In a surprising turn of events, the Italian government recently imposed a ban on ChatGPT, an artificial intelligence system known for generating human-like text and computer code. This decision was made due to concerns about privacy violations, as ChatGPT is trained on vast amounts of data collected from the internet. While the ban is still under investigation, there have been unintended consequences resulting from it.
One of the key questions surrounding large language models like ChatGPT is how they are being utilized by humans to enhance productivity or even replace human workers entirely. The potential of these models is evident, but the extent to which they have infiltrated the workplace remains a mystery. However, researchers David Kreitmeir and Paul Raschky from Monash University in Australia saw an opportunity to shed light on this issue arising from the Italian ban.
The specific area of work they focused on was computer coding. While there is anecdotal evidence suggesting coders heavily rely on ChatGPT for code development and correction, concrete evidence has been hard to come by. Kreitmeir and Raschky recognized that if Italian programmers had truly become dependent on ChatGPT, the impact of its absence would quickly become apparent. To investigate this, they analyzed data from the popular software repository GitHub, which captures the hourly coding output of 8000 programmers in Italy as well as two other European countries, both before and after the ban.
The findings were quite surprising. According to Kreitmeir and Raschky, the ban on ChatGPT led to a significant decrease in coding output by Italian GitHub users, with a decline of around 50% during the first two business days following the ban. However, output levels eventually returned to normal. The researchers speculate that coders found ways to circumvent the ban, such as using virtual private networks or Tor Relay to access ChatGPT from other countries. In fact, they observed a noticeable increase in Google searches for “virtual private networks” immediately after the ban was imposed.
These findings suggest that the usage of ChatGPT extends far beyond initial expectations, with Kreitmeir and Raschky describing it as evidence of the active employment of ChatGPT for high-skilled tasks in the economy. Considering that ChatGPT was just launched in November of the previous year, this may be just the beginning of a wave of evidence to come regarding its widespread impact.
Reference:
“The Unintended Consequences of Censoring Digital Technology – Evidence from Italy’s ChatGPT Ban”: arxiv.org/abs/2304.09339