INFORMS Open Forum

I just came across a short post by Radhika Kulkarni on his article "Synergizing AI and OR published in IJDB. – it mentions the Simon Article – Two heads are better than one.

I jotted down a short history of this Simon Article

The material was originally presented as the plenary Address at the TIMS/ORSA meeting, Miami, October 1986.  From the start of the 1980s Peter Norden (past TIMS president and founder of IJAA – aka Interfaces – aka Bulletin) was leading the charge in the INFORMS community to recognize the importance of DSS and then AI/Expert Systems.  Yes, DSS.  We take interactive applications for granted, this was not standard.  See Peter Keen, DSS – A modest proposal.  Peter Norden was the AI/Expert System topic editor for Interfaces.  He had access to many applied examples with IBM and IBM customers – in typical fashion practice was way ahead of theory.  Peter knew Herbert Simon, spoke with him after the presentation, and suggested he publish this material in IJAA (Interfaces).  Dr. Simon said he would give it some thought and asked for written short list of reasons - why Interfaces.  The list was provided a week later, and the paper appeared in July 1987.  The reader should note, it was the applied side of INFORMS that would drive this innovation.  The list can be found in the Herbert Simon material now available online.  It is an interesting read.

To understand the current AI rage and its impact on organizational decisions, it is important to just read this article but understand the 1980s.  Tools such as interactive graphical packages were emerging – GRAFSTAT for example.  A must "review" is Ed Feigenbaum's book The Rise of the Expert Company.  If Peter and his contemporaries were still alive – they tell you the key to success is to focus on the decision that need support, the decision premises (powerful Herbet Simon concept), and the complexity of the application area.  This includes your data supply lines.  Irv Lustig is an extremely articulate and powerful advocate for "data first", before you model.  I lean to an iterative investigation of decision and data.  If you start with the model – the project is DOA.

All these complex interactions and how dominating these drive succces are demonstrated each year in the Edelman Competition – spring INFORMS meeting.  Many of the 2025 finalists made clear that success comes from the intelligent interaction of data science and methods from Operations Research.  My 2025 favorite was Flipkart.

Consider applying.  Peter was critical of getting Edelman launched.  Irv has been critical in driving its continued success.  Having heard the original Simon presentation and tagging along with Peter when he chatted with Dr. Simon, I am certain Dr. Simon would want INFORMS to see this as one viewpoint, not the final viewpoint.

------------------------------
kenneth fordyce
director analytics without borders
Arkieva
Wilmington DE
------------------------------

Regarding the ongoing discussion on synergizing AI and OR/MS, I am somewhat confused about "current AI rage". Any discussion can be meaningful only if the terms being discussed are clearly defined and have the same meaning for everybody involved in the discussion.  

So, Operations Research (OR)/Management Science (MS) can broadly be defined as a methodology for developing  managerial decisions for efficient allocating of material, human and financial resources needed for making the best possible decisions within given constraints using various mathematical and computer simulation methods.  The main OR/MS tools are linear/non-linear deterministic or stochastic optimization, sequential decision-making in random environment, discrete-event, system dynamic or Monte-Carlo simulation, and some other mathematical techniques. This is a rather developed area with numerous examples of successful and practically relevant applications.

Now, what is artificial intelligence (AI)? Despite a lot of AI talking and AI rage, I have not seen yet a satisfactory definition that would help everybody be on the same page discussing the same thing. Indeed, one general and strong definition is that AI is the simulation of human intelligence by computers. However, it is not clear how to define human intelligence in the first place. 

On the other hand, there is a weak AI definition:  it is the narrow use of available AI technology, like machine learning or deep learning, to perform very specific tasks, such as playing chess, recommending songs/purchases based on past history, or steering cars. Also known as Artificial Narrow Intelligence (ANI), weak AI is essentially the kind of AI we use today. It can be, say, Generative AI that is a kind of artificial intelligence capable of producing original content, such as written text or images, in response to user inputs or "prompts." Generative models are also known as large language models (LLMs) because they're essentially complex, deep learning models trained on vast amounts of data that can be interacted with using normal human language rather than technical jargon.

So, how can OR/MS and AI defined above be synergized? And why such a synergy is needed at all? To enhance each other? In what way? 

