So-called "generative AI" is attracting the attention of economists, independent organizations and consulting firms, that studies are pouring in and converging on one key conclusion: depending on the nature of the task, this tool brings significant productivity improvements. Our article takes a critical look at one of the fastest-adopting dimensions of artificial intelligence. Despite its marked effect on technological integration in the professional world, we take a close look at its true significance for human resources, and give you the keys to assessing its value in your specific context.
Analysis of how Generative AI can improve productivity
Like me, you doubt the extent of a phenomenon amplified by an often divisive, rarely objective press. I'd like to remind you that studies are sometimes mistaken, often in the sense of minimizing the impact of a trend. We recall the case of McKinsey, who estimated that the cell phone would be a niche (link here), or the great Paul Krugman on the subject of the Internet: "Its growth will shrink drastically. By 2005 or so, it will be clear that the Internet's impact on the economy will have been no greater than that of the fax machine".
How does the GPT model boost productivity?
There are 7 main advantages to TPM that explain its positive impact on productivity, from the generation of quality content to its ability to personalize customer service. Let's take a look at these sources of growth:
- Content creation assistance : GPT generates quality content, accelerating communication while maintaining relevance.
- Optimized search processes : GPT simplifies information retrieval, speeding up decision-making thanks to improved access to relevant data.
- Supporting creativity and innovation: GPT stimulates innovation by generating original ideas and pushing back traditional boundaries.
- Improved decision-making: GPT provides relevant information for informed decisions, anticipating the potential outcomes of the options under consideration.
- Automation of repetitive tasks: GPT speeds up routine tasks, freeing up time for more complex activities.
- Rapid information processing: GPT rapidly analyzes large quantities of data, facilitating faster, more informed decisions.
- Customer service personalization and automation: GPT provides precise answers for more efficient customer service, allowing agents to focus on complex problems.
What about studies on the contribution of Generative AI to productivity?
To answer this question, we'll take a closer look at 2 studies in particular, one by the National Bureau of Economic Research and the other by Goldman Sachs:
The study by the National Bureau of Economic Research (US)
The NBER claims that the use of the Generative Pre-Trained Transformer can bring a 14% productivity gain on a population, its base of exploration is that of a population dedicated to customer assistance. This result can be explained by the GPT's ability to provide precise, consistent answers to uncomplicated customer questions, as long as these are simple interactions. This additional efficiency benefits less-skilled workers in particular , and reduces the need for supervisors.
Current fears about the disappearance of professions are understandable, but they require a different view of the circumstances:
We need to distinguish between the notion of task and that of job.
AI may automate certain tasks, but it doesn't do away with professions. Depending on the nature of the tasks within a trade, several scenarios will emerge: either the trade is really threatened, or it is impacted in the sense of complementarity, or, finally, it simply remains protected.
Goldman Sachs study: "The Potentially Large Effects of Artificial Intelligence on Economic Growth".
To understand the scope of the phenomenon and its limits, we have evaluated the global economic analysis proposed by Goldman Sachs, which offers an international dimension since it covers data from both the USA and Europe.
What are the conclusions of this analysis on the impact of Generative AI on productivity?
- Around 300 million jobs could be impacted by the latest wave of AI. In the USA and Europe, almost two-thirds of current jobs are exposed to partial AI automation.
- The majority of these jobs have a significant proportion of replaceable tasks (25-50%). Although the overall impact of AI is notable, most jobs and industries are partially exposed to automation, favoring complementarity rather than replacement.
- As a result, around 7% of current jobs in the US could be replaced by AI, 63% completed and 30% unaffected.
The current state of Generative AI in the enterprise
A need for clarification of the various AI terminologies
The figure of 36% AI adoption rate in the enterprise was given by Michael Page in 2020, an assessment itself based on an IBM study. Why is that?
Quite simply, because artificial intelligence is often misunderstood and covers very different fields. Here's a glossary of the different terms used today, to help you better identify what's happening today:
- AI : Computer science subcategory focused on the creation of systems capable of performing tasks that normally require human intelligence.
- Machine Learning : A subdivision of AI that focuses on the research and development of mechanisms that learn from experience without being explicitly programmed.
- NLP (Natual Language Processing): This is a segment of Machine Learning that focuses on the interaction between computers and human language.
