- detail
- Published: Wednesday, July 12, 2023 09:55
Peter Ruffley explores the ethical issues affecting enterprise AI adoption, the importance of trust, and the need for robust data sets to support robust bias checks.
Artificial intelligence (AI) is changing lives. Transforming diagnosis and treatment across healthcare to improve patient outcomes. This has the potential to accelerate drug discovery, significantly improve road safety, and unlock new manufacturing productivity and quality through robotics. However, the speed at which emerging technologies such as ChatGPT are being adopted by individuals has pushed the ethical and political implications of AI adoption to the top of the agenda.
Despite all the benefits that AI can arguably offer, if an algorithm does not adhere to ethical guidelines, is it safe to use its output with confidence? Where’s the trust? Where’s the value if there’s no way to check if it’s ethical? And how big is the risk?
Technology leaders have called on the industry to be aggressive in developing AI, but it’s too late. Pandora’s box is wide open, with ChatGPT, everyone is playing with AI now, and individual employee adoption is outstripping corporate capacity. Currently, managers don’t know if an employee is using her AI, and have no way of knowing whether the work was performed by an individual or by technology. And some employees now claim to have multiple full-time jobs with these tools, completing tasks like content creation and coding in half the time with the tools. companies need to quickly understand their AI policies.
Aside from the ethical issues raised by individuals who might deceive employers by not devoting time to their full-time jobs, ChatGPT’s current output may not pose much of a risk. Emails and marketing copy generated by chatbots must also be subject to the same level of rigor and approval as manual content.
But this is just the tip of a very rapidly expanding iceberg. These tools are being developed at a staggering pace, creating new unaccounted for risks every day. For example, you can have a chatbot create Excel rules, but if you have no way of demonstrating what rules were used or what data was changed, can you trust that data? With the tendency to hide their use of AI from employers, companies are completely unaware of the rapidly evolving business risks.
This is just the beginning. What if an engineer asks her ChatGPT to create a list of safety tasks, or does an attorney use this tool to check case law before providing a client’s opinion of a disaster? The possibilities are endless.
ChatGPT is just one side of the enterprise AI story. Businesses are also rapidly embracing the power of AI and machine learning (ML) to accelerate automation in areas such as healthcare and insurance. These technologies are very exciting. From healthcare to education, fraud prevention to self-driving cars, these AI and deep learning solutions demonstrate superior data recognition and prediction performance, especially in visual recognition and sequential data analysis/prediction tasks. .
But this is a very big question, can businesses trust those decisions if they have no way of understanding how AI came to its conclusions? Where are the rigorous checks done? To realize the potential of AI, tools must be robust, secure, resistant to attacks, and importantly, how conclusions and decisions are made. provide some form of audit trail that indicates whether
Without this ability to “show your work”, companies face a legal and corporate social responsibility (CSR) nightmare. What if an algorithm is found to work against an organization’s Diversity, Equality, and Inclusion (DEI) strategy, resulting in bias and discrimination embedded in decision-making?
Rather than call for an unattainable slowdown in AI development, it is now imperative that data professionals come together to mitigate risk and ensure that these technologies can be used effectively and reliably. It is the duty of data professionals to develop technology that supports the safe and ethical operation of AI. This ensures that both the data used and the outputs of AI and ML activities are supported by appropriate data governance and data quality procedures, including the use of accurate and accessible datasets to check AI outputs for bias. can only be achieved if
In practice, this involves using AI to provide critical transparency to help companies understand how the AI reached its conclusions, what sources were used, and why. You need to develop trustworthy components across your production pipeline. Clearly, such “AI check” technology is inherently usable, alerting in the face of exposed bias, discrimination, or use of questionable source data, and allowing AI to review the entire process. It should be a simple data governance and risk monitoring framework. if necessary.
Creating simple tools that can bridge the gap between enterprise domain experts and AI experts will make AI systems easier to understand and trust, allowing enterprises to confidently embrace AI and trust its output. You will be able to
Additionally, there is a global need for data collaboration and data sharing within and across organizations to expand the available data and add more context and accuracy to the morass of internet-only information. This collaboration will be an important part of the process of combating the stigma and discrimination created by AI, and together with the “explainability” of AI, we will create a trusted world where AI can deliver the tangible business value organizations are seeking today. It creates a view.
Conclusion
Of course, these changes have to happen while AI continues to innovate at a staggering pace. So while collaboration and technology to enable trust in AI are on the agenda, the next few years will not be without risks. Large-scale corporate bankruptcies are almost inevitable due to mismanagement of AI use at both the individual and corporate levels.
As such, it is imperative that organizations now focus on CSR and their impact on enterprise risk, and escalate the creation of robust strategies to safely manage the adoption and use of AI. So while some organizations won’t progress as quickly as others who plunge headlong into AI and ML, taking an ethical approach to AI will protect their stakeholders and perhaps the future of their business in the process. will be protected.
author
Peter Ruffley is CEO of Zizo.
