Beyond Generative AI Hype: Avoid the Bubble Through Selective People Engagement

Less than a year ago, the world witnessed the birth of a revolution: Generative Artificial Intelligence, with ChatGPT as its most notable exponent.

Due to the enormous buzz, many companies started looking at AI with increased interest, aware that this technology could be a strategic competitive factor with the potential to significantly improve business performance.

A Gartner research note from last May (see link 1 below) reports that 45% of executives interviewed credited the hype around ChatGPT as the reason for increasing their AI investments. Additionally, the same research shows that 70% of organizations are currently exploring use-cases for Generative AI, and 20% have already developed applications that or are in pilot phases or have rolled out in production environments already.

As is often the case in similar situations, the substantial acceleration towards the desire to introduce a new technology throughout the company must take into account the technical feasibility of such a demand, as well as, even more simply, the number of technical professionals that the company has (i.e., the AI teams' capacity). It is extremely easy, in such a context, to run the risk of exceeding with the exuberance of wanting to generate many ideas and then expecting that they can be realized almost immediately (and, usually, at no cost).

The situation that typically occurs in these cases is the start of a large number of working groups aimed at identifying use cases in which artificial intelligence can provide support, which see the contribution of many people and many areas of the company and, in which, the participation of AI technical staff is necessary to ensure that the discussions started actually make sense from a technological point of view. This is a classic "bottom-up" approach to innovation, which, while admirable for valuing everyone's contributions, poses several challenges:

• "Bottom-up" work often results in very specific ideas with limited potential benefits, mainly because participants in brainstorming sessions usually have a view confined to their area only,
• Qualifying the business case for these use-cases is very labor-intensive, requiring the support of the limited number of AI experts who are usually already very busy with other tasks, causing bottlenecks and delays,
• The disillusionment of people, as only a fraction of the submitted use-cases get approved for development, directly driven by the above two points: limited impactful ideas and higher than usual workload that lead to longer waiting times.

From my perspective, the most significant structural issue is that identifying very focused AI-based use-cases risks reducing AI to just another technology, rather than leveraging it as a transformative business factor.

Obviously, none of this is to say that specific use-cases should be discouraged. Many have been developed pre-ChatGPT and have certainly yielded excellent results, such as process automation and manual activity reduction. My message is simple: excessively rely on traditional approaches when we are looking for an innovative transformation may not be the best option, lest we face disappointment from unmet expectations.

Leveraging traditional approaches also stems from a fundamental misunderstanding. When the various departments of a company initiate brainstorming sessions to generate AI-based ideas, they are often actually referring to Generative AI, which is transformative by its nature. Thus, it necessitates innovative approaches to maximize its sustainable and lasting value, mitigating the risks of seeing enthusiasm deflate in a bubble in the short run (see links 2, 3, 4 below).

While a "bottom-up" approach exposes to the issues just mentioned, also a purely "top-down" approach could be not completely effective because it neglects the perspectives of those closest to operational processes and value creation (e.g., exposure to, and understanding of, the end customers).

Not only that: Generative Artificial Intelligence systems, such as ChatGPT (or similar: e.g., Google's Bard, Anthropic's Claude), are not tools that perform a specific and limited task: they can rather be considered as advanced and (almost) general-purpose personal assistants. They have the advantage that people in a company do not have to invent them: they already exist. Rather, people need to learn how to use them to understand by direct experimentation how they can add value to the organization (see link 5 below).

For these reasons, a "selective bottom-up" approach, where ideas are generated not through large brainstorming sessions, but through daily use of these tools by specific individuals, could be more effective. This approach could be organized as follows:

