AI is the Next General-Purpose Technology
Microsoft’s Chief Data Scientist Reveals How AI Is Transforming the World at the 2024 AI for Good Global Summit
When you hear “GPT,” your mind might not immediately jump to “General-Purpose Technology.” However, GPT not only refers to the Large Language Model (LLM) that got people excited about artificial intelligence (AI), but it also stands for inventions that are so revolutionary as to dramatically change the world.
During his 20-minute presentation on Day 1 of the 2024 AI for Good Global Summit, Microsoft’s Juan M. Lavista Ferres makes the case that AI is the world’s newest general-purpose technology (or GPT) that is poised to significantly impact our economy and society.
Lavista Ferres provided compelling insights into AI’s transformative power. This article summarizes eight key insights from his talk, offering a glimpse into how AI is already reshaping our world.
The presentation opens and closes with specific examples of how AI is already making positive advancements in education, disaster preparedness and response, health, and environmental protection. For more on AI’s role in advancing the Sustainable Development Goals (SDGs), be sure to check out this upcoming article from SDGCounting — AI in 2030: Broad Impacts and Specific Solutions for a Better World.
General-Purpose Technology Defined
A “General-Purpose Technology” is a technology that has a profound impact on the economy and society, with the potential to drastically alter the way we live and work. GPTs are characterized by their pervasiveness, potential for improvement, and ability to enable innovation in various sectors. Examples of GPTs include the wheel, electricity, the internet, and now, AI.
8 Insights
To ensure that the changes AI creates will benefit everyone, we need to not only understand this technology’s transformative power for the future but also the limitations and challenges inherent in today’s LLMs. To do this, Lavista Ferres offers the following key insights.
Insight #1: LLMs Are Just Math
LLMs, despite their sophisticated behavior, are fundamentally mathematical constructs. They perform extensive linear algebra and mathematical calculations. Their seemingly human-like interactions arise from training on human data. Lavista Ferres highlighted, “No matter how sophisticated these AI models appear to be, behind the scenes, these models are just math.”
Insight #2: LLMs Do Not Have Reasoning or Intentionality
LLMs generate outputs based on statistical correlations rather than deliberate thoughts or reasoning. They predict the next word in a sequence based on learned probabilities. Lavista Ferres explained, “Their outputs are driven by statistical correlations, not by deliberate thoughts or reasoning.”
Insight #3: LLMs Do Not Know What Is True
LLMs lack inherent knowledge of truth. Their responses depend on the data they were trained on, which can include inaccuracies. Thus, it’s crucial to use reliable datasets for training.
Insight #4: LLMs Can Hallucinate
LLMs can produce inaccurate or nonsensical outputs, known as hallucinations. This occurs because LLMs infer data based on probability distributions, not on actual knowledge. Lavista Ferres noted, “They don’t know they’re hallucinating because otherwise, it would be relatively easy to solve the problem.”
Insight #5: Generative AI Models Can Be Affected by Bias
Generative models can reflect societal biases present in their training data. Lavista Ferres highlighted studies showing skewed gender distributions in generated images of various professions, emphasizing the need to detect and mitigate such biases. Because these models are learning from our biased dataset, he urged the importance of training on diverse and representative data to avoid perpetuating biases.
Insight #6: LLMs Can Be Very Powerful
LLMs have the potential to convert vast amounts of internet data into actionable knowledge. They have demonstrated capabilities such as passing professional exams and understanding complex instructions.
“Zero-shot learning,” or an LLM’s ability to perform a task or solve a problem without being explicitly trained on that specific task, is another example of the powerful capability of LLMs like ChatGPT-4, which can handle a wide range of tasks without requiring task-specific training data.
Insight #7: LLMs Can Improve Productivity
LLMs can significantly enhance productivity across various fields. Experts routinely cite coding as a skill that LLMs are greatly improving, especially since 99.95% of the world’s population does not know how to code. Lavista Ferres also highlighted a BCG study of 758 consultants that showed marked improvements in task completion speed and quality for consultants who had access to ChatGPT-4 compared to those who didn’t.
AI’s potential to significantly improve global productivity is crucial since this is the first time in human history when the working-age population in developed countries is declining.
Insight #8: Generative Models Can Help Close the Digital Divide
Lavista Ferres emphasizes that AI can help close the digital divide by making advanced capabilities accessible to a broader range of people, including the 1.3 billion people around the world who live with a severe disability. “LLMs have the power to make anybody an AI power user,” he states.
In the education sector, AI can tutor students in a language they actually understand while reducing teacher time spent on administrative tasks to create more time for student-teacher interaction.
Conclusion
The AI for Good Global Summit presented nearly 100 demonstrations on how AI can improve our world. Juan M. Lavista Ferres’ presentation provides a helpful framework for understanding the strengths and limitations of the technology and for placing its importance in the larger global history. With this understanding, we can better harness its power to drive positive change.
“AI is a new general-purpose technology and provides amazing power to change the world. Together we can build a future AI that benefits everybody.”
Learn More
This AI for Good Global Summit 2024 presentation — and many more — are available to watch on the organization’s YouTube channel. Check out the playlist or directly access Lavista Ferres’ presentation on Day 1.
Our StartingUpGood team believes that events and conferences are great places to learn, share ideas, and innovate. We are committed to using our innovative tech tools to share key insights and learnings from top conferences. This article uses Otter.ai to create transcripts and various LLMs to help summarize and proof-read. All content for the article was hand-curated and checked for quality.
StartingUpGood supports fresh entrepreneurial approaches to social impact. FOLLOW US on social media:
Check out SDGCounting for the latest news on tracking the progress of the Sustainable Development Goals. #SDGs #GlobalGoals
