The discussion about the benefits and limitations of generative artificial intelligence in the industry has become heated in the corporate office and boardroom. The first fundamental focus is if generative AI is genuinely a game changer, as indicated by fundamental measures like productivity improvements. We discover that it is, but that the advantages vary greatly from case to case, implying that managers must do their research to determine their most advantageous position in relation to generative AI.
MidJourney has already completed 50% of the corporate visual advertising presentation for Slidor, a French communication firm founded a few years ago.1 This substantial application of generative AI is also seen in the code generated by Copilot on GitHub in the same proportion.2
Needless to say, this widespread use of generative AI is rapidly infiltrating the corporate sector. In the process, it may have created its “I-phone moment”: while most digital technologies have focused on “routine” technologies, new generative AI systems such as MidJourney, Stable Diffusion, You, OpenAI’s ChatGPT, and DALL-E are automating creative tasks such as content image generation or software coding that were thought to be largely isolated by the first generation of neuronal AI.
Automation in technology is not a zero-sum game. Technology replaces duties to increase production and, eventually, to provide a larger pie for everyone to share.
Star economist Daron Acemoglu had already warned in recent scholarly articles in world-renowned journals such as the American Economic Review3 and the Journal of Human Capital4 in 2018 that the traditional assumption that “high-skilled workers are protected from automation because they specialize in more complex tasks requiring human judgement, problem-solving, and analytical skills” may be a dubious narrative. ChatGPT, a strong generative writing tool, may readily replace some sorts of writing (such as press releases and blogs), engage in meaningful conversations with clients, conduct medical diagnostics, and debug software code.
Is this the end of employment, even for highly qualified workers? We have our doubts for a number of reasons.5 The most important argument, though, is that technological automation is not a zero-sum game. Technology replaces duties to increase production and, eventually, to provide a larger pie for everyone to share. Several studies have shown that technology often generates more jobs6 than it removes, and that organizations that embrace technology may wind up expanding faster than their rivals.7 In other words, technological adoption benefits both employees and stockholders.
Despite the buzz around GPT chat, there are few studies8 that look at productivity gains in businesses that adopt these technologies. We report on one such potential research and offer crucial conclusions for CEOs debating whether to invest in generative AI.
The experiment employs an external tool, such as ChatGPT for context-based information/dialogue (nlp) and Dall-E or Stable Diffusion for content development. We neutralize disparities in tool choice as a factor in productivity differences by focusing on some of the important tools. We also looked in three other contexts: coding (including sub-activities like re-coding, debugging, and documentation), content creation (for media and advertising), and customer contacts (through social media, blogging, email, and customer support). According to numerous observations9, these are the most often utilized tasks so far (except for activities like as medical diagnosis, which are not yet prevalent), therefore we should see some gains from employing generative AI.
We compare the productivity benefit versus not utilizing generative AI. We also gather information on the user’s degree of expertise with these tools, age, employment, impression of these tools (“enriches work”, “impoverishes work”, “neutral”), and the motivation for usage (“curiosity”, “peer pressure”, “fun”, “efficiency”).
Three Key Takeaways
Three major findings stand out:
1.Usage. “Business use is already relatively high and beginning to take hold.”
Approximately 26% of respondents report utilizing technology as part of their profession, and approximately 16% use it on a daily basis. Finally, 3% of respondents say they used it and then stopped using it.
Let us not forget that these technologies are still relatively new, and that the rate of adoption at the corporate level is rather high — three to five times quicker than the last wave of so-called corporate 2.0.10 Second, conversion to an enterprise habit (16% to 26%) is also quite strong in a couple of months, while other Enterprise 2.0 technologies, with the exception of messaging and certain collaboration tools, required years.11
Leaders in adoption. “Curious digital natives are the drivers of adoption.”
We connected numerous indications gathered on whether or not employees utilized the tool. Although we can only explain less than half of the adoption rate, we discovered that the drivers of adoption are, in order of explanatory power, (a) age, (b) profession, (c) curiosity, and (d) efficiency.
Media and software workers are more likely to gain from the technology since these sectors have a history of disruption via digitization and are, in all probability, the most vulnerable to the risks and advantages of generative AI. Let us not forget that, although low-code has made coding easier, it is still a complicated activity that takes time and effort. For many years, AI has been used to attempt to automate code creation, from Amazon’s CodeWhisperer to IBM’s Wisdom. OpenAI is the next step in the growth of simple apps built from natural language instructions.12
Curiosity encourages exploration and eventually usage, but more conservative employees tend to reject technology, both due to the learning cost and the concern that technology would harm their profession and position.
