By AI Trends Staff
In the era of deep learning, cloud compute power is being concentrated in the hands of elite universities, at the expense of efforts to “democratize” access to AI technology.
A team of AI researchers from Virginia Tech and Western University conducted an analysis of 171,394 research papers from 60 prestigious computer science conferences to reach their conclusions, according to an account in VentureBeat.
The effect of the concentration is to crowd out students at mid- to low-tier research organizations, according to the analysis of accepted papers on topics including computer vision, data mining, machine learning and natural language processing.
Noting that the rise in use of GPUs since 2012 has resulted in wider availability of the powerful computing needed for AI research, the papers’ authors state, “We find that AI is increasingly being shaped by a few actors, and these actors are mostly affiliated with either large technology firms or elite universities.” They suggest this divide will need to be bridged with the help of government policy. “To truly ‘democratize’ AI, a concerted effort by policymakers, academic institutions, and firm-level actors is needed to tackle the compute divide,” the authors state.
Study authors Nur Ahmed and Muntasir Wahed summarized their findings and recommendations in the paper entitled, “The De-democratization of AI: Deep Learning and the Compute Divide in Artificial Intelligence Research.” The paper was published recently on arXiv and presented in late October at Strategic Management Society, a business research conference.
The cost to access the computing resources needed for AI training and research can go into millions of dollars. The top six contributors at leading AI research conferences were found to be Google, Stanford University, MIT, Carnegie Mellon University, UC Berkeley, and Microsoft.
Smaller schools often lack the financial resources to consider deep learning applications. This limitation tends to accelerate the brain drain to Big Tech companies of academics with the talent to teach AI, the study found. Some leave the prestigious universities for high-paying industry jobs.
Brain Drain from Academia to Industry Documented from 2004 to 2018
This trend is also found in a paper entitled “Artificial Intelligence, Human Capital and Innovation”. From 2004 to 2018, more than 200 academics versed in AI left for industry positions. Top universities, Ph.D. students, and startups in deep learning were found to have benefited the most from a shortage of talent in AI in the overall job market. Carnegie Mellon University, MIT, and Stanford University ranked highest among colleges whose alumni go on to launch AI startups.
Universities ranked 301-500 by U.S. News and World Report have published on average six fewer papers at AI research conferences since the rise of deep learning, the study found. “To the best of our knowledge, this is the first study that finds evidence that an increased need for specialized equipment can result in ‘haves and have-nots’ in a scientific field,” stated the authors on the study of the Computer Divide in AI Research.
The history of AI can be divided into two eras, authors Ahmed and Wahed suggest. The first stretches from the 1960s to about 2012, when general purpose hardware was used to train AI. In the second era, deep learning models running on specialized hardware such as GPUs have defined the industry.
The findings point to a need for a national research cloud, and shared public datasets that can help train and test AI models, accessible to resource-constrained organizations.
US National Research Cloud Initiative Making Way Through Congress
Legislation to fund a national cloud did move along in the US Congress over the summer. More than 20 major tech companies and universities joined the National AI Research Resource Task Force Act, which aims to spur and democratize AI-centered studies and applications by developing a national asset for scientists and students to use, according to an account in NextGov .
“We must maintain our AI leadership,” stated Sen. Rob Portman, R-Ohio, a member of the Senate Armed Services Committee “I heard from constituents and stakeholders about how vital this is for cutting edge AI research that will benefit the entire country.”
If passed, the bill would require the National Science Foundation and Office of Science and Technology Policy to establish a task force of experts from government, academia, and industry to pursue a “coordinated roadmap and implementation plan” for forming and sustaining the AI-focused research resource.
The policymakers aim to pave the way to lower the barrier for entry to researchers across the nation, especially those outside the major tech companies and elite universities, by opening up compute power, time and datasets.
“For the U.S. to maintain its global leadership in AI, researchers must be enabled to access high-power computing, large datasets, and educational resources that are required for AI research and development,” stated Rep. Anna G. Eshoo, D-Calif., a member of the House Committee on Energy and Commerce, who co-sponsored a House version of the bill. “This effort is critical for our economy and national security.”
UKCloud Initiative Helping Public Sector Deliver on Digital Services
Elsewhere, the UK founded UKCloud in 2011 aimed at helping the UK public sector deliver better digital services. All the UKCloud’s infrastructure technologies and services are hosted in UK-based data centers and supported and managed by staff located in the UK.
Working with partner companies who are independent software vendors, system integrators, and managed service providers, UKCloud offers software capabilities including AI, cybersecurity, big data, disaster recovery and backup, according to an account in CIO.
“UKCloud was founded on core values that include doing what’s right to improve public services for UK citizens and protecting UK data, like the national asset it is, with sovereign cloud services,” stated Leighton James, Chief Technology Officer at UKCloud. “The needs of our public sector differ from the private sector in that there is more emphasis on data governance. From patient data and citizen records all the way to military details, everything must be handled safely and in a secure cloud environment.”
Read the source articles and information in VentureBeat, in a paper entitled “Artificial Intelligence, Human Capital and Innovation,” in NextGov and in CIO.
