When AI infrastructure was a big idea, we had a long time to wait
Posted On June 15, 2021
In a new report, a team of international scientists has warned that the “AI infrastructure” required to develop artificial intelligence could take much longer than anticipated.
The team, led by Dr. John Robinson, director of the Center for Artificial Intelligence at Carnegie Mellon University, made the startling claim in a report released today by the non-profit Future of Humanity Institute.
Robinson and his team’s analysis of existing AI research projects was based on their analysis of data from the U.S. National Institute of Standards and Technology (NIST) and the International Institute for Strategic Studies (IISS).
The research also found that, despite the hype, there was not enough progress to justify the cost of building AI hardware, the report said.
Instead, the researchers said that a “lack of confidence in AI hardware is undermining the ability to deliver reliable and secure AI services to consumers.”
“The fact that most AI software projects are already underway at an incredibly early stage is not an indication that AI hardware development is imminent,” Robinson wrote in the report.
“The time needed to develop AI hardware and software has grown significantly in the past few decades, and that trend is likely to continue.
The time needed for hardware development to make it into consumer products and services has also grown rapidly, and we expect this trend to continue.””
Given the limited resources available to develop a high-level, self-driving, computer vision-aware artificial intelligence, a large-scale deployment of AI hardware in the near term is unlikely to be feasible,” he wrote.
Roberson and his colleagues wrote that AI is expected to become a key component in most industrial processes by the end of the decade.
“This means that the development of AI infrastructure will have a large impact on the long-term viability of the technology,” he and his co-authors wrote.
“There is an urgent need to develop more AI hardware to enable the widespread adoption of AI technologies and for the provision of a secure and reliable AI platform to make these technologies available to consumers,” they added.
“In the long term, the potential for AI hardware investments to increase and the pace of innovation to accelerate will be crucial to the success of AI, and will have important implications for the future of the human race.”
The report is the first of its kind to highlight the slow pace of AI development, which is seen as essential to the future survival of the species.
Robson and his collaborators concluded that AI will likely need to wait until 2035 for its first breakthrough.
“The time is now,” Robinson said in the release.
“There is no time like the present.”
Robinson said that AI has been a “key driver of technological progress” in the 20th century, with the technology’s development occurring at a rate of three times per year.
“Our projections are based on our understanding of the future trajectory of AI technology and on the likelihood of achieving a critical mass of AI that will be able to compete for and advance in the emerging global economy,” Robinson and his associates wrote.
“It is imperative that we continue to build AI infrastructure and ensure that the necessary investment is made to make AI affordable and available to the global population.”
The authors said AI would be “a fundamental part of our industrial and social fabric.”