Subscribe to DSC Newsletter

Someone asked a question in a LinkedIn forum on the distinction between Decision Science and Data Science.

My 2 cents. Others may disagree, given their particular orientation. Given a choice, we are in Decision Science domain; Given a firehose we are in Data Science domain. I would suggest that Decision Sciences ends where Data Science begins. Decision Scientists don't necessarily work with Big Data. Data Scientists are specialized to work with Big Data (or recognize when something isn't truly Big Data) and all the problems associated with that domain.

Decision Scientists build decision support tools to enable decision makers to make decisions, or take action, under uncertainty with a data-centric bias. Traditional analytics falls under this domain. Often decision makers like linear solutions that provide simple, explainable, socializable decision making frameworks. That is, they are looking for a rationale. Data Scientists build machines to make decisions about large-scale complex dynamical processes that are typically too fast (velocity, veracity, volume, etc.) for a human operator/manager. They typically don't concern themselves with whether the algorithm is explainable or socializable, but are more concerned with whether it is functional, reliable, accurate, and robust.

When the decision making has to be done at scale, iteratively, repetitively, in a non convex domain, in real-time, you need a Data Scientist and lots of compute power, bandwidth, storage capacity, scalability, etc. The dynamic nature of the generating process which leads to high volume, high velocity data, in my mind, is the differentiating factor between the two domains.

The transition from manually input data sources to sensor-driven real time data sources is the underlying theme of the "Internet of Things" and this is especially true with the "Industrial Internet of Things" with complex machines interconnected with hundreds of sensors relaying data continually. The underlying math may be somewhat sharable, but doing it at scale, at velocity, etc. requires an end-to-end approach, and a greater emphasis on high speed algorithms. This is particularly true when you are talking about IoT, and especially IIoT, where you do not want humans to be making decisions "on the fly".

A Decision Science problem might be something like a marketing analytics problem where you are segmenting a customer base, identifying a high margin target market, qualifying leads, etc. Here the cost of a bad 'decision' is relatively low. I view Decision Science from the perspective of "decision making under uncertainty" (see Decision theory) which suggests a statistical perspective to begin with. A Data Science problem might be more like "How do I dynamically tweak a Jet engine's performance in flight to ensure efficiency/uptime during flight, to achieve stable and reliable output, to optimize fuel consumption, to reduce maintenance costs, and to extend the useful service life of the engine?" The cost of a failure in flight can be pretty high. Here, you would want to have a reliable machine making those computations and decisions in real time, relying on both ground and in-flight streaming data for real time condition monitoring.

In reality, practitioners tend to be "Agnostic" Scientists that are transferring skills from one domain to another, with varying degrees of comfort with the different tools out there. However, a Data Scientist is more likely to have a diverse background that spans multiple disciplines including statistics, computing, engineering, etc. In these examples, the decision maker is a key differentiating factor in my view. A Decision Scientist typically provides a framework for decision making to a human being; a Data Scientist provides a framework for decision making to a machine.

When machines become more human, this distinction may be called into question, but for now, I think it works.

What do you think? Did I just give you a choice? Or, did I just deluge you with data?  Your call.


Note: In the above marketing analytics example, I am not referring to clickstream data; rather, I refer to historical records stored on an RDBMS somewhere.

Views: 14622


You need to be a member of Data Science Central to add comments!

Join Data Science Central

Comment by Neil Raden on March 5, 2015 at 8:14am

And one more thing - informing decisions is quite different from making them.

Comment by Neil Raden on March 5, 2015 at 8:13am

Perhaps the most poorly-understood area of analytics/data science etc. is the whole area of decision-making. It's about 10% math, 25% technology and 99% everything else (thanks Casey Stengel for the quip). To me a decision SCIENTIST is a researcher studying how decisions are made, not someone building BI or DSS systems. If the term "data scientist" is an inflation (and I believe it is), then DECISION scientist as used here is hyper-inflation. 

Comment by Anthony Barnes on February 17, 2015 at 7:56pm

I have worked for a long time with one of the top machine learning experts in the world. He has always told me that it was my job to understand the problem and the use of the solution as it interfaced with the real world; that he could not waste the time to understand the application of his math while keeping up with the changes in the math itself. I consider him a data scientist, and myself a decision scientist. I often had to understand the nuance of the validation data and the sensing environment to ensure the math did not give an erroneous answer. I have had to use my background in sensor design as well as training in economics and statistics to properly use his advanced algorithms and feature selection procedures. Perhaps there are three steps in this problem - algorithm designer, data scientist (applied), and decision scientist or "interfacial consultant". I need to understand what the math can do as well as what was actually done in order to perform my delivery of the decision models to the end users.

