discussions and interactions, repeating questionable ones more carefully. Imagine a friend or company utilizing your past data to discriminate between your truths and falsehoods, significant and trivial objects, and things you care about and don't care about. This gives investigators, advertisers, the cautious, the paranoid, vigilantes, and everyone with Internet access a new weaponry. Each of us will need to maintain this new data-driven public record of our responses. The urge to know what someone is thinking, what they feel, what they will do, and what everything means to them is a recurring motif in films. We all know the world isn't perfect, but internet help is available. What happens when counsel is reinforced by a torrent of poorly understood data? What will happen if this new data is used in the hiring process and software is used to detect whether candidates lied during the interview? What happens if the same procedure is used for school admissions, jury selection, and other interviews, or if the results are shared with employers? As such situations proliferate, we must consider when our pulse becomes private. Is knowledge of our internal reactions fundamentally private? considering that until recently only a few sensitive people could tell. Communities typically choose the path of least resistance, creating a divide between those who can navigate the new digital record and others who can't. Imagine therapists actively recording cognitive dissonance, news programmers assessing in real time whether a guest believes what they're saying, and companies re-framing interviews using active face analysis. Rising sensor capabilities herald in a post-lying age, or the end of our comfort with lying. As always, the benefits won't be evenly distributed. We might see lie detection advance toward brain-computer interface, in which case the right to privacy must address when our thoughts are private. If we can reliably discern between lying and truth feelings in courtrooms, should witnesses keep this knowledge private? These technologies may change the courtroom's nature. It's unreliable, thus witnesses aren't offered polygraphs. With a portable analytic device, someone may determine their vitals or analyses a video stream remotely, then publish the findings. How should past behaviour be interpreted? Social norms allow us to hide information about ourselves and others when we develop nudges, establish public locations, and negotiate social situations, job offers, and personal relationships. What should we do with technology that reveals hidden data? Is a world based on facts better than ours? Will we vote? Advances in AI and the democratization of data science make the hypothetical question of what kind of society we want all-too-real and urgent. Data has been the subject of several literary works since the 1900s. As more corporate sectors embraced AI and big data, the number rose. Due to the numerous times data breaches have wreaked havoc in the financial, educational, and healthcare sectors, these works address big data and the ethics of maintaining personal data. In 2018, the EU General Data Protection Regulation (GDPR) addressed data gathering, storage, and processing. The 2018 UK Data Protection Act accompanies the GDPR. (DPA). The GDPR functions seven principles which are listed below;
• Lawfulness, fairness and transparency
• Purpose limitation
• Data minimization
• Accuracy
• Storage limitation
• Integrity and confidentiality (security)
• Accountability
These guiding principles are now firmly established as the standards that will be used to direct the development of big data, most notably artificial intelligence and analytic s based on AI. The General Data Protection Regulation (GDPR) and the Data Protection Act (DPA), which are applicable to all other economic, social, and private organizations, are also relevant to all government parastatals in their deployment of AI technology in providing services to residents. The primary concerns of citizens who are bothered by AI technology and the way data is handled by the government are similar to those that have been observed in other fields. These concerns have been categorized into four categories, and they are as follows: privacy, ownership, security, and AI biases (White and Ariyachandra, 2016; Kerr et al., 2020).
2.1 Privacy
Because of social norms, it is feasible for us to hide information about ourselves as well as information about other people when we design nudges, establish public forums, and negotiate social situations, job offers, and personal relationships. Which course of action is best for us to do with the technology that uncovers top-secret information? Is it possible that a world based on facts may be a better place to live than the one in which we now find ourselves? Will we vote? The question of what kind of society we want to live in is no longer a theoretical one as a result of the advent of artificial intelligence (AI) and the democratization of data science. This has pushed the topic into harsh and urgent reality. Imagine if therapists actively recorded cognitive dissonance, if news shows could tell in real time if a guest believed what they were saying, and if companies could use active face analysis to change the way interviews were framed. Now, it's possible to do all of these things. With the development of sensor technology comes the start of a new era, which means the end of an era in which it was easy to lie. As is usually the case, the benefits won't be shared fairly. In the near future, brain-computer interface could be used to find out if someone is telling a lie. If this happens, then the right to privacy needs to say when our thoughts are private.
2.2 Security
One of the biggest fears people may have while dealing with AI and data gotten from lie machine is data breach. If data are not sufficiently protected, they may get into the hands of unauthorized parties who may exploit them for discriminatory or harmful purposes (White and Ariyachandra, 2016). Citizens are likely to assume that insufficient data protection might result in instability and injustice in the distribution of social and economic interventions within communities (Henman, 2005). The creation of enormous quantities of data has been aided and hastened by the development of AI technology. The probability of data breaches grows as the quantity of data increases (Ronzhyn and Wimmer, 2019). Data obtained from several sources and stored in enormous data banks or clouds are often at risk of being hacked or breached due to insufficient security measures or because it is generally difficult to protect big data due to its size (Lyon, 2014). Therefore, the government must develop ways to gain the public's confidence in its ability to appropriately protect their data.