1. Identify task force participants. Select a limited number of people with specific qualities, such as being highly skilled and talented, open-minded, with a desire to experiment and innovative thinking, who can influence and promote solutions, etc.,
2. Deliver dedicated training sessions for these participants. Deliver trainings focused on: i. Mindset towards Generative AI, as it is crucial to convey that (e.g.) ChatGPT is not merely a substitute for human labor, but should be considered an advanced personal assistant, useful for highlighly valuable tasks like work review, problem-solving or coaching; ii. Technical usage, as without proper training Generative AI can yield disappointing results, technical sessions on (e.g.) prompt engineering are indeed required for extracting maximum utility and long lasting satisfaction,
3. Experiment and reflect. Allow time for these key people to experiment with (e.g.) ChatGPT to understand its actual utility and how it could add widespread value. This could range from basic supporting tasks, like drafting documents, to profound applications like rethinking professional roles and processes in the company,
4. Qualify benefits and set objectives. After an agreed-upon period (e.g., 3 months), ask participants to qualify such ways to utilize these tools to improve performances, specifying the expected benefits and timelines for achieving them. The objectives should then be discussed with the managers, approved, communicated and monitored.

In addition to the above, two useful enhancements could be made:

• Sustaining a community for exchanging ideas, achieved results, ask for support, etc., to accelerate benefits realization and enable "cross-fertilization" (e.g., ideas from one area could be shared and reused in another),
• Regularly repeating this entire process with newly identified people (e.g., in other areas of the organization) to gradually expand the business functions participating in the change. A "train the trainer" approach may also be considered for faster scaling.

Adopting Generative AI systems in this manner would offer several advantages:

• It could be executed in full compatibility with the "more traditional" approach of demand management for specific AI use-cases,
• It would contain the costs of using ChatGPT (or equivalent), as access would be granted only to selected people chosen for involvement in the process outlined,
• Users would master a "general-purpose" tool that they could apply to their area of expertise, where they are presumed to have maximum competence and therefore the ability to identify where and how benefits could be realized,
• AI technical teams would no longer be a bottleneck, as the use-cases definition would essentially be delegated to the staff selected for such experimentation,
• Individuals responsible for executing the business case that supports the identified use-cases would be motivated to achieve the objectives, as this would be seen an opportunity for high visibility,
• It would ensure value-added utilization of the tool (i.e., ChatGPT or similar), mitigating the risk of its discontinuation due to suboptimal use,
• Periodically proposing the process to additional stakeholders would refine the training technique, allow for more ideas to be shared (e.g., from previous sessions), progressively engage all areas of the company in a granular way, and deliver value exponentially.

Lastly, if we truly aim for a sophisticated and comprehensive approach, we should parallelly take steps to avoid potential issues stemming from a perception of "elitism" by those not involved in the aforementioned task force. Specifically, we should:

• Provide all company employees with access to ChatGPT (or an equivalent tool - in any case, with all the needed protections, for example to prevent data leaks), albeit in a more basic or "downgraded" form (for free of usage costs, if possible), as a standard productivity tool. This would prevent disillusionment among those not selected in the aforementioned process, preventing them from feeling excluded and demotivated when they learn that some colleagues can use an empowerd version of the same tool but they cannot,
• Offer all employees access to a structured educational program on AI, ensuring everyone has a basic understanding of the subject with no discrimination, and promoting a diffused positive attitude towards AI.

This proposal is indeed rather complex and non-trivial to implement, but it can certainly solve the problems of organizational impacts that are inherent in other, more traditional approaches. However, it has the disadvantage of not being able to predict in advance the extent of the benefits that can be achieved, which could lead to limited managerial commitment in the early stages.

On the other hand, the execution cost is low "by design", given the limitation of access to Generative AI systems (for a fee) to a restricted number of participants.

From my point of view, it is therefore worth testing.

The critical success factor is undoubtedly the correct identification of the people to involve in this task force. These individuals, thanks to their skills, creativity, and pragmatism, can maximize the chances of finding the desired value, which, once found, will become the engine of subsequent iterations.

If this were to prove true, it would once again demonstrate that Artificial Intelligence's success ultimately depends on Human Intelligence.

I leave here below the links to some additional article that you may find interesting:

1. Gartner Poll, May 3, 2023
2. Boom di investimenti nell’intelligenza artificiale. Ma per alcuni esperti è la prossima grande bolla
3. ChatGPT Is Losing Users. Is The Artificial Intelligence Craze Over?
4. What’s the future of generative AI? An early view in 15 charts
5. Reid Hoffman: “L’Italia impari a usare l’Intelligenza Artificiale”