Goldman Sachs economists said they continue to expect significant economic upside from generative AI.
Citing its baseline estimates, the bank predicts a 15% cumulative increase in US labor productivity and GDP, assuming that the capital stock evolves to match the increased labor potential from widespread AI adoption.
Meanwhile, recent work by MIT economist Daron Acemoglu presented a more conservative outlook, estimating only a 0.7% boost to total factor productivity and a 1.1% increase in GDP.
According to the bank’s economists, the discrepancies between their and Acemoglu's projections stem from different assumptions about the scope of AI automation and the potential for labor reallocation.
Goldman Sachs assumes that 25% of all work tasks will eventually be automated by generative AI, compared to Acemoglu's 4.6%. This difference is largely due to Acemoglu's belief that only 19.9% of all tasks are exposed to AI and only 23% of those are cost-effective to automate within the next decade. Furthermore, Acemoglu does not account for the potential reallocation of labor or the creation of new tasks, which Goldman Sachs includes in their estimates.
“These assumptions explain over 80% of the discrepancy between our estimates, with the remainder reflecting smaller differences in assumptions regarding cost-savings and production potential for non-displaced workers,” Goldman said in the note.
"We are quite sympathetic to Acemoglu’s view that automation of many AI-exposed tasks is not cost-effective today and for many tasks may not become so within the next ten years," they added.
However, economists also stressed that the potential for cost savings and the rapid decrease in technology costs will likely lead to more widespread adoption and automation over time.
Goldman Sachs disagrees with Acemoglu's dismissal of labor reallocation and new task creation as irrelevant. They argue that history shows technology-driven resource reallocation and the expansion of the production frontier are main sources of economic growth.
"We anticipate that AI will raise output both by increasing demand in areas where labor has a comparative advantage and by creating new opportunities that were previously technologically or economically infeasible," economists wrote.
Therefore, they view Acemoglu’s less optimistic outlook as highlighting valid concerns that notable macroeconomic impacts of GenAI are not imminent “and may be more backloaded than is commonly appreciated.”