2.3 Ownership
People desire to feel in control of their personal information and details, and they may be possessive about how the government or commercial organizations acquire and utilize it. At each level of data collection and use, the objective must be explained and the persons impacted must consent (Olumoye, 2013) Big data comes from several sources, making it challenging to handle and monitor (White and Ariyachan-dra, 2016). Government agencies and companies muzzle dissent by asking people to sign a disclosure agreement that transfers data ownership and future alterations to them (Tene and Polonetsky, 2013). People feel they own their data and should be free to choose what companies or governments do with it (Olumoye; 2013;) White and Ariyachandra, 2016). AI systems are developed with predictive skills to analyze human behavior and form conclusions using in-built algorithms (White and Ariyachandra, 2016). It was assumed that computer algorithm-based assessments might lead to uneven deployment of development approaches and incentives (Tene and Polonetsky,2013). Compartmentalization based on pre-supplied information may harm disadvantaged populations. People who know how to fill out paperwork on a computer may have an edge. The bulk of the responsibility for doing the right thing falls on the data analyst's ability and desire to follow laws and ethical guidelines (Kerr et al., 2020).
.
3 Trust and governance
Data Lee and See (2004) defined trust as "the attitude that an agent will assist attain an individual's aims" Trust is comparable yet different with robots and AI systems. When a person trusts another, they expose themselves to the other's actions. When someone trusts an AI agent, it's unclear whether the computer is making its own judgement or following a script. Companies would have to create system that would be trusted by the public, find useful, and can pay for. Businesses have a reason to make systems that people can trust. Trust is made up of a very complicated set of ideas. To trust a system, users must be sure that the system will help them or do good things for them. That is, it will treat them with kindness. They will need to know that the system won't hurt them or hurt their interests in some other way (e.g., breach their privacy or cause them embarrassment). People will need to believe that the system can be fair within the limits of what it can do. This is much easier to do if there are clear rules and laws and no moral disagreements. At the moment, these functional scopes are small because people don't agree on what moral theory is. Even so, there are many real-world applications that have moral implications and can be handled well by AI. People are unconcerned about responsibility when nothing goes wrong and no one is injured. This idea is only relevant when something goes wrong. Some harm is done, people are wounded, stuff is taken. Then, they want to know who is accountable, who is held accountable, and who is liable for damages or compensation. What applies to the physical world also applies to AI. Only when these technologies produce difficulties will it be appropriate to hold someone or anything accountable or liable. And these difficulties may be substantial, i.e., problems that are experienced not only in a single instance, but often and systematically. What does "go bad" in the context of AI technology entail?
3.1 Explicability of the artificial lie system
Explicability isn’t about transparency or maximum openness of AI algorithms and codes may not solve problems and may create new ones. Why make millions of lines of code transparent? First, even experts would have trouble grasping the program's purpose. Second, software transparency might threaten competition and discourage investment. Due of these variables, some argue over "explicability. ( Floridi et al,2018) say explicability involves understandability and accountability. In moral applications, those using AI systems or whose interests are influenced by AI systems want to "understand" how an AI made a certain decision. Intelligibility means humans can understand AI's processes. The system's inner workings aren't mysterious. Even a master developer can explain the system to judges, juries, and users. The EU enacted the "right to information" as part of the GDPR (formerly known as the "right to explanation"). Those whose interests are affected by an automated judgement might request an explanation. This is problematic for "incomprehensible" machine learning approaches like neural networks. Some don't mind machine learning's "inscrutability." They think it can be tested. As long as it works in practice, they don't care if they can't explain it (Weinberger 2018). This may be adequate for machine learning. In ethically delicate situations, it may be required to defend the decision. Justification may be needed. Moral functioning involves doing the right thing and defending it. Moral justification can't be "inscrutable." "Explainable AI" study seeks to explain neural networks' conclusions (Wachter et al., 2017). Such research may help machine learning justify its results. Despite this, the court has tried COMPAS for estimating recidivism risk and affecting probation. In Loomis v. Wisconsin, the plaintiff claimed he was deprived "due process" due to the proprietary nature of the COMPAS algorithm, which precluded his defense from questioning its scientific basis. His appeals failed. The courts ruled that COMPAS risk rating don't impact sentencing. Judges may utilize risk ratings with other factors to determine recidivism risk.
