Balancing AI implementation with return on investment

A representative abstraction of artificial intelligence
(Image credit: Shutterstock / vs148)

For CIOs, the identification and rollout of Artificial intelligence (AI) is one of the most discussed topics. In particular, how to balance implementing the technology effectively while understanding the issues with its rollout including bias, governance, integration and security.

Mounting pressure now sits on these CIOs and other technology leaders to move swiftly in regard to AI implementation. Organizations that are slow to act could be at a competitive disadvantage, and as a result there’s a struggle to balance the fears of missing out and messing up when it comes to AI integration.

Sharon Mandell

Senior Vice President and Chief Information Officer, Juniper.

How accurate is AI

Understanding AI’s accuracy is a major consideration for technology teams. For AI to produce accurate outputs, analysis must be based on precise, unbiased data. With new generative AI and the best large language models becoming increasingly popular, the wider industry has rushed to implement them. In turn, this has impacted how AI is generally perceived, largely due to the negative focus on AI hallucinations and data biases used to train LLMs.

A recent study by Wakefield Research – a leading global polling, insights, and research provider – found that 99 percent of participants felt “some, most or all” AI data outputs were impacted by bias. Additionally, 87 percent questioned if it is even possible to know whether AI outputs are accurate.

In reality, biased outcomes are inevitable if AI is fed previously discarded or outdated data. Therefore, it is crucial to differentiate between various types of issues to assess if, and when, bias may occur. To train LLMs there must be an assessment of the uses for the data, how the AI is tested and who is using the model. Techniques such as retrieval augmented generation (RAG) can be used to train AI and ensure accuracy.

Building trust in technology

To implement AI efficiently, CIOs and other decision makers must limit employee’s blind trust in the technology. It is beneficial to see some acceptance and trust of new technology by employees and team members. However, debatably, this degree of blind trust can be a more serious business concern than any possible hallucinations or biases.

To limit associated harms, regular, up-to-date employee training on the different AI concerns and use cases is critical. From the Wakefield’s findings, 84 percent of participants felt their employers should increase AI training. Business leaders should prioritize the education of these employees, helping them to make informed decisions when using AI in high- and low-stakes circumstances.

An evolving AI landscape

As AI and its capabilities have evolved so quickly, understandably, IT professionals feel pressured to deploy AI as a priority. In fact, 87 percent of respondents claimed they felt rushed to implement AI and 74 percent felt their company’s policies could not keep up with the potential risks and benefits. Meanwhile, legal and company policies are often considered to be stumbling blocks to rapid deployment.

To increase return on investment, AI policies should be seen as a positive progression, rather than a regression. For example, some organizations have policies addressing what data employees can or cannot share with third parties. Depending on circumstances, companies can also rewrite policies to apply to external generative AI solutions. However, teams should be aware of software purchase policies and addendums for further reviews of all solutions embedded with AI.

Workforce disruption

Like other innovative technologies, AI will inevitably change how the workforce currently functions. Nevertheless, this change will undoubtedly introduce new roles and skills to the workforce. This is reflected in 85 percent of respondents believing AI will be somewhat impactful in creating future advancement opportunities.

For those looking to progress in an AI enabled job market, individuals should prioritize skills related to tasks which cannot be accomplished by statistics, as AI is largely generated by statistics. The capacity to enhance both personal and collective effectiveness through AI tools will be highly valued. Workers who remain engaged and continually refine their skills will find themselves with abundant opportunities.

Moving ahead with AI

Despite AI existing as a concept for several decades, the technology has only truly entered the mainstream in recent years. Particularly in the last few months, AI has transitioned from something of science fiction to a fundamental business necessity.

Throughout 2024 and beyond, organizations that do not have the technology in place will risk falling behind. However, implementation must be done in a measured and sensible way. Ultimately, stakeholders must fully understand how to utilize AI effectively to better business operations, processes and employee wellbeing, without jeopardizing putting client relationships at risk.

We've featured the best AI phone.

This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro

Sharon Mandell is the Senior Vice President and Chief Information Officer leading Juniper’s global information technology team.