Age is inversely associated to use, as it is with technology in general,13 but curiosity is a strong motivator of utilization. The latter two are more significant than the scope of generative AI and have repercussions. This means that a digital divide can form between young, digital natives and other workers — and the psychographic characteristics of workers: curiosity drives experimentation and then use, whereas more conservative workers tend to oppose technology, both because of the learning burden and fear that technology will harm their work and status.
2. Increased productivity. “They are real, but they take time to manifest due to both learning and incentives.”
The productivity effect is the sum of (a) technology users, (b) the percentage of activities where AI is used, and (c) productivity improvements in those activities.
Our findings for (a) demonstrate that 16% of employees use it on a regular basis — and that the momentum for growth is tremendous — and quicker than for any preceding technology. Regarding b), we discover that AI generative tools account for almost one-third of the actions of employees who utilize the technology every day in their workflow. This rate ranges from 12% to 44% for marketing activities (including blogging and social media communication); it is approximately 23% to 36% for coding (including documentation, code testing, and debugging; incidentally, the usage rate corresponds to the GitHub report that “46% of coding in Copilot-enabled languages is done by the AICodex wizard”).14 Finally, we see that the time spent on content creation ranges between 23% and 41% (particularly for special effects, sponsored material, and so on).
Finally, we conclude that (c) the time reduction for AI-assisted jobs is in the 30% to 60% range, which is consistent with several general experiments8 as well as some particular research with GithubCopilot.15
Adding these three factors together, overall company productivity in the professions included by the study is between 1% and 4% at a minimum. This may not seem to be much, but it is dependent on how use evolves, and this productivity increase is already more than Europe’s average labor productivity growth in recent years.16
Technology should supplement but not replace human labor, and the quality and effective use of these technologies must be worked out at the organizational level.
Outside of use, the productivity gap ranges from 1 to 4, and correlation research reveals that productivity varies depending on the activity (e.g., debugging vs writing new software code, a specific backdrop effect for content versus developing entire content, etc.). Furthermore, productivity accumulates over time — it takes an average of 6 to 8 weeks to achieve stable productivity gains using these tools (so-called learning effects), whereas productivity is higher for workers associated with companies that have long promoted the use of AI, including generative AI today.
3. How to Begin Using Generative AI
We are living in an age of business ownership of generative AI. Some executives purposefully restrict its usage, whilst others feel it should be liberated and tested among staff (see, for example, JPMorgan vs. Morgan Stanley).
While this is one of the first (relatively simple) studies on generative AI productivity, it confirms early studies of different types of AI17 that AI can be particularly powerful and adds to the parallel evidence that generative AI can provide significant gains, at least in some industries and for some tasks. As a result, these technologies may provide advantages that make the returns on investment appealing enough to warrant inclusion in any company’s technology portfolio.
However, like with Gartner’s development cycle, the future may include some setbacks, so businesses must assess their options. At least three unapologetic motions are seen. The first is experimenting, since there is a learning effect to getting the most out of technology. The second step is to research non-problematic usage scenarios. These technologies can be used to improve the efficiency of internal human resource communication, better predict customer reactions in commerce, speed up information retrieval in service contracts, virtualize an architectural project, an advertising campaign, or redesign a new website, and so on. The third step is to develop the appropriate framework for employing these technologies: they are still not transparent, are not always correct, and may introduce biases and dangers of copyright infringement. All of this argues that technology should supplement but not replace human labor — and that the quality and right use of these technologies must be sorted out at the organizational level. Finally, as general-purpose technologies have shown, technology has the potential to disrupt workflow productivity. Companies must research disruptions and change themselves in response.
This new AI moment may appear chaotic, but evidence suggests that early adopters aren’t necessarily taking risks — they’re aware of the technology’s limitations, they’re working on more explainable AI and source transparency, and they’re working internally to comply with Europe’s recent AI law. They are also preparing for new competitors and benefits, as seen by Salesforce’s rollout of EinsteinGPT as part of its data cloud, which improves the business insights supplied to its clients while also responding to Microsoft’s investment in OpenAI.