Mohanty is an award-winning AI leader & entrepreneur. She is an Applied Science Manager at Amazon & Vice Chair of IEEE Women in Engineering.
Artificial intelligence (AI) adoption is rising and evolving. As AI develops, its capabilities grow, and it is continually integrated into business workflows in new ways. However, while AI can indeed be used for a vast number of business tasks, is it really a prudent use of AI investment to do so?
Companies, both small and large, are slowly finding out the hard way that a large-scale adoption of AI into their workflows may be doing more harm to their bottom line than good. To make AI adoption economically feasible, companies need to take a measured approach in both the way they adopt AI and what they expect from their employees. The practice of “tokenmaxxing,” or measuring employee productivity by the number of AI tokens they use, is now being frowned upon, and for good reason.
AI usage tracking and leaderboards aren’t typically accurate metrics for employee productivity. To truly get a return on their AI investments, businesses instead need to reimagine which workflows will actually benefit from AI integration, and to stop pushing for indiscriminate wide-scale adoption without assessing the token cost against the productivity gain it brings. There are several aspects to be considered, and we will take a look at them all to understand a better approach to AI integration for the maximum return on investment.
AI Use Or Abuse
Before we delve into the granular aspects of properly integrating AI into your business, let us talk about the more general trends we are seeing in AI adoption. A recent Harvard Business Review article laments how generative AI gifts are double-edged swords. While generative AI can speed up processes and boost the volume of their outputs, knowledge decay can occur if companies aren’t “enforcing quality controls.” This can have a cascading effect, where errors and inaccuracies multiply into “workslop” and a loss of trust in the processes themselves, and ultimately the loss of the business’s credibility.
Another obvious issue of this AI misuse is the labor-intensive process of verifying AI-generated content after it is produced, requiring “critical thinking, additional searches, and revision,” negating the productivity gains of the AI output.
The misuse of AI takes other forms, too, including the employee use of AI for “tasks they dislike rather than tasks most valuable to the company.” With the integration of leaderboards, we even see some employees using AI to check the weather.
Others point out that using the “most powerful or expensive AI model” as a one-size-fits-all approach can skyrocket costs, and instead, using the right model for the task is the way forward.
This high-cost-to-return can compound when one takes into account that most generative AI service pricing scales through activity, instead of SaaS-style seats or subscriptions, resulting in a charge for every query rather than a fixed monthly cost. The net effect is that unless companies properly mandate AI use, costs can balloon.
These issues can be likened to the productivity paradox of information technology we saw around the turn of the century, where the paradigm shift of computing took years to properly integrate and result in actual productivity gains.
How To Properly Integrate AI Into Your Business
As I have said before, correctly implementing AI into business workflows requires reimagining the way we work. Leaders need to identify the processes suitable for AI integration, define what value is being added by the AI and determine if it actually is delivering output that doesn’t require additional man hours to validate. A part of the reimagining is instituting guardrails into every stage of the system, preventing “workslop” from insufficient supervision and cascading effects from bad data.
The aim of integrating AI into businesses is to boost productivity and free up resources for higher-order strategic thinking. AI usage leaderboards and measuring employee output based on AI use have not worked on both cost and productivity fronts. Instead, leaders need to identify tasks for automation and which AI models should be used for execution, thus preventing knowledge decay while keeping token costs where they balance productivity gains. The idea is not to replace the human workforce with AI, but to make them digital collaborators.
The reimagined workforce needs to be trained on knowledge verification, validation and entropy. Apart from ensuring the output is free from errors, companies need to validate any information they use, whether research, reports or even the resumes they receive, to ensure they are getting content that has been authenticated by humans. Such validation prevents knowledge base degradation affecting future business decisions.
This leads directly into the challenge of knowledge entropy, where systems will gradually decline into disorder, even if strictly scrutinized and maintained. This is a function of how generative AI and large language models (LLMs) operate: You are playing a game of telephone, where AI content will degrade from the original if it is further iterated by AI systems. This problem can worsen in cases of model collapses, where LLMs trained on data from another LLM can have issues with accuracy and variability.
As I’ve mentioned in a previous article, AI will only work as well as the data it is trained on and works with. Data quality and accuracy are functions of its provenance, and retaining the original, unmodified data is another vital step in this direction.
AI Is The Future: Developing A Utopia Instead Of A Dystopia
The advent of AI is a paradigm shift that’s swiftly changing the way human society works. Whether we like it or not, AI is here to stay. As it evolves, its adoption and consumer behavior are evolving alongside, in ways that may not be in our comprehension to fully predict. This mother-of-all paradigm shifts is called the technological singularity, and we’re swiftly passing the event horizon into the undefinable. It is in our power, however, to try to guide that progress as ethically and responsibly as possible to ensure a future that benefits all stakeholders.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