- LLM (Large Language Model): Statistical calculation models developed to anticipate the next word or sentence based on the prior analysis of huge quantities of textual data.
- GPT (Generative Pretained Transformer): Modeling developed by OpenAI, in which Microsoft has invested 10 billion. It has been pre-trained on a corpus of tens of billions of words from a wide variety of sources. It generates text in response to queries or "prompts" that require a methodology to produce relevant results.
Why is generative AI accelerating enterprise AI adoption?
While AI is often misunderstood, the current explosion among the general public, reaching a million users in 5 days, can be explained by factors that can be transposed to the enterprise:
- Firstly, the model's ability to be driven at low cost.
- The availability of HR data has also increased tenfold: in addition to the corpus of texts on which GAI relies for its power, there is now data provided by users. On this subject, read our recommendations for making your HR data more reliable.
- The advent of Generative AI is also due to the ease of interaction with the "machine". What will change and dictate the adoption of this tool is its ability to integrate into users' daily lives in an invisible, intuitive way, in the "flow of work".
AI technology is like a compass that guides us through the vast landscape of data to uncover hidden nuggets of talent knowledge. Alexandra Levit, Founder and CEO of Inspiration at Work.
Generative AI and task automation: an 8-criteria reading grid
To hear HR analysts and influencers tell it, this technology completely frees HR from core talent management activities. Alain Goudey states that "a 2022 study showed that using AI to analyze CVs can reduce recruitment time by 75%": which study? On what sample population?
The networks are teeming with glowing claims: resume analysis, pre-selection of candidates, generalization of job descriptions, construction of reference systems, employee appraisal, anticipation of departures. These promises are sometimes verified, and sometimes overestimated. So how do you assess the benefits of integrating generative AI to support your business tasks?
In order to estimate the real potential of generative AI for boosting performance, let's base ourselves on a pragmatic methodology that takes into account the criteria for automating a task.
A simple methodology for identifying whether a task can be automated
The productivity improvements provided by AI are based on several criteria that you yourself can use to decide whether or not to opt for a solution incorporating AI:
1. Degree of complexity :
It's important to understand the inherent complexity of the task. Simple tasks have a higher potential for automation.
Example: A neurosurgeon, although benefiting from major advances in AI for the doctor, remains a protected profession.
2. The level of human interaction required:
AI can handle tasks that require basic human interaction, such as answering frequently asked questions. However, for more complex interactions, requiring deep understanding, empathy or consideration of cultural or emotional contexts, AI's current capabilities are limited.
For example, a social worker or clinical psychologist, whose job is based on listening and sharing, is a profession whose tasks can hardly be automated.
3. Level of expertise required :
Generative artificial intelligence can reproduce past reasoning, but cannot match advanced levels of expertise. This is illustrated in particular by the hallucination bias, which consists in producing information that appears credible but is in fact incorrect or unfounded in reality. This phenomenon is particularly worrying in applications where the accuracy and reliability of information is crucial.
4. The amount of data processed:
The more a task requires the manipulation of large quantities of data, the easier it can be performed with AI. For example, a bioinformatician, whose tasks rely on the processing of large amounts of data to be cross-referenced, can benefit from a major contribution from AI.
5. Regulatory constraints:
To a certain extent, the in-depth knowledge of regulations available to generative AI can tend to automate a number of tasks. Whereas ChatGPT relied until the end of August on data prior to 2022, it is now obsolete: AI can fetch data in real time.
6. Human risk :
One limit to the spread of AI remains: that of fully trusting an AI to carry out tasks where the human risk of an error is significant. Let's take the example of transport as a general rule.
7. Strategic and creative nature: Generative AI doesn't really create anything new, since it proposes the generation of any kind of content on the basis of probabilities. Use cases are still in their infancy and are gradually spreading, but new ideas will be needed to enrich its operation sooner or later.
8. The physical dimension: Guess which is the least automatable job in the world?
The tiling profession: it requires physical qualities, expertise and a form of strategy. AI cannot help to automate physical tasks, and this coincides with a strong appetite among French employees for retraining in manual trades (37% according to OpinionWay).
Conclusion
In conclusion, it's important to note that the impact of generative AI on productivity varies depending on the business sector, the way the technology is implemented and the skills employees to take full advantage of its functionalities. In addition, challenges remain, particularly with regard to trust in the results produced by generative AI and the need to ensure ethical use of this technology.
