Study Suggests AI is Not Yet Ready to Replace Jobs

Artificial Intelligence May Not Replace Jobs as Quickly as Predicted, MIT Study Finds

A new research paper published by MIT researchers suggests that the automation of jobs by artificial intelligence (AI) might happen at a slower pace than previously anticipated. While Silicon Valley executives have been touting the capabilities of cutting-edge AI systems, the study examines whether it makes economic sense for firms to replace human workers with AI in the broader context of the labor market.

The researchers focused on computer vision AI, which is capable of automating tasks that account for 1.6% of worker wages in the US economy (excluding agriculture). However, they found that only 23% of those wages (equivalent to 0.4% of the entire economy) would be cheaper for firms to automate instead of paying human workers at current costs. The authors conclude that AI job displacement will be significant but gradual, leaving room for government policies and retraining programs to mitigate unemployment impacts.

The study highlights that certain tasks, such as analyzing medical images or examining trays, are technically feasible for AI to perform. However, these tasks are often fragmented and uneconomical to automate. Neil Thompson, the lead author of the study, reassures that the rapid displacement of jobs will not lead to immediate chaos, as there is time to adapt to the changes.

It is important to note that the study focused solely on computer vision AI and did not consider more flexible systems like multimodal large language models. A recent study from OpenAI estimated that 19% of US workers could see 50% of their workplace tasks impacted by more advanced AI systems. This raises the question of whether the MIT study’s findings apply to “general” AI tools capable of automating various forms of cognitive labor.

The researchers also found that fine-tuning computer vision systems for specific specialized tasks can be expensive for companies. While larger companies may find it economically viable, smaller businesses often find it more cost-effective to retain trained human workers. However, it remains uncertain whether this dynamic will apply to language tasks. Fine-tuning advanced language models for specific tasks can be as simple as providing them with detailed written rules, making them more adaptable and potentially leading to faster and cheaper automation in language-related economic tasks.

Although the MIT study predicts a slower pace of automation, the introduction of customizable advanced AI models, like OpenAI’s GPT-4, may accelerate the process. However, the costs and safety challenges associated with integrating these models into company workflows are still factors to consider.

In conclusion, the MIT study suggests that while AI job displacement will occur, it will happen gradually, allowing time for adaptation and the implementation of policies and retraining programs. The impact of AI on the workforce will depend on various factors, including the type of AI system and its economic viability in replacing human workers.