×
Written by
Published on

A groundbreaking new study conducted by researchers at Harvard Business School and Boston Consulting Group provides an unprecedented glimpse into the impact of artificial intelligence (AI) on complex knowledge work. Through experiments involving hundreds of management consultants, the study investigates the effects of giving skilled professionals access to advanced AI technologies like GPT-3. The findings offer intriguing insights into the potential transformation of work as AI capabilities continue to advance.

Surprising Productivity Gains

The study found significant productivity and quality gains for professionals within the current capabilities of AI. When performing tasks such as data analysis, persuasive memo writing, and creative idea development, consultants completed 12% more subtasks, 25% faster, and achieved over 40% higher quality results on average. It is noteworthy that these improvements were observed in highly skilled workers, rather than novices or entry-level employees. Even experienced consultants experienced significant performance boosts through AI collaboration.

An Uneven Frontier of Capabilities

But the study highlights AI has an uneven “frontier” of capabilities. While fantastically helpful for some tasks, it falters at others that seem superficially similar. Researchers intentionally designed one task to be slightly outside the frontier to test this. When consultants relied too heavily on inaccurate AI for these tasks, performance suffered – success rates dropped 19 percentage points lower. The frontier between what AI can and (yet) can’t do well remains jagged.

Learning to Navigate the Frontier

Higher performing consultants adopted strategies the study likened to “Centaurs” and “Cyborgs” from mythology. Like Centaurs, consultants split tasks strategically between human and AI based on relative strengths. This meant delegating data analysis or memo writing to AI, while retaining strategy and recommendations for human judgement. Cyborg-style consultants tightly integrated AI throughout their workflow, continually guiding the tech via validation, edits and nudging. Both models effectively played to complementary strengths of man and machine.

Cause for Optimism and Caution

On one hand, the potential of AI goes beyond automating routine work, as evidenced by the productivity and quality gains. Knowledge workers can benefit from enhanced performance. However, blind reliance on AI outside its capabilities has led professionals astray. As technology advances, organizations and workers must strike a balance between the advantages and risks.

While not definitive, the findings suggest the emergence of templates like the Centaur and Cyborg, which blend human and artificial intelligence. The future may involve hybrid management and design of mixed human-AI teams. This also highlights the importance of continuously evaluating tasks where AI excels, struggles, and everything in between.

Just as pioneering organizations transformed work for the digital age, companies must once again adapt for an era of algorithms and bots. However, it’s not as simple as complete automation. This research offers an exciting glimpse into the nuanced integration that can unlock the full potential of thinking machines while maintaining human oversight.

Report Download: HBS

Report Authors: Edward McFowland III, Technology and Operations Managment & Karim R. Lakhani, Technology and Operations Managment

 

Key recommendations from the report:

  • Organizations should evaluate AI adoption not as all-or-nothing, but based on specific tasks and workflows
  • Focus on integrating AI into tasks within current capabilities frontier to boost productivity and quality
  • Avoid over-reliance on AI for tasks beyond current capabilities frontier without human validation
  • Develop ways to identify tasks within vs. outside of AI’s capabilities frontier as it evolves
  • Create training on how to successfully leverage AI as “centaurs” and “cyborgs” for specific tasks and workflows
  • Be aware of risks like over-reliance on inaccurate AI output and homogenized ideas from AI
  • Continuously update understanding of AI capabilities frontier as it rapidly expands
  • Monitor for negative impacts of AI like training deficits from reduced junior work responsibilities
  • Consider diversity of AI systems used to maintain range of ideas and innovation
  • Study how successful “centaurs” and “cyborgs” integrate human and AI capabilities at subtask level

Recent Articles

Pika 1.5: A Leap Forward in AI Video Generation

Pika 1.5: A Leap Forward in AI Video Generation

Move over, Hollywood – this AI can create blockbuster-quality videos with a few clicks
Guide to Google’s NotebookLM

Guide to Google’s NotebookLM

A very meta in depth guide and audio example of how NotebookLM makes a guide on using Perplexity.