The involvement of AI in our daily activities is now more profound than ever. It is as simple as media applications like YouTube gathering data to suggest a playlist based on search and streaming history or shopping applications suggesting items based on purchase and search history. The adoption of AI in the security industry has also seen fast growth in unique automation, like the launch of countermeasures against missiles within a specific range, one of AI’s numerous uses. The positive changes AI brought to the defence system are undeniably impressive. However, there are lots of concerns when observed from an ethical point of view.
Understanding The Role Of AI Security In Surveillance Systems
Using the OODA Loop Model as our first case study, let’s explore ethical concerns regarding AI use in surveillance. Airforce pilot Col. John Boyd, the developer of the OODA loop model, created the model to help make a faster decision that prompts the opponent to act in a predictable way that ensures victory. However, there was no consideration of the possibility of the opponent being a machine. Understandably, machines have become so advanced that humans can’t outrun their processing power, making it eminent to commit some decision-making to machines, but at what point should humans take charge?
The reality remains that machines still have a long way to go before attaining that level of human complexity. However, the ethical concerns of AI security in surveillance systems have become a part of the forces responsible for AGI (Artificial General Intelligence) and a major dilemma for some researchers.
The application of AI to surveillance promotes better data analysis and pattern recognition, which sometimes boosts overconfidence in the ability of AI to predict crime without error. This misconception has become the failing point of many AI-powered surveillance systems. AI provides valuable insight into situations based on patterns and history, which invariably means AI can’t accurately foresee criminal activities even though it excels at handling vast amounts of data, which can help solve crime.
Rising Concerns Of AI Security In Surveillance Systems
- Lethal Autonomous Weapon Systems
Lethal Autonomous Weapon Systems are weapons systems designed with sensory abilities that make them environment conscious, allowing them to identify a target and decide on engaging it without human input. An example of its use was in the March 2020 event in Libya. An unmanned aerial system autonomously engaged human targets.
Although effective, it could not differentiate between people’s intentions. This particular concern remains one of the most difficult ones to eliminate because only some believe in achieving such a high level of sophistication where AI-driven weapons meant to enhance security can conveniently and thoroughly differentiate between friends and enemies regardless of the situation. Even those who believe in the possibility of achieving it don’t see it happening soon.
- Privacy ConcernÂ
The growing concern about AI infiltrating privacy prompted the development of several privacy protection laws and regulations, such as the GDPR (introduced in 2018 for EU regions) and the Privacy Act in Australia (Introduced in 1988). In response to privacy matters, China introduced strict laws on AI surveillance technology usage, which covers data collection, storage and use. Despite the regulations in place to solve the issues of privacy, there is still an unending debate about governments secretly harvesting citizens’ data and invading privacy in the most unimaginable ways.Â
- Cybersecurity Risks
Committing decision-making into the hands of an AI system isn’t such a bad idea, but one must wonder what could go wrong if the system gets hacked. Unlike home gadgets, organizational servers, or computers that hackers or cybercriminals only focus on secretly exporting data and carrying out attacks such as DOS or identity theft, hacking a weapon or surveillance system goes beyond data theft. It could lead to multiple casualties within seconds.
- Bias and Discrimination
AI algorithms used in surveillance systems can inherit biases in their training data. It can lead to discriminatory outcomes, disproportionately targeting certain groups based on race, gender, or socioeconomic status, amplifying societal inequalities. So, how credible are the data sources used in training the AI model?Â
- Lack of Transparency and Accountability
Making AI models simple and easier to understand is vital to promote trust and confidence. The use of an AI surveillance system requires more transparency. The public should know the purpose and scope of surveillance, which should be backed by a solid accountability mechanism to prevent data misuse.
Possible Solution To AI Security In Surveillance Systems
- Privacy by Design
This approach focuses on embedding privacy into an AI model’s design and development system right during development. Engaging privacy by design involves implementing privacy-enhancing measures at every stage of the AI lifecycle, which involves data collection, model development, data processing, user control and transparency, and regular assessments and updates. An example is China’s regulations implemented to enforce the consent requirement regarding data collection.
- Enhanced Cyber Defense
Cybercriminals are improving their Tactics. Just as artificial intelligence has been helped in our daily lives, it also has its downside when used in launching more sophisticated attacks. Therefore, organizations need to increase their investments in cybersecurity because of the ever-changing threat landscape. It will allow faster threat and vulnerability detection while improving protection.
- Eliminate Bias and Discrimination
Eliminating bias in an AI model is a continuous process involving training the model with data from credible sources. Facial recognition technologies require more attention in this aspect. One way to make this a success is by following the four-step procedure: Pre-processing, In-processing, Post-processing, regularization and fairness control.Â
The pre-processing stage involves adjusting the training data to reduce bias before feeding it into the model, the in-process deals with modifying the learning algorithms to decrease bias during model training, and the post-processing stage focuses on applying the corrections to the model’s output to mitigate bias after generation. The last stage, the regularization and fairness constraints, involves adding constraints to the model’s training process to ensure fairness and reduce prediction bias. The old saying remains the same: regularly updating the algorithm isn’t optional to ensure effectiveness.Â
Conclusion
Addressing these ethical implications requires a balance between the legitimate security concerns that may warrant surveillance and the protection of individuals’ rights and freedoms. Striking this balance necessitates ongoing discussions, ethical frameworks, and responsible deployment of AI technologies in surveillance systems.