Artificial Intelligence (AI) and Machine Learning (ML) are transforming entire industries because of higher performance and faster time to market. Part of the success is due to researchers creating and open sourcing datasets, frameworks, and algorithms (e.g., ImageNet, Caffe). Current leaders are following suit by opening up their own developments (e.g., DeepMind Lab and Sonnet, OpenAI Gym and Universe). Despite this generosity, operating and developing on these components still requires large amounts of expertise and vast computational resources.
Jack Clark of OpenAI believes that this situation seems to benefit large-scale cloud providers like Amazon, Microsoft, and Google. All of these companies have an incentive to offer value-added services on top of basic graphics processing unit (GPU) farms. This is also why our data center people are working with NVIDIA to add GPUs to our Unified Computing System (UCS) line (Dec 2016).
The addition of GPUs makes it likely that each cloud/appliance will specialize around one or more particular frameworks to add value as well as services that play to each provider’s strengths. For instance, Amazon: MXNet integrated with the AWS suite. Microsoft: CNTK integrated with business-process automation and LinkedIn data. And Google: TensorFlow integrated with ecosystem ML services.
While this makes business sense in the short to medium term, it might severely limit the potential of AI in much the same way the von Neumann architecture got us all stuck onto CPUs until now. To uncover paths that might be 100 times more powerful than the current state-of-the art, it’s necessary to still provide researchers, startups, and hobbyists the possibility to experiment with low budgets.
Distributed AI Development Using Blockchain
French researchers have outlined Morpheo: A distributed data platform that specializes in ML—especially transfer learning—and uses the blockchain for securing transactions. Thus, creating a trusted compute economy. The system (outlined in this research paper) lets researchers access large amounts of distributed computers, using cryptocurrencies to buy and sell access to compute and data. The researchers announced that, “The first stable release with a blockchain backend is expected in Spring 2018.”
On the same vein, albeit not specialised in AI, there’s Golem: A decentralized, open source supercomputing system. It lets people donate their own compute cycles into a global network, and will rely on Ethereum for transactions. Every compute node in the network will have a reputation, a measure of trust that changes according to how well it completes the jobs associated with it. Using reputation as a token of value, this system is also poised to create a trusted compute economy. Golem will release its first iteration in a few months under the form of a Computer-Generated Imagery (CGI) rendering system.
The development of a “dual AI economy” is worthwhile to follow because significant opportunities might arise in both camps. Additionally, the synergy with blockchain makes these initiatives twice as interesting. You never know what you’ll find in the long tail, especially if you’re busy trying to sum it all together…