3.2 Degree of professional competence
The polygraph examination, which is also known as a lie detector test, is one of the most intriguing techniques used in the domains of criminal justice and criminology. However, it is also one of the least understood. Even if the tests are founded on straightforward scientific concepts, it is nevertheless impossible for anybody to immediately start asking questions after strapping a subject into an instrument and doing the exam. Polygraph examiners are specialists who have received extensive training and are responsible for conducting lie detector tests. (Jeffrey et al 2016). There are some factors the examiner must consider, such as the patient's condition of health. The security of the location where the result has been determined, including the ability to refuse gifts from the examiner. The result's confidentiality shall not leak or be shared with a third party.
4 AI-governed framework for lying machines
in Framework for synthetic liar detectors in ethics in Governance. It is necessary for the government to in-still confidence in its population to guarantee the seamless deployment of disruptive technologies such as AI technology; otherwise, it would always be challenging to make constructive changes in society via these methods (Kerr et al., 2020). Consequently, the following tactics, which are described in the sub-sections below, may be effective for ensuring the successful application of AI technology.
4.1 Integrate AI as a citizen-centric programmed
The deployment should be a practical way to solve problems that people can see, and it shouldn't be done just as a formality without clear laws (Mehr, 2017). Measures like training for the officers in charge of lying machine must have been taken to set up the way for it to be done. It should also be proven that using artificial intelligence in lie detectors is the best way to deal with and try to solve the problems people have (Mehr, 2017; Kerr et al.,2020). Sensitizations can also be used to help people get used to the procedures as they are used. Also, it must be made sure that lying machine interactions are demographically inclusive and can take into account people from different socioeconomic groups (Mehr, 2017).
4.2 Participation of citizens
During the introduction of artificial intelligence in lie detectors, channels for public input should be established by assigning participation responsibilities to representatives selected by citizens (Mehr, 2017). The citizen's representative may explain the notion of artificial intelligence in lie detectors to the general public without raising suspicions of a hidden goal. Aside from this, the government may organize a public think-tank session or a public conference where direct questions about deployment and operations can be addressed and answered in order to educate the public and allay their anxieties over the abuse of artificial intelligence in lie detectors. Keeping these channels of citizen input accessible will assist AI technology get the finest insights and user preferences (Mehr, 2017).
4.3 Leverage existing resources
Rather than being a whole new programmed, artificial intelligence in lie detectors should be an inventive change in detecting deception (Mehr, 2017). In other words, artificial intelligence in lie detectors should help in the development of processes from a more sophisticated way to one that is simpler, hence facilitating the execution of tasks and processes (Bearden, 2014; Chow-White et al., 2015). Also, it may be advantageous to consolidate comparable jobs on the same AI platform in order to reduce indifference caused by the repetition of identical activities for various reasons (Mehr, 2017).
4.4. Be data-ready and apply privacy prudence
When using artificial intelligence in lie detectors, it should be of the utmost importance to equip all relevant agencies for data collecting, storage, and analysis. Handlers should be taught on data administration, the kind of data that must be gathered, and the manner in which such data should be used while training the lie detector system (Mehr, 2017). The duration of data storage and the date on which it will be deleted should be explicitly defined, and different quality controls should be implemented to identify errant employees. During the deployment of artificial intelligence in lie detectors (Mehr artificial intelligence in lie detectors), the government should maintain procedural openness to prevent any kind of reaction.

4.5 Reduce ethical hazards and eliminate AI bias in decision-making
Due to its training, the lie detector's AI is biassed (Davis, 2016). Since this bias affects the trainer and everyone involved in its implementation, a multidisciplinary team should train and deploy lie detector AI (Chessen, 2017; Mehr, 2017). Before the lie detector is released, AI-using ethicists should test it. When employed in decision-making, lie detector AI should be human-supervised to reduce ethical risks. Artificial intelligence analyses and proposals for lie-detection systems should not be used without human validations (Mehr, 2017). AI should be a complement to human problem-solving efforts, not a solution in and of itself due to prejudice and ethical problems (White and Ariyachandra, 2016).
5 Conclusion and Recommendation
AI and lying machines can transform government. Secondary evidence may help detect lying. As an exploitable technology, the GDPR must regulate its use. If the government builds on existing platforms and adopts ethical approaches, AI might speed up governance development. Training is needed to utilise a laying machine effectively. As technology changes, the operator should get ongoing training and instruction. To be ethical, candidates should be able to justify their result that happen
During the analyization of their testing .

References

1. Acono, W.G. and Lykken, D.T., 1997. The validity of the lie detector: Two surveys of scientific opinion. Journal of Applied Psychology, 82(3), p.426
2. Baker, P. and Sunday, M., 2002. Polygraphs, Renewing Suspicions. Washington Post
3. Chin, C.W., 1989. Protecting Employees and Neglecting Technology Assessment: The Employee Polygraph Protection Act of 1988. Brook. L. Rev., 55, p.1315.
4. Cios, K. and William Moore, G., 2002. Uniqueness of medical data mining. Artificial Intelligence in Medicine, 26(1-2), pp.1-24.
