By now, most everyone understands that “artificial intelligence” does not mean human-looking robots bent on world domination. Despite decades of movies, books, TV shows, and other media depicting sentient computers as dangerous, the reality is that AI is an exceedingly valuable tool that can, in the right applications, all but eliminate errors while improving efficiency.
Still, the use of AI is not entirely without risk. Though generative AI platforms have only been available for a handful of years, AI has already altered business practices, causing a significant amount of change — both intended and unintended, both positive and negative. Experts and laypeople alike expect even more significant change as AI adoption becomes more widespread, and it is imperative that businesses strive to control that change, ensuring the greatest benefit to human society.
Most organizations implement AI with the expectation that the technology will benefit their bottom line, but IT operations need to make sure that AI technologies, especially those deployed in data archiving, are benefiting human workers and customers, as well. The key to adopting an AI application for data archiving in an ethical manner is to remain cognizant of the potential harms it may cause and take steps to mitigate those harms now and into the future.
Understanding AI
Archiving data is not a new concept for businesses; most companies have collected and utilized data for decades, so most IT teams already have various non-AI systems in place to ensure effective data storage. Thus, many IT leaders might be tempted to ignore the issue of AI entirely, which allows them to sidestep ethical concerns.
Unfortunately, this will prove all but impossible. In the dynamic landscape of data management, efficiency is king, and cutting-edge AI solutions promise to streamline the archiving process in myriad ways, not only increasing speed and reducing errors but also enhancing data storage systems. AI will be essential to data archiving going forward, so the sooner IT management adopts AI tools into their data archiving practices, the better.
Already, there are a handful of AI-driven tools that have revolutionized data archiving, becoming essential to enterprise data management systems. These include:
Natural language processing tools, which are machine learning applications that more accurately interpret human language. These can aid data management systems in understanding data context. The result is greater data accessibility across an organization, as non-IT workers can use natural language to search and retrieve relevant data without assistance from IT or data management teams.
Automated metadata tagging tools, which assign tags to data based on context derived through machine learning algorithms. Tags enhance the categorization process and make the storage and retrieval of relevant information more efficient, and the automatic creation of these tags allows IT professionals to focus on more complex tasks associated with data management.
Predictive archiving algorithms, which help IT teams plan for future data storage needs. By tracking historical archiving patterns and usage trends, these tools can indicate how much storage space will be necessary and when, so IT can make more strategic data management decisions as the business grows.
IT operations that have not already deployed these or similar AI solutions in data archiving are certain to do so soon. Yet, with the adoption of AI in data management comes the assumption of certain ethical risks — namely, bias and privacy.
Recognizing Bias
AI bias arises when an AI system produces results that are influenced by human preferences, preconceptions, or prejudices. Unfortunately, biases in AI most often reflect the biases already present in human society, which means that AI tools can work to perpetuate existing social inequalities. For businesses, biased AI tools offer fewer benefits; when AI-driven systems are limited in their understanding of or applicability to marginalized groups, like women, BIPOC, the queer community, or people with disabilities, they offer less potential for business growth.
There are two places within the development and implementation of AI where bias can occur:
Training Data
Perhaps most often, bias is found in the data used to train an AI tool. When some groups are over- or under-represented in the dataset or when the data is inconsistently labeled, the AI will be improperly trained. One of the most popular examples of this is in facial recognition tools, which have been less capable of reading the faces of people of color because they were trained predominantly with white faces.
Algorithm
AI algorithms that consistently generate skewed results are typically considered to be biased. Biased algorithms can arise due to a developer’s inability to identify or rectify biased training data, but algorithmic bias can also be the result of programming errors, such as a developer unfairly assigning weight in algorithm decision-making. A good example of this comes from applicant tracking systems, in which the AI algorithm tends to favor vocabulary most often used on men’s resumes, resulting in lower recruitment of women or non-binary applicants.
Bias can be difficult to detect and mitigate, even for AI experts. IT operations should strive to work with AI platforms that are transparent in their programming processes. Ideally, tools used in data archiving will be trained with large datasets that appropriately reflect the diversity contained in an organization’s data.
It might also be useful for IT to introduce a “human-in-the-loop” step, wherein an IT worker’s approval is required for certain automated tasks before the AI makes and acts upon a decision. Short of developing a bias-free AI tool in-house, IT teams limit bias by being aware of its possibility and acting quickly to intervene should signs of bias emerge.
Noting Privacy Concerns
Security should always be top of mind for IT operations responsible for data management. Business data is always under threat, either from thieves who want to sell personal information or from saboteurs who want to disrupt business operations — or both. Unfortunately, the use of AI tools can interfere with existing security precautions and put sensitive information at greater risk.
AI models rely on training data to learn how to behave properly, and that training data can include all manner of personal information, from names, phone numbers, and addresses to health records, banking information, Social Security numbers, and more. Plenty of AI systems allow users to elicit training data with a simple prompt, meaning anyone with access to the AI program may gain access to this valuable data.
Even if the data archive itself is protected with several layers of cybersecurity checks and balances, any AIs trained with a business’s data could become an easy target for a data breach. Online AI tools pose an even greater threat.
To ensure data privacy, security must be a priority at every level of development for AI-driven data archiving systems. When IT teams outsource AI tool development, this may be difficult to control. Again, transparency becomes essential. Firstly, AI developers should be transparent regarding how they secure sensitive data. Secondly, and arguably most importantly, businesses should be transparent with workers, customers, and any others whose data may be utilized in AI training and deployment.
IT may put systems in place that allow users to decide whether AI tools can have access to their personal information. Finally — and this should go without saying — IT needs a strong incident response plan in place to protect the entire data archive and related AI tools should an attack occur.
Accepting Responsibility
IT might be tempted to offload the responsibility of AI’s ethical considerations onto various other groups. It is easy to argue that AI developers shoulder the majority of the obligation to ensure AI does not perpetrate societal harm; indeed, developers have control over the data and algorithm of AI solutions, and they should be held accountable for any glaring bias or insecurity in their products. One might also suggest that business leaders should assume some blame for the wanton adoption of AI, pushing for the implementation of unvetted tools in pursuit of greater efficiency or lower costs.
However, IT operations do need to accept their responsibility in mitigating the negative ramifications of AI. IT workers typically have a greater understanding of the systems impacted by AI implementation and the risks that AI may pose to those systems. IT also has a greater capability to intervene if and when an AI goes wrong.
As IT adopts more AI solutions for its data archive or elsewhere, managers need to be certain that every worker is aware of the ethical considerations of the technology. Investment in strong ITOM practices can help ensure that IT remains watchful for the development of AI ethical concerns and is capable of preventing harm from befalling the company, customers, or society at large.
AI might not be a walking, talking, gun-wielding robot — yet — but in the wrong hands, it can still be dangerous. The public’s fears of AI are increasing, and if organizations want to benefit from the advantages of AI, they need to be respectful of the real ethical concerns associated with the technology. The larger data archives grow, the more imperative it is that IT teams take meaningful precautions to prevent harm caused by AI.