Is Generative AI Really Helpful at Work? (‘IISIA Technology Blog’ Vol. 4)
Some have already begun to spread statements such as ‘the generative AI boom has already peaked,’ but even so, how we should approach this so-called ‘generative AI,’ and in particular how we should proceed with its application in the workplace (social implementation), remains a major challenge. While there are various discussions developed, in my humble opinion, the direction is slowly converged into two. At first, this ‘IISIA Technology Blog’ is written from the AI planner’s point who brainstorms for social implementation rather than as an AI engineer. As such, this time, I face the problem of ‘generative AI is really helpful at work’. When discussing this issue, many commentators first seek evidence to support their argument [Brynjolfsson et al. 2023]. The abstract of the thesis titled ‘Generative AI at Work’ is stated below.
We study the staggered introduction of a generative AI-based conversational assistant using data from 5,179 customer support agents. Access to the tool increases productivity, as measured by issues resolved per hour, by 14 percent on average, with the greatest impact on novice and low-skilled workers, and minimal impact on experienced and highly skilled workers. We provide suggestive evidence that the AI model disseminates the potentially tacit knowledge of more able workers and helps newer workers move down the experience vurve. In addition, we show that AI assistance improves customer sentiment, reduces requests for managerial intervention, and improves employee retention.
You may have noticed if you learnt how to read through an academic thesis, that this thesis cannot blindly be stated ‘generative AI is helpful at work’. In the first place, even if it says generative AI, the target here is text generative AI. That is, image-generative AI based on diffusion models is not in its scope. This is the first thing you need to ascertain. When you look closely with that in mind, while the learning effect of unskilled workers by using ‘text generative AI’ is recognised, the impact on skilled workers is less noticeable. The reason for this evaluation is that tacit knowledge is made explicit by the ‘Text-based AI,’ and by being able to learn it efficiently, unskilled workers are able to rapidly improve their abilities.
Actually, [Brynjolfsson et al. 2023] even stated this. As yet, the author is not aware of any evidence that any discussion highlighting this point is taking place in Japan.
Finally, our findings raise questions about whether and how workers should be compensated for the data that they provide to Ai systems. High-skill workers, in particular, play an important role in model development but see smaller direct benefits in terms of improving their own productivity.
This thesis considers GPT as an analysis materials and the learning of Large Language Model (LLM) is taken as a discussion point. With customer centres being used as the experimental site, various data will be extracted from skilled operators and injected into the LLM to train it, and this data will then be transferred to unskilled operators, bringing great benefits to the latter. However, for the former, the know-how they have built up is simply taken by the company, and it is argued that compensation should be made for this. It seems there is no problem from the management view, however, from the employee’s point of view, this is the very point that could be said to be the problem. Therefore, at this point, especially the ‘propulsion clique’ of social implementation in ‘text generative AI’ needs to explicitly provide an answer.
In response to this, [Eapan et al. 2023] discuss generative AI from a broader perspective of ‘creativity’ and in a more general sense, not limited to text generative AI. The article column titled ‘How Generative AI Can Augment Human Creativity’ states as follows at the beginning.
The term “democratizing innovation” was coined by MIT’s Eric von Hippel, who, since the mid-1970s, has been researching and writing about the potential for users of products and services to develop what they need themselves rather than simply relying on companies to do so. In the past two decades or so, the notion of deeply involving users in the innovation process has taken off, and today companies use crowdsourcing and innovation contests to generate a multitude of new ideas.]
Von Hippel is a famous advocate of the concept of ‘stickiness’. Knowledge and experience in the workplace each have a unique ‘stickiness,’ meaning that they are often not easily transferable from one worker to another. That impedes the company’s growth as an organisation and it needs to strive to minimise it. In this connection, the ‘SECI model’ by Ikujiro Nonaka is on the extended line, as well as the discussion of ‘skill and knowledge transmission’, which has arguably been disputed in Japan.
Since Eapen stands on the extension of these discourses, this thesis is written from a basic position that generative AI has a positive effect on creativity at work. However, the discussion here is not as ‘test generative AI’ which has the advantage of (kind of) restoring past text data, but keeps ‘image generative AI’ which features diffusion models (reference [Okanohara 2023]) learning generative models from a finite number of practice data and realising various samples except from practice data in mind. And the result is as argued below, that generative AI is the one bringing ‘democratising innovation’ as Von Hippel advocated before.
Humans have boundless creativity. However, the challenge of communicating their concepts in written or visual form restricts vast numbers of people from contributing new ideas. Generative AI can remove this obstacle. As with any truly innovative capability, there will undoubtedly be resistance to it. Long-standing innovation processes will have to change. People with vested interests in the old way of doing things-especially those worried about being rendered obsolete-will resist. But the advantages-the opportunities to dramatically increase the number and novelty of ideas from both inside and outside the organization-will make the journey worthwhile. Generative AI’s greatest potential is not replacing humans; it is to assist humans in their individual and collective efforts to create hitherto unimaginable solutions. It can truly democratize innovation.
As stated above, there are two directions regarding the effectiveness of generative AI at work. Especially when considering text-generative AI, it would be of great significance to enable the efficient transfer of tacit knowledge possessed by skilled workers to unskilled workers. However, if this is the case, there should be some compensation for skilled workers, and even more so, if there are still elements that cannot be transferred, then they will be truly ‘unique,’ and the existence of a very small number of professionals who continue to earn high incomes in areas that generative AI cannot reach will inevitably be revealed. On the other hand, when considering image-generative AI, it can be said that there is a reason to hold a slight expectation to be realised ‘democratising innovation’. The presence of generative AI, which enables ordinary people, as well as consumers, to ‘create from zero’, ultimately ‘bouncing off ideas’ necessary for innovation, has been manipulated by only a part of the creators. However, it is something unforgivable for those who consider these parts of creative activities as vested interests. Therefore, ‘democratising innovation’ will not progress easily, and we will need to accept the fact that completing the above is the situation closer to revolution.
However, nothing will happen until we try out generative AI ourselves, just like with any other technology that humanity has acquired so far. As such, considering this, our Institute has launched virtual PR ‘Mariko’ by stable diffusion technology. Looking at the example above, what would you see in the future of generative AI (Please note that the text read in the video is fictional.)?
26th August, 2023, Marunouchi, Tokyo
CEO/Global AI Strategist
Written by Takeo Harada
(References)
Brynjolfsson, Erik, Danielle Li, and Lindsey R. Raymond. Generative AI at work. No. w31161. National Bureau of Economic Research, 2023.
Eapen, Tojin T., Finkenstadt, Daniel J., Folk, Josh, Venkataswamy, Lokesh. How Generative AI Can Augment Human Creativity, HBR, July-August, 2023.
Okanohara, Daisuke. Diffusion Model: Mathematics of Data Generation Technology. Iwanami Shoten, 2023.