My previous (long) post attempted to make the case that the innovator has access to a far richer universe of data than the observable (even conceivably observable) data available to data science. The innovator taps the missing data: the null records that will never enter the data store for big data technology to exploit.
Even after allowing for the emergence of even superhuman artificial intelligence, I argue that the innovative human will still succeed to find and exploit vulnerabilities before the AI would. The reason is that the universe of missing data is far larger than the universe of observable data. The super-intelligent AI is overwhelmed by the magnitude of missing data, leaving plenty of space for human intelligence to make the discovery first.
After writing the previous post, I thought about how to describe the vastness of this universe of missing data. I then realized I already had written a good analogy: the vastness of missing data is like the vastness of empty space between planets representing observable data. In that post, I marveled at the scale of the nothingness between planets. That is comparable to the missing-data space available to discover innovations. This view of missing data gives me confidence that human ingenuity will remain competitive even against the most advance AI. Missing data resides in too much space for AI to discover before a human does.
In my previous post, I presented the case of the innovative criminal who only needs to get lucky once. The same applies to the entrepreneur who introduces a disruptive new business. An advanced AI technology will probably not hamper human innovation because the AI can not explore all of the possibilities that humans might discover first.
We do not yet have advanced AI and so I can not test my theory. I have confidence that humans will continue to innovate even in the presence of advanced AI in place with the intent to interdict that innovation. This will occur even if AI is so advanced it will beat humans every time when faced with a common problem.
An analogy may be chess games where even modern computers can compete at the grand master level. A human player can still win by playing the multitude of games that do not involve the computer.
The humans will succeed in solving a problem that the advanced AI has not yet considered. I assume a scenario that employs advanced AI to protect the established or status-quo of business or government. In this scenario, the innovative criminal will provide vivid reminders of the human capacity to surprise the establishment with unexpected innovations not yet predicted by that AI.
I have been writing several posts exploring a concept of government by data, or dedomenocracy. In the earlier posts, I proposed that government by data is strictly limited to working with observable data and knowledge-domain-independent statistical analytics. My concept of dedomenocracy retains a population-participation role as data scientists in the way I define the field as being skeptical about that data and analytics. Also, I wanted to exclude domain-knowledge from decision making in order to make faster discoveries of new phenomena. The access to vast amounts of observable data permits analytic consideration of facts of nature exposed by the data itself instead of relying on hard-coded domain-knowledge.
In my previous post, i conceded the introduction of advanced AI in dedomenocracy as a countermeasure against the innovative criminal. The innovative criminal is not constrained by observable data. Recent history demonstrates that the innovative criminal is capable of catastrophic dame. To combat the innovative criminal, I allowed for the introduction of advanced AI to explore the same missing data.
However, I concluded that there is simply too much missing data for AI to be effective at combating the innovative criminal. The innovative criminal can always find something before AI has a chance to find it first. Perhaps, it would be more cost effective to leave AI out entirely because we’ll still be suffering from the consequences of innovative crime at about the same frequency.
The key benefit of dedomenocracy is its agile responsiveness to new conditions. Although unable to prevent the human innovation, the more agile government can respond quickly to mitigate the effects, to prosecute the criminal, and to augment the data to observe future copycats. The demonstrated benefits of this agility is the key to the society’s acceptance of dedomenocracy.
Dedomenocracy will not be effective in preventing innovative crime, even if we allow the addition of superhuman AI. It will, however, be more effective in responding to the new instances. Dedomenocracy is agile because it is completely automated from sensor to decision. This type of government is not encumbered by human debate involving persuasion of human-understandable theories.
Dedomenocracy will be valuable despite its inability to predict human innovations even when we have access to the most futuristic technologies of data and AI.
This impossibility of even futuristic data technologies to predict human innovation challenges the deterministic theory of mental processes or psychology. This theory suggests that all human experience is wholly deterministic based on physical structures of the body, the available genetic code, and the specific history of the individual’s encounters with the social and physical environments.
All of the physical contributions to mental determinism is observable data. Modern medical scanners provide detailed observations of internal body structures. DNA sequencers permit cheap enumeration of individual’s genomes. Social-media and Internet of things provide detailed records of the environment encounters.
Much of this available observations is redundant. We only need a sample of the totality of possible observations in order to reconstruct the necessary influences on future human decisions. In addition, our recent experience with the power of multidimensional analytics informs us that we can make accurate predictions based on even fewer number of data samples in exchange for observations in a sufficient number of dimensions.
A dedomenocracy, theoretically, will have all of the information it needs to accurate predict all future behaviors of a deterministic mind. The system’s analytics will have sufficient information about the environmental inputs and internal structures to make a skillful prediction based on statistics alone. With big data and analytics, there is no need to build a human-understandable model of the physical and psychological processes that govern future before. Big data analytics should be able to make the prediction based on multidimensional statistics of extensive observations.
If the theory of a deterministic mind were correct, then we should expect that a futuristic dedomenocracy should be effective at preventing the innovative crimes. At a minimum, dedomenocracy should be able to prevent the most damaging innovative crimes and then predict the impact of less damaging innovate it knows can occur but it lacks resources to prevent.
This futuristic dedomenocracy offers a way to falsify the deterministic theory of mind by simple observation of innovative crime that occurs despite the power of the available data. Observations of successful innovative crimes that result in unacceptably damaging consequences, or that occur in unacceptably high numbers would be evidence that we lack sufficient data for the deterministic model. However, I am assuming a future where we will have observations of just about everything that is observable. If dedomenocracy continues to experience innovative crimes of unacceptable magnitude or frequency despite having access to everything that is observable, then the innovative part of the mind must be accessing something that can not be observed.
I have no doubt that the adherents of deterministic mind will find fault in this conclusion. We may never have technology that will record for every living human the microsecond resolution details of every molecule. There may not be enough rare-earths and energy on earth to implement such a magnitude of machinery.
I assert that big data analytics of multidimensional data does not need extensive data to come up with accurate predictions even if the individual dimensions contain coarsely sampled observations. Practical data science (especially of futuristic capabilities) should be sufficient to predict how an individual deterministic mind will behave. If despite that capability, we continue to be surprised by the innovative criminal, then I would doubt the validity (and certainly the practicality) of deterministic theory of mind.
Such arguments never really ever get resolved, but personally I already see doubt emerging in recent news of major crimes occurring despite our present investments in big data. Despite our huge investments in data collections and analytics over the past two decades, we have encountered countless dramatic surprises in areas supposedly being studied most closely: such as various states in the Middle east, North Africa, Southwest Asia, and Ukraine.
The hasty and sloppy evacuation of Yemen is the latest example of something that surprised us despite our access to bountiful information. Certainly, there had to be some expectation of risk but that assessment apparently underestimated the magnitude of the risk as evidenced by our inability to evacuate American citizens in that country. This is just one of many recent examples that raise my doubts that data can predict large-scale human innovations.
This failure may be a result of lack of maturity of our data technologies, but it seems to me that our technologies should be most mature for this area and for this type of outcome. It failed.