SITUATIONAL AWARENESS: The Decade Ahead - What to Watch in AI

SITUATIONAL AWARENESS: The Decade Ahead - What to Watch in AI

The race towards developing Artificial General Intelligence (AGI) is well underway. We are on the path to building machines that can think and reason, with projections showing that these machines will surpass the capabilities of many college graduates by 2025/26. It is predicted that by the end of the decade, we will have achieved true superintelligence, where machines will be smarter than human beings. Currently, there are only a few hundred individuals, mainly concentrated in San Francisco and AI labs, who possess situational awareness of this progress.

The advancement in AGI is incredibly plausible, with the development of models like GPT-2 to GPT-4. These models have taken us from the abilities of a preschooler to that of a smart high-schooler in just four years. By tracing the trendlines in compute power, algorithmic efficiencies, and the removal of limitations, we can anticipate another qualitative jump in AGI capabilities by 2027, equivalent to the leap from preschooler to high-schooler.

It is essential to understand that AI progress will not stop at the level of human intelligence. With the automation of AI research through hundreds of millions of AGIs, we can compress a decade's worth of algorithmic progress into less than a year. This rapid advancement will push us beyond human-level capabilities, leading to vastly superhuman AI systems. The power and potential dangers of superintelligence are monumental.

As the revenue generated by AI continues to grow exponentially, we can expect trillions of dollars to be invested in building the necessary infrastructure, including GPUs, data centers, and power supply, before the end of the decade. The industrial mobilization required for this expansion will be intense, with a significant increase in electricity production in the United States alone.

In the world of AI, there are specific areas that we need to closely monitor. One of the primary challenges is the exponential rise in knowledge and the distributed nature of work. This has led to an increased amount of time needed to find existing knowledge, making the process of "searching for stuff" at work inefficient. To address this issue, intuitive work assistants like Glean have become critical tools in driving employee productivity. These assistants help individuals navigate the vast amount of information available, ensuring that they can access the right knowledge at the right time.

Furthermore, enterprises face obstacles in shipping AI applications to production due to the lack of appropriate governance controls. Ensuring that applications understand what end-users are allowed to access, determining where inference is performed, and defining ownership of the source data leading to model outputs are crucial for maintaining ethical and legal standards. Overcoming these obstacles will enable enterprises to leverage AI effectively and responsibly.