5. Cios, K. and William Moore, G., 2002. Uniqueness of medical data mining. Artificial Intelligence in Medicine, 26(1-2), pp.1-24.
6. Crawford, C.C., 2002. The polygraph in agent interrogation. Polygraph, 30, pp.28-32.
7. Eysenck, H.J., 1971. Relation between intelligence and personality. Perceptual and motor skills, 32(2), pp.637-638.
8. Faigman, D.L., Fienberg, S.E. and Stern, P.C., 2003. The limits of the polygraph. Issues in Science and Technology, 20(1), pp.40-46.
9. Gaggioli, A., 2018. Beyond the truth machine: emerging technologies for lie detection. Cyberpsychology, Behavior, and Social Networking, 21(2), pp.144-144.
10. Gershman, B.L., 1998. Lie Detection: The Supreme Court's Polygraph Decision. New York State Bar Journal, 70, p.34.
11. Handler, M., Nelson, R. and Blalock, B., 2008. A focused polygraph technique for PCSOT and law enforcement screening programs. Polygraph, 37(2), pp.100-111.
12. Hinkle, C., 2020. The Modern Lie Detector: AI-Powered Affect Screening and the Employee Polygraph Protection Act (EPPA). Geo. LJ, 109, p.1201.
13. Honts, C.R. and Reavy, R., 2015. The comparison question polygraph test: A contrast of methods and scoring. Physiology & behavior, 143, pp.15-26.
14. Horvath, F., 1977. The effect of selected variables on interpretation of polygraph records. Journal of Applied Psychology, 62(2), p.127.
15. Horvath, F., 1977. The effect of selected variables on interpretation of polygraph records. Journal of Applied Psychology, 62(2), p.127.
16. Kircher, J.C., Packard, T., Bell, B.G. and Bernhardt, P.C., 2001. Effects of prior demonstrations of polygraph accuracy on outcomes of probable-lie and directed-lie polygraph tests. UTAH UNIV SALT LAKE CITY.
17. Lansang, M. and Umpierrez, G., 2008. Management of Inpatient Hyperglycemia in Noncritically Ill Patients. Diabetes Spectrum, 21(4), pp.248-255.
18. Levetan, C., Passaro, M., Jablonski, K., Kass, M. and Ratner, R., 1998. Unrecognized Diabetes Among Hospitalized Patients. Diabetes Care, 21(2), pp.246-249.
19. Lykken, D.T., 1978. Uses and abuses of the polygraph. In Psychology: From research to practice (pp. 171-191). Springer, Boston, MA
20. Nelson, R., Handler, M. and Morgan, C., 2012. Criterion validity of the Directed Lie Screening Test and the Empirical Scoring System with inexperienced examiners and non-naive examinees in a laboratory setting. Polygraph, 41(3), pp.176-185.
21. Nowrojee, B., 1993. Divide and rule: State-sponsored ethnic violence in Kenya (Vol. 3169, No. 102). Human Rights Watch
22. Pittas, A., Siegel, R. and Lau, J., 2004. Insulin Therapy for Critically Ill Hospitalized Patients. Archives of Internal Medicine, 164(18), p.2005.
23. Raskin, D.C. and Honts, C.R., 2002. The comparison question test.
24. Rowe, W.B., Yan, L., Inasaki, I. and Malkin, S., 1994. Applications of artificial intelligence in grinding. CIRP annals, 43(2), pp.521-531.
25. Segrave, K., 2003. Lie detectors: A social history. McFarland.
26. Siegelaar, S., Hoekstra, J. and DeVries, J., 2011. Special considerations for the diabetic patient in the ICU; targets for treatment and risks of hypoglycaemia. Best Practice & Research Clinical Endocrinology & Metabolism, 25(5), pp.825-834..
27. Tricco, A., Ivers, N., Grimshaw, J., Moher, D., Turner, L., Galipeau, J., Halperin, I., Vachon, B., Ramsay, T., Manns, B., Tonelli, M. and Shojania, K., 2012. Effectiveness of quality improvement strategies on the management of diabetes: a systematic review and meta-analysis. The Lancet, 379(9833), pp.2252-2261.
28. Umpierrez, G., Isaacs, S., Bazargan, N., You, X., Thaler, L. and Kitabchi, A., 2002. Hyperglycemia: An Independent Marker of In-Hospital Mortality in Patients with Undiagnosed Diabetes. The Journal of Clinical Endocrinology & Metabolism, 87(3), pp.978-982.
29. Vinik, R. and Clements, J., 2011. Management of the Hyperglycemic Inpatient: Tips, Tools, and Protocols for the Clinician. Hospital Practice, 39(2), pp.40-46.
30. Widacki, J. and Szuba-Boroń, A., 2017. Polygraph Examinations of Civil Servants in Poland.