After all, OR/MS is clearly defined and is based on a solid practically tested methodology. The AI (any version) is fuzzy and is focused on some narrow specific tasks with rather limited business outcomes and business needs. AI can help in performing some tasks, but, to me, it is mostly a nice toy rather than necessary methodology to make a business difference.

Thank you for the discussion.

------------------------------
Alexander Kolker
ge healthcare
MILWAUKEE WI
------------------------------

The referenced article and this discussion are both timely and germane to the design of decision intelligence systems.

While the OR/MS literature addresses a huge range of decision support and evaluation needs, it lacks a standardized method for tagging research and models to specific decisions. A standard 'decision model,' akin to the APQC Process Models or the SCOR models, would be of significant value. One might even imagine an 'INFORMS Decisions List.'

The adoption of such a decision list is likely to be hindered by a lack of familiarity. Systems dynamics (boxes-and-arrows diagrams "lit up" with data visualizations) and simulations could serve as effective bridges for businesspeople to overcome this barrier.

The challenge of encoding Simon's 'decision premises' is non-trivial and likely represents a market opportunity for experimentation in creating viable decision-support systems. At a minimum, variations in decision premises will necessitate different decision lists for different industries.

Currently, discussions about AI often serve as a proxy for discussions about LLMs. I think LLMs can serve as a bridge for people to understand and utilize decision-intelligence functionalities. In my work, we call this bridge "interaction intelligence" or a user-experience controller.

Numerous other non-LLM AI algorithms are employed in optimization, decision support, evaluation, simulation, forecasting, diagnostics, etc. However, these are so deeply intertwined and overlapping with traditional OR that, in my view, it is not practical to draw a clear distinction between them.

------------------------------
Rahul Saxena
RevInsight.com
------------------------------

Original Message:
Sent: 07-12-2025 13:44
From: Alexander Kolker
Subject: A bit of History in Simon's Article on Two Heads are better than one

Regarding the ongoing discussion on synergizing AI and OR/MS, I am somewhat confused about "current AI rage". Any discussion can be meaningful only if the terms being discussed are clearly defined and have the same meaning for everybody involved in the discussion.  

So, Operations Research (OR)/Management Science (MS) can broadly be defined as a methodology for developing  managerial decisions for efficient allocating of material, human and financial resources needed for making the best possible decisions within given constraints using various mathematical and computer simulation methods.  The main OR/MS tools are linear/non-linear deterministic or stochastic optimization, sequential decision-making in random environment, discrete-event, system dynamic or Monte-Carlo simulation, and some other mathematical techniques. This is a rather developed area with numerous examples of successful and practically relevant applications.

Now, what is artificial intelligence (AI)? Despite a lot of AI talking and AI rage, I have not seen yet a satisfactory definition that would help everybody be on the same page discussing the same thing. Indeed, one general and strong definition is that AI is the simulation of human intelligence by computers. However, it is not clear how to define human intelligence in the first place. 

On the other hand, there is a weak AI definition:  it is the narrow use of available AI technology, like machine learning or deep learning, to perform very specific tasks, such as playing chess, recommending songs/purchases based on past history, or steering cars. Also known as Artificial Narrow Intelligence (ANI), weak AI is essentially the kind of AI we use today. It can be, say, Generative AI that is a kind of artificial intelligence capable of producing original content, such as written text or images, in response to user inputs or "prompts." Generative models are also known as large language models (LLMs) because they're essentially complex, deep learning models trained on vast amounts of data that can be interacted with using normal human language rather than technical jargon.

So, how can OR/MS and AI defined above be synergized? And why such a synergy is needed at all? To enhance each other? In what way? 

After all, OR/MS is clearly defined and is based on a solid practically tested methodology. The AI (any version) is fuzzy and is focused on some narrow specific tasks with rather limited business outcomes and business needs. AI can help in performing some tasks, but, to me, it is mostly a nice toy rather than necessary methodology to make a business difference.

Thank you for the discussion.

------------------------------
Alexander Kolker
ge healthcare
MILWAUKEE WI
------------------------------

I agree with Rahul Saxena's statements: "Currently, discussions about AI often serve as a proxy for discussions about LLMs....Numerous other non-LLM AI algorithms are employed in optimization, decision support, evaluation, simulation, forecasting, diagnostics, etc. However, these are so deeply intertwined and overlapping with traditional OR that, in my view, it is not practical to draw a clear distinction between them.".

