A Data Science Central Community
The brave new world of machine managers
original blog post: http://sctr7.com/2014/10/07/manager-machine-analytics-artificial-in...
Advances in analytics and artificial intelligence increasingly encompass decision making traditionally associated with middle management. With business analytics systems increasingly able to make complex decisions, the traditional scope of management is being squeezed, forcing contemplation of the future scope of management practice.
As a result, management as a profession will increasingly need to become more creative and flexible in terms of asserting its value proposition in tension with emerging advanced decision technologies. As many traditional industrial age business functions become subject to disintermediation by automation and expert analytics systems, new industries and value propositions will need to be framed by an emerging generation of leaders.
This transition will take place over the next two decades, so those most affected will be early- and mid-career professionals. In the mid-term, it is clear that quantitative and technical fluency in working with analytics and machine intelligence proctored decisions will be an increasingly essential skillset for the modern manager. While there are certainly overblown pronouncements concerning the speed of this change, ‘shift’, as they say, is already happening. As per the adage about the frog boiling in slowly warming water, this is a trend to track over the course of years. The end result, though: boiled frog - the scenario to avoid for forward-looking management professionals.
Prevalence of artificial intelligence
Artificial intelligence (AI) as a term dropped out of favor in the 1990’s after failing, at the time, to live up to much of the hype it generated. To be fair, the high expectations were in part framed by unrealistic science fiction depictions of autonomous robots and super-intelligent, conscious computers resident in popular culture from as far back as the 1950’s utopian futurism movement.
However, the last decade has quietly and steadily seen the re-emergence and hybridization of a number of factors which are now on the cusp of delivering on the promise of ersatz AI. We are seemingly, according to some informed estimates, within three decades range of being able to realize the intelligent, interactive ‘agent’ as popularized by fictional vehicles such as Hal 9000 from 2001: A Space Odyssey (though hopefully less the psychotic inclinations).
Indeed, much of what Hal evidenced in the way of ‘intelligence’ is available today:
Hal was basically an aggregate of all these systems inter-operating, and thus, given sufficient effort, could be evidenced today as a composite.
Facilitating technical factors in this ‘rebirth’ of AI include:
This allows for learning systems such as IBM’s Watson, systems which gather, sift, test, and refine information and modify their resulting behavior. The addition of prolific low-cost and persistent sensors in the form of the Internet of Things, leads to the capacity to design and deploy sensing, ‘thinking’ (seemingly, albeit ‘artificial’), and acting systems.
Google search itself is not only an effective solution for web-based search and retrieval, but by relentlessly applying analytics to search patterns and results, it is a constantly refined, closed-loop learning system. As a cyclical model, albeit human mediated, Google search captures an ‘understanding’ of what people are interested in, how they seek information, and, therefore, how they ‘think’ to some degree. The ability to generate autonomous hypotheses - to reason - by extension, is a linked feasible capacity within sight.
The business analytics movement
Analytics is a practitioner movement combining several formal disciplines including BI, statistics, econometrics, operations research, machine learning, and data engineering. At its boundaries, analytics encompasses organizational decision management and decision automation.
We can consider analytics as a discipline interested in managing decisions as much as creating insights from data. It may be useful to think of business analytics as addressing analytics-focused decision making as a ‘frame’ around data analytics, which focuses on the application of technologies and formal analytics methods (as below).
A core implication of the analytics discipline is closing the gap between data analysis and insights. A robust analytics solution encompasses a process which includes gathering data, generating hypotheses, testing these hypothesis, reporting the results to guide a decision, and applying subsequent results to revise the decision model based on new data (i.e. to revise a machine learning model). This implies that robust analytics solutions are capable of systemic inductive reasoning, a mainstay of the scientific method and traditionally a facility associated uniquely with human creativity.
When a complex decision is sufficiently described and encoded as an analytics solution, it is a reasonable leap to automating the decision itself. As such, designing advanced BI reporting and alerts mixed with pattern discovery, diagnostics, predictive analytics, and prescriptive techniques is essentially a composite process whose capacity approaches that of human reasoning. Embedding a closed-loop, self-improving learning model at the core is an aspect which puts the ‘I’ into AI.
I am often struck by this notion when demonstrated advanced analytics and visualization: in realizing a robust recommendation system, we often, essentially, are presenting a decision engine itself. The results are increasingly demonstrably better than human judgment, which are beset with agency interests and behavioral decision biases. The upshot is that a sufficiently automated analytics solution questions whether the ‘manager’ is, strictly speaking, necessary in the process of identifying value driving insights, except as a steward.
To take a practical example, if a complex supply chain management system is able to autonomously specify an optimal transport and delivery route, why is the supply chain manager necessary, except in terms of addressing the formality of ensuring the results are applied? There is a point where the pilot of the airplane is no longer flying, but is there simply to reassure the passengers and to ensure that no person touches the airplane controls.
Rage against the machine
If an analytics system, as a decision process, is sufficiently delivered to a manager such that the decision becomes self-evident, does the manager not become in some sense obsolete, at least in terms of picking the best decision, especially when the system clarifies the formal proof and refutes objections at a level beyond the capacity of individual experts?
When I have raised this notion in the past with management colleagues, there is always a rapid flurry of objections. Blustery qualifications follow: “leaders will always be needed” and “machines only frame the decision, they don’t make it.” However, as cited, the advent of IBM’s Watson and systems which can autonomously develop and test research hypothesis increasingly call these objections into question.
At some point, do the refutations not risk sounding like histrionic objections against the, admittedly, emotionally disturbing, notion of obsolescence, if not a major assault on assertions of human uniqueness, creativity, and intelligence?
