For now, public attitudes seem surprisingly positive towards AI — at least in relation to workplace applications.

Several months ago, Microsoft released its work trend index, which surveyed 31,000 people in 31 countries. Nearly three-quarters (70%) of respondents said they would like to delegate as much work as possible to AI to lessen their current workload.

For its part, Microsoft said: “AI is poised to create a whole new way of working.”

While there appears to be broad agreement on that point, there is less consensus on what the future workplace will look like. At least in the near-term, AI will augment people to improve productivity, avoiding large-scale workforce displacement.

Generative AI a de facto workplace partner by 2025

A majority of workers surveyed recently by The Conference Board reported they were already using generative AI on the job. This includes tasks such as AI writing assistance, intelligent search, data analysis and image creation.

According to Gartner analyst Mary Mesaglio: “We are moving from what machines can do for us to what machines can be for us. Machines are evolving from being our tools to becoming our teammates. Gartner predicts that by 2025, gen AI will be a workforce partner for 90% of companies worldwide.”

AI autonomy

This “teammate” reality could be a permanent state — or it could be a transitory one. One indication this could be a transition phase comes from a recent New York Times story about the emerging capability for chatbots to become autonomous agents: “In time, many researchers say, the AI agents could become far more sophisticated, and could replace office workers, automating almost any white-collar job.”

The article quotes Jeff Clune, a computer science professor at the University of British Columbia: “This has a huge upside — and huge consequences — for society.”

Should it come to pass, such an outcome would certainly satisfy those who want to delegate as much work as possible to AI.

In his newsletter last week, Jack Clark —  a cofounder of frontier model company Anthropic — discusses one effort to benchmark the capabilities of large language models (LLMs) to perform complex multistep tasks, work similar to what an autonomous agent would need to do. Known as “SWE-bench,” the benchmark evaluates the ability of an autonomous agent to perform multistep software development tasks. Clark notes that solving SWE-bench tasks requires models to be able to deal with diverse long inputs, edit code in different contexts and explore a very wide scope of potential solutions.

Solving SWE-bench tasks is a crucial step in assessing the ability of an AI to work autonomously in complex scenarios, as it requires models to manage diverse inputs and explore potential solutions independently. The results to date show that Anthropic ‘s Claude2 performed best but was able to resolve less than 5% of the benchmark tests. ChatGPT performed even worse.

Augmentation, for now

These results suggest that AI engines do not currently have the capability to be highly effective autonomous agents and will continue to serve as an assistant to augment human employees — in this case, software developers. While current AI is not yet sophisticated enough to fully automate complex jobs, this capability could change if language models continue their rapid advancement.

Clark notes: “If a language model were to get much higher scores on SWE-bench (perhaps 90% would do it?) then you could imagine entirely automating some chunk of work, turning language models into virtual full employees … no human required.”

It seems certain that LLM systems will continue their rapid advance, although we do not know how long it might be before autonomous AI agents achieve this level of capability. There is speculation, as reported by Reuters, that OpenAI could soon announce technology to build autonomous agents, presumably on top of GPT-4, that can perform tasks without human intervention.

Certainly, tech companies are racing to release the next generation of LLMs, perhaps before the end of the year, with the expectation they will be substantially more powerful. This improvement does not automatically mean they would perform autonomous activities better. But if they did, the ramifications would be significant.

Future in focus

The near future is coming into focus, and AI will perform an increasing role in the workplace and offload or augment a substantial amount of the work that humans currently perform. In this approaching world, humans will manipulate and orchestrate multiple AI assistants to perform tasks and achieve business goals.

To illustrate the practical impact of AI in the workplace, let me share a recent experience. I have served for 10 years as a board member of a non-profit organization focused on development work in eastern Africa. The founder asked me to help launch his new book. I decided to leverage AI’s assistance.

From days to hours

I asked a gen AI model to provide a concise synopsis of the book, which it delivered accurately. With this summary in hand, I turned to another AI tool to help craft a compelling pitch and email outreach for reporters. Additionally, I used an AI-powered search tool to identify relevant journalists who cover related topics.

This AI-powered workflow significantly expedited the process, reducing what could have taken several days of manual work to just a couple of hours — providing a remarkable efficiency gain. Moreover, the founder was thrilled with the quality of the work. This experience exemplifies the evolving role of AI in augmenting human productivity, emphasizing the potential for humans to orchestrate AI assistants effectively in the workplace.

What emerges from this is an analogy where the human-in-the-loop becomes a conductor, and the AI assistants are members of a symphony. But even as the “musicians” improve, it is still up to the conductor to ensure the assemblage plays beautifully.

This is where the future, once again, begins to look murky. If everyone at work — at least in certain professions where gen AI will have outsized impact — were to become the conductor of an AI symphony, it is unknown how many will be needed to conduct relative to the current workforce. This realization is even more salient as the assistants become increasingly autonomous.

The only certainty is uncertainty

Autonomous agents have the potential to reshape the workforce and how we approach work. While there are significant opportunities for increased efficiency and innovation, there are also challenges related to job displacement, ethics and education.

At this point, we simply do not know if the AI-infused workplace will be utopian or dystopian or combine elements of both. It is certain that some jobs will vanish with the hope that new jobs will appear. It will be essential for businesses, policymakers, regulators and society to navigate these changes thoughtfully and responsibly to maximize the benefits of AI while minimizing potential downsides.

One thing is clear: These impacts are coming soon, so the time available for our institutions to plan is vanishingly short before they will have to react. As we step into an AI-augmented future, embracing change and continuous learning will be key to maximizing the benefits and minimizing the challenges of the AI-infused workplace. By cultivating curiosity, learning and adaptability, people can maximize the benefits and minimize the challenges of the AI-infused workplace.

Gary Grossman is EVP of technology practice at Edelman and global lead of the Edelman AI Center of Excellence.