Microsoft’s latest Future of Work Report comments on how AI is reshaping how we work, presenting generative AI in an overwhelmingly positive light.
The report concerns itself mainly with how workforces will change and how we’ll be using generative AI for a host of different tasks. Unsurprisingly, there’s a strong focus on Microsoft or Microsoft-sponsored generative AI tools and products such as Bing Chat and the GPT-4 LLM. In broad strokes terms, the report paints quite a positive picture that AI is transforming the way we work, often making us better in our typical workflows. It also highlights some key challenges.
Microsoft started releasing its Future of Work report back in 2021 after the COVID-19 pandemic induced a paradigm shift in how we work (remotely now). Their latest report focuses on generative AI. This report is a culmination of different research papers from the last decade or so, mixed with new information and current advancements, that aim to tell us about the direction we’re moving in.
Most notably, there is no major point being made about how it will affect hiring or the composition of workforces itself. How will generative AI affect the jobs of people, particularly jobs that are easily replaceable? Job roles that require rote or repetitive tasks or roles that require only a grasp of vast amounts of information are both better handled by generative AI. It’s also feared that job cuts are looming large for the majority of big tech companies, and the more we innovate in the direction of generative AI, the more it can replace human beings, even in artistic or more creative job roles.
To summarize the main slides, Microsoft says that people in organizations do not always accept new technologies (based on previous research papers), which is undoubtedly true, especially if the new technology can replace parts of their job (the whole Luddite thing, basically). What the company fails to recognize is that generative AI in 2024 is so much bigger than the impact electricity, internet, or computers had in their own times. The majority of the modern workforce in advanced countries is based on knowledge and thinking – two concepts rooted in the idea of generating ideas and having intelligence. It’s a no-brainer that advanced generative artificial intelligence can replace this. It’s in the name.
Impact of AI on Work
On page 34, Microsoft does comment on how AI tools are perceived by “knowledge workers” – basically people whose jobs involve handling or using information. Here, too, the company wants current knowledge workers to adopt new tools. The report cherry-picks the papers that side with AI/technology throughout its entire body, but it becomes a little more difficult to digest here. This page quotes the study where radiologists, without working with AI in the past, had incorrect expectations from the tools. The problem is that this report completely fails to cite any of the many peer-reviewed papers that hint that people who are on the verge of replacement by AI or have already been replaced by AI had no incorrect expectations. They were just more expensive to maintain.
A paper by the National Bureau of Economic Research, the first empirical and qualitative study about “Generative AI at work” (PDF here) concludes that using a generative AI tool in a real-world customer support workplace increases worker productivity and customer sentiment. Though the researchers stop short of pretending to know how this will impact the demand for human customer support employees, I think it’s pretty evident that at least the less-skilled job roles with lower satisfaction rates can be entirely eaten up by generative AI, or by a configuration such as hiring a few high-skilled workers to oversee generative AI tools, that essentially outperforms having to pay more of less-skilled workers or taking risks in hiring.
Key Points in the Report
35 out of the 41 slides in the presentation deck are the main content of the report. Each of these slides has a title that describes the data being discussed with charts and cited research papers. You can read these titles to get a gist of what Microsoft believes is the future of work:
- Lab experiments show LLMs can substantially improve productivity on common information work tasks, although there are some qualifiers.
- Copilot for M365 saves time for a variety of tasks in lab studies and surveys.
- The evidence points to LLMs helping the least experienced the most.
- Critical thinking: LLM-based tools can be useful provocateurs.
- AI can enhance microproductivity practices.
- Analyzing and integrating may become more important skills than searching and creating.
- Constructing optimal prompts is difficult.
- But constructing effective prompts is becoming easier.
- People are also learning to prompt more effectively.
- Complementarity is a human-centered approach to AI collaboration.
- Appropriate reliance on AI is a key challenge in human-AI interaction.
- Uncertainty visualization can help create appropriate reliance.
- Co-audit tools help users check LLM outputs.
- Generative AI demands greater metacognition from users but also has potential to support it.
- LLMs have made giant steps forward in multilingual performance, but there is still much to be done.
- New systems point to how LLMs can aid creative activities.
- Bing Chat is frequently used for professional and more complex tasks.
- “Fast AI” and “Slow AI”: Different LLM experiences require different latencies.
- For software engineering, benefits of LLMs depend on the task.
- New research highlights some of the benefits of LLMs in education.
- GPT-4 excels at core examinations for medical licensure and practice.
- LLMs will change the way social science research is done.
- Instant AI feedback may improve real-time interactions in meetings.
- Retrospective AI feedback may improve long-term meeting interactions.
- AI may help leaders and teams plan and iterate on workflows.
- Digital knowledge is moving from documents to dialogues.
- LLMs may help address one of the greatest problems facing organizations: knowledge fragmentation.
- The introduction of AI into any organization is an inherently sociotechnical process
- How AI tools are perceived by knowledge workers and whether they fit their work context can determine if they will be effectively adopted.
- Human-AI working: Monitoring and takeover challenges.
- We need to work to mitigate increased risk of “moral crumple zones”.
- Early evidence shows disparities in adoption follow traditional digital divide.
- Most jobs will likely have at least some of their tasks affected by LLMs.
- Innovation is the secret sauce to job creation with new technologies.
- The future of work is a choice, not a predetermined destiny.
The bottom line is we are truly at the precipice of a transformation. Companies are changing how we work remarkably. But it’s not enough to show only one side of the coin and say that workers need to adapt to the new technology. We have to focus on the people being slowly replaced by AI. At a time like this, if a company like Microsoft only wishes to protect its vested interests by publishing a one-dimensional report, cherry-picking research papers that encourage the idea that AI is not something to be worried about, all you need to do is just make a few tweaks to be fine or even become more productive, then it’s more problematic than the actual impact of generative AI on work.