light trails into the horizon

Artur Debat/Getty Images

It will soon be possible to program artificial intelligence (AI) to perform a wide variety of work that is initiated with just a one-sentence prompt. While this development will likely have a material impact on job roles quite quickly, there’s a decent chance that all human tasks will become highly automatable by the year 2116, according to a survey of leading AI researchers. 

While this reality might seem a long way off, it’s important to note that the timeline for this prediction has been moved up by half a century from the original estimate of 2164, which was made just a year ago. Plus, there are developments rapidly emerging that are destined to reshape IT and business professionals’ jobs within the next five to 10 years.

Also: Generative AI now requires developers to stretch cross-functionally

These are some of the takeaways from the survey of 2,778 AI experts, which identified a number of short-term and long-term capabilities that will be made possible through AI. 

The survey’s team of authors, led by Katja Grace and Harlan Stewart, both from the University of California at Berkeley, sought to measure the progress toward high-level machine intelligence, which is achieved when unaided machines can feasibly “accomplish every task better and more cheaply than human workers.” In both 2022 and 2023, the survey’s respondents gave a wide range of predictions for how soon such intelligence will be feasible. The aggregate 2023 forecast predicted a 50% chance of high-level machine intelligence by 2047, which was 13 years earlier than the 2060 date prediced in the 2022 survey. 

Also: Implementing AI into software engineering? Here’s everything you need to know

But let’s move away from the long-range forecasts and focus on the kinds of things the survey suggests we can expect from AI during the next few years. A synopsis of the views of 2,000-plus AI researchers finds many capabilities could be just around the corner. 

Within the next five to 10 years, AI systems will be fully capable of delivering the following results, often in response to one-sentence prompts. AI will even clean up the large language models (LLMs) that are feeding these capabilities:

  • Find and patch security flaws: “Given a one-sentence description of the task and no more input from humans, find and patch a security flaw in an open-source project with over 100,000 users.”
  • Build payment processing website: “Given a set of specifications, build a website from scratch that can handle payment processing, including the frontend, backend, and secure payment integration, without more input from humans.” 
  • Offer phone banking services, including unique tasks, on par with human operators: “Provide phone banking services as well as human operators can. This includes many one-off tasks, such as helping to order a replacement bank card or clarifying how to use part of the bank website to a customer.” 
  • Write readable Python code for algorithms like quicksort from specs and examples: “Write concise, efficient, human-readable Python code to implement simple algorithms like quicksort. That is, the system should write code that sorts a list, rather than just being able to sort lists.”
  • Finetune large language models: “Given a one-sentence description of the task, download and finetune an existing open-source LLM, without more input from humans. The finetune must improve the performance of the LLM on some predetermined benchmark metric.”
  • Run machine-language study and write paper: “Given a one sentence description of a research question in machine learning, conduct a study that would inform the answer to that question and write a paper of a quality that could be accepted at a leading machine learning conference, without more input from humans.”

The researchers suggest the drop in costs for computing will have the greatest effect on the development of AI. If AI progress continues at its current pace, “the chance of unaided machines outperforming humans in every possible task was estimated at 10% by 2027, and 50% by 2047,” the survey’s co-authors estimate: “The latter estimate is 13 years earlier than that reached in a similar survey we conducted only one year earlier [in 2022].” 

Also: These are my 5 favorite AI tools for work

The study’s authors also assimilated predictions pertaining to the “full automation of labor”. They predict, “the chance of all human occupations becoming fully automatable was forecast to reach 10% by 2037, and 50% as late as 2116 — compared to 2164 in the 2022 survey.”


Source link