The Singularity Sneaking Up Behind Us
Cloud AI, Edge AI, Fog Computing and Autonomous Systems are all part of the day to day discussion on the future of Artificial Intelligence but so far the real progress delivering the vision of ambient intelligence has been fairly limited.
Many have taken this slow progress as proof that AI hype is unwarranted. Others see this as proof that there is time for societies to gradually adapt to new work and life paradigms. Ray Kurzweil explained that while change appears to be linear, in fact, change is an exponential process in his seminal essay “The Law of Accelerating Returns”.
With this in mind, the current observations of organic or arithmetic change in AI advancement are likely to be flawed, hiding the true exponential changes well underway. This paper will evaluate the current technology and societal trends to establish a benchmark rather than evaluating compute power, speed, energy consumption, etc. The goal is to establish how close we are to the singularity.
Testing Our Understanding of AI Adoption
We will look at the current impacts of AI on the average developed world citizen (Trailing Indicators) as well as recent or emerging developments (Leading Indicators). This approach provides a sort of dead reckoning to establish the current position.
Trailing Indicators
The trailing indicators are easily observable and quantifiable social changes that we are seeing today. They are “trailing” as the technology that enables them has already been developed to some degree, and the early effects are visible today. Trailing indicators include:
- The dramatic growth of and reliance on Internet search is resulting in a technology-driven form of “Distributed Cognition”. While humans have always sought to offload cognitive workloads to social networks (asking friends, structured roles and responsibilities, etc.) and the environment (post-it notes, road signs, etc.), researchers are suggesting that the internet and the AI that makes it easily searched is changing the way that our brains are wired (World Economic Forum, 2016)
- The rise of the “Recommendation Engine” is changing nearly every aspect of modern life, and determining the outcomes of traditionally human endeavors like shopping, dating, choosing entertainment and news content, route finding, and finding employment. Parul Pandey posted a good deep dive on this topic. The degree of hidden influence that these algorithms wield is significant and growing.
- Growth of “Digital Workers” or Bots as detailed in “New World Order of the AI Economy” is rapidly changing the workplace. There are already millions of unattended digital robot workers, and nearly half of jobs in the United States are predicted to be automatable ( Policy Prescriptions for the AI Social Contract, 2019). These digital workers use Vision AI and Natural Language Understanding to read, Machine Learning optimized decision making, and Text to Speech AI to communicate. Think this is science fiction? Just ask your Amazon Alexa this question “Alexa — Are you Skynet?”
- Growth of traditional robotics into new fields including self-driving cars and other autonomous vehicles and drones, social robots, and specialized platforms that can automate work that has been difficult to automate like food service.
Leading Indicators
Leading indicators are early or pre-emergent technology innovations expected to build on and expand the impact of AI in society. These include:
- The Standardization of AI hardware, architectures, and Supercomputers are setting the stage for faster inference, in new parts of the network or AI fabric. Recent examples including:
- The Openfog Consortium in partnership with the IEEE Communications Society launched a new standard called IEEE 1934. This standard enables the idea of a continuum of intelligence from edge devices up thru to the cloud. Further, it enables near edge AI fabrics to be formed where devices can share resources to execute inference workloads that were previously impossible. Recently the OpenFog consortium merged with the Industrial Internet Consortium which will drive more adoption and innovation in this area.
- AI model portability thru partnerships like Open Neural Network Exchange Format ( https://onnx.ai/ ) led by cloud leaders including Microsoft, Facebook and Amazon Web Services. Already this effort is breaking down barriers that have previously slowed AI research and deployments
- The Supercomputer arms race is accelerating as leaders in the field race to deliver Exascale performance. The most recently published performance place the United States (IBM Summit) at the top 2 spots with 143.5 and 94.6 petaFLOPS with China close behind at 93 petaFLOPS. HPE’s acquisition of Cray is likely to spark a new round of competition and innovation. This will accelerate AI training and Artificial General Intelligence (read Skynet) workloads
- Quantum Computing is still very early, but the promise is so great that we need to factor it into any discussion on the future of computing and AI. If this technology can be made practicable then the timing of the “singularity” jumps to the near term
Conclusion for the Future of AI
Most researchers have focused on the AI Singularity concept as the point in time where a supercomputer or AI exceeds the power of human cognition effectively ending the “Human Era”. The research is often focused on Moores Law, algorithm maturity, and the idea of the emergence of artificial superintelligence.
Wikipedia provides this definition “The technological singularity … is a hypothetical future point in time at which technological growth becomes uncontrollable and irreversible, resulting in unfathomable changes to human civilization” ( Link to definition).
The trailing indicators already point to changes in human society that would be considered unlikely 10 years ago, and unfathomable to an observer 20 years ago. So under this definition, we are well into the singularity. Researchers are now documenting changes in human cognition, social function, and our institutions are under immense pressure to adapt to the speed of change.
The more extreme definition of the singularity “ nonbiological intelligence will match the range and subtlety of human intelligence” (Singularity Q&A, Kurzweil, 2011) is still challenged on a number of fronts. Exascale supercomputers will be needed to move to the next phase since the current cloud infrastructure is poorly suited for the AGI workloads. Green power production for supercomputers is another practical consideration that will need to be addressed so we do not end the human era with carbon emissions. Organizations like Open AI are pushing the boundaries and creating the foundation for this “nonbiological intelligence” with a goal of ensuring safety for humankind.
So we are navigating an interesting sea, full of both icebergs and opportunity. Keeping in mind the illusion of long term exponential growth looking linear in the near term, we must be aware that we are likely already well into the singularity. Its gravitational power is bending our perceptions of reality, history, and indeed our biology.
About the author:
Matt Vasey is a Senior Director responsible for AI Business and Corporate Development at Microsoft. He is focused on expanding the ecosystem of technology partners, standards bodies, and other innovation enablers that are required for the new generation of AI Applications, Services, and Systems that serve both individuals and businesses. Technology interests and expertise include Cognitive Workplace Automation, Robotics, Mixed Reality, Virtual Assistant Capabilities, Vision AI, Content Intelligence, and Edge/Fog AI.
In addition to his work at Microsoft, he serves as the Chairman of the OpenFog Consortium, Member of the Steering Committee of the Industrial Internet Consortium, Board member at the OPC Foundation, and on AI advisory boards at A3 Automation, Myplanet, and startups in the AI and IoT field.
Connect with Matt at https://www.linkedin.com/in/mattvasey/
