A Data Science Central Community
I just finished reading Nick Bostrom’s book titled “SuperIntelligence: Paths, Dangers, Strategies”. It was a tough slog to say the least. The gist of the book is that superintelligent machines (computers capable of doing what our human mind does, sans consciousness, only a 1000 times or more faster) are feasible and the consequences of creating them could be either the best or the worst thing ever to happen to humanity. This book argues that superintelligence will inevitably be the worst thing, our last invention, if one believes a second book by James Barrat called “Our Final Invention: Artificial Intelligence and The End of the Human Era”. Bostrom argues for safeguards to be built in to stop an intelligence explosion. Then again one can believe Ray Kurzweil who believes in a Singularity where humans and computers merge and all our problems are solved.
AI seems to have finally found its sea legs after several false starts. Of its four general areas of research (systems that think like humans, act like humans, think rationally or act rationally), systems that act rationally (or rational agents) is dominating AI research. A key component to the design and operation of rational agents is data though the importance of big data to the effective design and operation of a rational agent is not discussed at length or addressed explicitly.
Myself and a colleague have spent several years looking at the problem of autonomous systems, i.e., systems that combine big data, artificial intelligence, many types if sensors (aka the Internet of Things), implemented in various business contexts (retail, health, financial services, …). Certainly the current versions of autonomous systems such as driverless cars and unmanned aircraft will have a significant impact to current business models. However, it is clear that big data is the oxygen for these autonomous systems. Regardless of what AI technique used to build an autonomous system, it cannot exist in isolation of a significant (big) data architecture. And it is not just a single data set but data whose relevancy can be less than a second long (price of the Euro on the spot market) to decades long (climate). AI systems do not come with their own data unlike a human baby who then builds their big data architecture rapidly. AI has to be fed the data, trained in it, ingest rules on how to use it and maybe (but not always) trained on what to do when the data is corrupted. And like a baby it has to continually be fed data because how it reacts to certain stimuli will and does change with time. An AI high-frequency trading system needs to know that the correlations at the beginning of 2015 between the 30 year bond futures and the S&P 500 futures is different from what it saw a year earlier. It would lose money if it used the 2014 rule set in January 2015.
Bostrom makes the case that the superintelligence existential risk be managed by research guided and managed within a strict ethical framework. But Bostrom does not consider, at least explicitly from what I could see, a solution that is big data-centric. The excellent work done at the Machine Intelligence Research Institute (MIRI) is also geared towards research on safety issues related to the development of AI. However, a perusal of its web site (https://intelligence.org/) does not turn up any work being done to consider if safety is not in the algorithms but in the data used to make the AI usuable. If big data is the oxygen for superintelligence, can we cut off the oxygen supply when the existential risk begins to emerge? I know a few AI researchers will say that the superintelligence will be so smart that it will anticipate and replicate the data. I consider this argument a red herring and an intellectual dodge to gloss over problems that are difficult to solve.
Obviously this is a work in progress for us and I will write more on it. A closing thought. There is a silver lining here that also seems to be lightly discussed. Combining one or more human beings with an AI program that acts as an assistant appears to be a powerful paradigm that addresses this existential risk and the employment issue where robots replace humans. Robots might be best for an assembly line but in an emergency room? I think a Human + AI + Big Data is the winner there. And a Human + AI + Big Data could be the first line of defense in case the superintelligence does emerge.