I myself tend to quickly leap to the human defense, having spent a good deal of my life studying and practicing the discipline of management. I deeply want human managers to not only be necessary, but essential. Yet, I must ask whether I am falling into the confirmation behavioral bias: selectively seeking evidence, and rejecting disconfirming evidence, to validate my deeply held belief that management is a uniquely human and thus unassailable practice.
Increasingly objections and qualifications are slipping under the relentless advance of AI and machine learning. As an example, a recent Wall Street Journal piece reported that IBM is prototyping a service where corporate executives can solicit a targeted interaction with Watson regarding business strategy. An example would be executives seeking guidance on feasible acquisition targets.
Much as Watson, via its internet access and natural language processing abilities, was able to handily trounce humans in the game show Jeopardy, it can similarly rapidly assemble a list of viable business acquisitions and crunch the financials to short-list a recommendation. This could be done in real time and would encompass supplying details documenting the assumptions, recommendations, and conclusions, including the extrapolated M&A financials.
The simple cost basis argument for computer expert decision making is clear: an hour ‘renting time’ with Watson versus hiring an expensive team of M&A consultants to generate essentially the same results. We can argue about quality, but conceivably this is an objection which will steadily fall away as the systems improve, much as chess playing computers quickly rose from mediocre playthings to beating grand masters.
Manager-machine uber alles?
At a recent Rotterdam School of Management (RSM) Leadership Summit on Big Data, an expert panel briefly discussed the implications of AI advances in terms of management. As an example, airline autopilots were raised as a domain where computer decision making surpasses human decision making. Similarly, with rapid advances in computer driven automobiles, such as Google car, we are now within generational sight of the obsolescence of human drivers. Simply, automobile accidents being a leading cause of non-health related mortality, prototype AI driver systems are increasingly demonstrably safer, as well as more efficient and sustainable.
Indeed, the future for airline pilots and motorists will increasingly become one in which humans may not be permitted to touch the controls or interfere with operation of a vehicle except under unusual circumstances.
When we relocate these type of advances in computer decision making to the sphere of management, we are forced to ask how much more complex an operating business is than an Airbus passenger airliner. Certainly domains such as strategy, innovation, and leadership are, at the moment, demonstrably human. Yet, if Watson can guide M&A, can it not also be configured for continuous managerial accounting monitoring, raising objections and guiding general tactics and strategy to optimize the operating business? Are not staffing decisions, growth targets, and budgeting all within scope?
Whither the manager?
If we take as a hypothesis that AI systems will increasingly disintermediate many traditional middle management decision functions in the coming three decades, we then must ask what will be left? Where should the new generation of managers invest their efforts and attention?
Some targeted thoughts on the future of management in the brave new world of analytics and AI mediation, including areas where humans will continue playing a central role:
The above examples are only a selection of areas where managers and entrepreneurs are most likely to continue to challenge themselves and grow outside the scope of growing automation and machine disintermediation. The themes are: integrated systems, deep social understanding, applying emerging technologies in innovative ways, and enhancing value by creating ecosystem platforms.
Choice towards the future…
The dark side of automation and disintermediation, unfortunately, is a much grimmer story. While the examples above have to do with pro-social goals and optimizing happiness, there are equal ‘commercial’ opportunities in the areas of security, policing, weapons, war, and social control. As technological development and globalization pressure and constrain traditional industries, affected communities will evidence tension and react negatively. Acting out and expressing social discontent can quickly lead to spiraling and cyclical violence and retribution between communities and authorities, a perverse ‘business cycle’ based on negativity.
Such cycles of violence and control can easily lead to situations whereby the weapons industry, terrorism, and security services create self-sustaining, self-reciprocating economic ecosystems. Just as a festival is driven by passion and people seeking fulfillment, a regional conflict is driven by animosity, hatred, violence, and revenge. Regional conflicts can degenerate into wars, terrorism, and generational struggles which inflict a terrible human cost.
Certainly AI and analytics can equally assist in the economic apparatus surrounding weapons production, security, and military action. Technology itself is not good or bad – it is put to particular good or bad uses by people. Thus, the choice is one for a new generation of mangers to make regarding the application of technology for economic development: will it be for maximizing human happiness and potential or a cynical cycle of control, violence, and oppression? Both can serve as engines of ‘profit’. Only one is self-sustaining.
We are heading toward a dangerous junction point where, ironically, our advancing technologies are disintermediating traditional forms of labor, plunging many into situations of economic uncertainty. A recent Economist article Technology isn’t working repeated the observation that the industrialized world is already experiencing spiraling negative effects from technical disintermediation. The article notes that disintermediation increasingly encompasses professional work: “technological advances are encroaching on tasks that were previously considered too brainy to be automated, including some legal and accounting work.”
The Economist article does qualify that there will continue to be top-tier opportunities, such as those cited previously: “over the next few decades demand in the top layer of the labour market may well centre on individuals with high abstract reasoning, creative, and interpersonal skills that are beyond most workers, including graduates.” The question becomes: will these ‘new generation’ professional elites focus their technical creativity and innovation on empowering new economic ecosystems for their brethren, or will it be easier to pursue the mantra of control, contain, and oppress using the very same mechanisms?
We must decide where human efforts will be focused while our traditional modes of economic interaction shift and transform beneath us. Technology grows in leaps and bounds while culture evolves slowly. If we treat disintermediation as a zero-sum game, in which we attempt to gain control of a shrinking iceberg, the future is likely to be one in which the quality of life declines and where our ‘work’ involves either struggling against control through disobedience or crime to gain a fair share of the economic pie, or working for those interests seeking to maintain control via a vast security apparatus. If we view the iceberg as being capable of renewal, we can allow our imagination to optimize startling new technologies to integrate social and economic systems, generating profit and maximizing happiness, producing more with less.