Human programmers might get a little more breathing room before AI fully takes over coding jobs. A new report from LessWrong suggests it will take about five to six years before we see AI automating all coding tasks. This is a significant delay compared to earlier predictions, which estimated the milestone could happen as soon as 2027 or 2028. The shift in timeline highlights how uncertain and quickly changing AI forecasts can be.

Revised Predictions for AI in Coding

The experts at LessWrong updated their outlook after reconsidering various factors. Their new AI Futures Model indicates that AI could reach a level called “superhuman coder” by February 2032. This means AI systems would be able to perform coding tasks at a level comparable to the best human programmers, but much faster. They also suggest that within five years of hitting that milestone, AI might develop into artificial superintelligence, or ASI.

The term “superhuman coder” refers to AI that can manage thirty times as many coding agents as a human organization, using only a small fraction of the available computing resources. Essentially, these AIs could work independently, completing complex tasks quickly and efficiently, similar to top-tier human programmers but at a much greater scale.

Reasons Behind the Updated Timeline

The delay comes from a variety of reassessments. The researchers were less optimistic about the speed at which AI R&D—research and development—would accelerate. They also introduced a new framework to evaluate how quickly AI might improve on its own, without needing more computing power. This “software intelligence explosion” could happen faster or slower depending on how well AI can set research directions and interpret experiments.

To make their predictions, the team used a method called “capability benchmark trend extrapolation.” This involves analyzing current AI performance on standardized tests and performance trends to estimate future abilities. They also used a tool called METR-HRS to project how much computing power AI will need to achieve general intelligence. While benchmarks can be imprecise, the researchers believe this method offers the best current estimate for predicting AI progress.

However, the team also factored in potential slowdowns. Growth in computing power, data, and labor inputs isn’t guaranteed to continue at current rates. Limitations in chip manufacturing, energy resources, and investments could cause these inputs to slow down over the next few years. The researchers estimate that progress in updating AI parameters might slow by about a year, and overall AI development could experience a two-year slowdown due to these constraints.

Overall, the revised forecast indicates that while rapid advancements are still expected, they will take longer than initially thought. The shift underscores how unpredictable AI development can be and how important it is to regularly update predictions based on new evidence and changing conditions.

In the end, these findings suggest that human programmers might keep their jobs for a few more years. While AI will continue to improve, reaching full automation in coding is likely to be a gradual process rather than an immediate revolution. Staying aware of these evolving timelines helps everyone better prepare for the future of work and technology.

Inspired by

• https://www.infoworld.com/article/4113574/forecast-ai-wont-replace-human-devs-for-at-least-5-years.html

Sources

• lesswrong.com
• aifuturesmodel.com
• cio.com
• lesswrong.com
• greyhoundresearch.com