Comment by Vincent Granville on February 15, 2015 at 8:40pm

The type of expertise (decision science versus data science) depends much more on the person, than her job title. Job titles nowadays mean very little. Heck, anyone can call herself president, lawyer, doctor, owner or unemployed. It does not mean that these people can deliver according to their job title.  

Comment by Pradyumna S. Upadrashta on February 15, 2015 at 7:34pm

"But a decision scientist has a deeper understanding of the consequences and uncertainty inherent to the decision situation he or she is dealing with."  ...How so?

Comment by Vincent Granville on February 15, 2015 at 3:28pm

It also depends on the size of the organization. In small companies, decision scientist and data scientist is the same person. I wear both hats, though I currently call myself data scientist.

Comment by Armin Trujillo-Mata on February 15, 2015 at 2:00pm

I would say that decision scientists work with the information that data scientists collect and analyze. But a decision scientist has a deeper understanding of the consequences and uncertainty inherent to the decision situation he or she is dealing with. I mean, decision science begins where data science ends.

Comment by Baguinebie Bazongo on February 13, 2015 at 1:48pm

Thank you for these calrifications

Comment by Pradyumna S. Upadrashta on February 12, 2015 at 5:37pm

I can partly agree.  However, there are ways to be 'close to' without ever really being close at all; for instance, take two particles in space separated by some negligible distance. We might say that they are 'close to' one another; yet, if we add in the time dimension, and I reveal that one particle existed 100 years prior to the other, they don't look so close any more.  You could say that this is unfair because my metric changes when I introduce additional dimensions; but, I think it illustrates my basic point that the difference may not be functional but ideological (if we accept 'mindset' as another dimension to evaluate these two POVs).  

A decision scientists world is based on the existence of an underlying linear, gaussian, stationary process that we can touch and feel, and from which we can arrive at great general truths.  A data scientist does not inherently *need* these types of comforters.  So, in that way, I question whether the fundamental ideology of a decision scientist is compatible with that of the data scientist.  

A decision scientist spends the majority of their time reducing complex problems into things that a human being can act on.  A data scientist gives up on humanity altogether as they do not try to reduce the problem into something we can understand, they are quite comfortable to live in the "black box".  So, strategically, the decision scientist is more likely to add the human element (an ingrained bias?) back into every business process and deployment.  The ML practitioner (my idea of a data scientist) tries to remove the human element as much as possible, trusting in the ability of the automated ML layer to continually adapt, learn, and grow to outperform the human element by leaps and bounds.  

Consider that a linear regression is just a special case of ML in which the underlying process is assumed to be stationary, gaussian, etc. so that you never really have to re-train the model, if your assumptions prove true.  One set of weights to rule them all.  An ML model does not have to care about the underlying generative process, because it is adaptive to new data.  A decision scientist that is used to static models requiring constant validation would be uncomfortable "letting go" enough to trust in the machine to learn, change, adapt and move on.  In a way, they live in the world of the "locally linear" process at all times, and always remain true skeptics when it comes to learning and generalization.  

They may be identically proficient in the mathematics; but I wonder that when it comes time to deploy the Algo., would they, in the deepest recesses of their hearts, find philosophical reasons to re-introduce the human layer to take the perceived "machine risk" out of the process?  If so, I don't see them as true 'strategists' if their entire strategy would be to walk-back all the real gains from an ML approach.  

I am certainly being a bit too broad here in my brush. However, I would say that there is a certain level of deep conviction that ML practitioners have in automation, that a non practitioner may never understand.  Some processes are by design too fast for human interference.  There are yet other processes that are just slow enough that a human decision maker could run interference (not necessarily improving the process by doing so).  For that subset of processes, the difference in their ideologies would probably come roaring out, and heavily influence their strategic outlook.  

So, I hesitate to equate the decision scientist to a strategists role.  Perhaps, it sort of long-windedly points back to your point about statisticians in earlier posts.  Perhaps I dice too finely. 

Comment by Vincent Granville on February 12, 2015 at 9:15am

I make the following distinction: strategic versus tactical data scientist. My strategic data scientist is close to your decision scientist, I believe, while the tactical data scientist is more like your data scientist. Bernard Marr might have been the first person to coin the word strategic data scientist. He uses the word operational data scientist for non-strategic data scientists.

© 2019   Data Science Central ®   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service