In addition, the referenced article "Synergizing AI and OR.." includes INFORMS 2024 Fellows AI Survey Questions. However, the AI term in the survey was not clearly defined at the beginning; therefore, participants could interpret this term differently in their answers.

------------------------------
Alexander Kolker
ge healthcare
MILWAUKEE WI
------------------------------

Original Message:
Sent: 07-13-2025 02:33
From: Rahul Saxena
Subject: A bit of History in Simon's Article on Two Heads are better than one

The referenced article and this discussion are both timely and germane to the design of decision intelligence systems.

While the OR/MS literature addresses a huge range of decision support and evaluation needs, it lacks a standardized method for tagging research and models to specific decisions. A standard 'decision model,' akin to the APQC Process Models or the SCOR models, would be of significant value. One might even imagine an 'INFORMS Decisions List.'

The adoption of such a decision list is likely to be hindered by a lack of familiarity. Systems dynamics (boxes-and-arrows diagrams "lit up" with data visualizations) and simulations could serve as effective bridges for businesspeople to overcome this barrier.

The challenge of encoding Simon's 'decision premises' is non-trivial and likely represents a market opportunity for experimentation in creating viable decision-support systems. At a minimum, variations in decision premises will necessitate different decision lists for different industries.

Currently, discussions about AI often serve as a proxy for discussions about LLMs. I think LLMs can serve as a bridge for people to understand and utilize decision-intelligence functionalities. In my work, we call this bridge "interaction intelligence" or a user-experience controller.

Numerous other non-LLM AI algorithms are employed in optimization, decision support, evaluation, simulation, forecasting, diagnostics, etc. However, these are so deeply intertwined and overlapping with traditional OR that, in my view, it is not practical to draw a clear distinction between them.

------------------------------
Rahul Saxena
RevInsight.com
------------------------------


Original Message:
Sent: 07-12-2025 13:44
From: Alexander Kolker
Subject: A bit of History in Simon's Article on Two Heads are better than one

Regarding the ongoing discussion on synergizing AI and OR/MS, I am somewhat confused about "current AI rage". Any discussion can be meaningful only if the terms being discussed are clearly defined and have the same meaning for everybody involved in the discussion.  

So, Operations Research (OR)/Management Science (MS) can broadly be defined as a methodology for developing  managerial decisions for efficient allocating of material, human and financial resources needed for making the best possible decisions within given constraints using various mathematical and computer simulation methods.  The main OR/MS tools are linear/non-linear deterministic or stochastic optimization, sequential decision-making in random environment, discrete-event, system dynamic or Monte-Carlo simulation, and some other mathematical techniques. This is a rather developed area with numerous examples of successful and practically relevant applications.

Now, what is artificial intelligence (AI)? Despite a lot of AI talking and AI rage, I have not seen yet a satisfactory definition that would help everybody be on the same page discussing the same thing. Indeed, one general and strong definition is that AI is the simulation of human intelligence by computers. However, it is not clear how to define human intelligence in the first place. 

On the other hand, there is a weak AI definition:  it is the narrow use of available AI technology, like machine learning or deep learning, to perform very specific tasks, such as playing chess, recommending songs/purchases based on past history, or steering cars. Also known as Artificial Narrow Intelligence (ANI), weak AI is essentially the kind of AI we use today. It can be, say, Generative AI that is a kind of artificial intelligence capable of producing original content, such as written text or images, in response to user inputs or "prompts." Generative models are also known as large language models (LLMs) because they're essentially complex, deep learning models trained on vast amounts of data that can be interacted with using normal human language rather than technical jargon.

So, how can OR/MS and AI defined above be synergized? And why such a synergy is needed at all? To enhance each other? In what way? 

After all, OR/MS is clearly defined and is based on a solid practically tested methodology. The AI (any version) is fuzzy and is focused on some narrow specific tasks with rather limited business outcomes and business needs. AI can help in performing some tasks, but, to me, it is mostly a nice toy rather than necessary methodology to make a business difference.

Thank you for the discussion.

------------------------------
Alexander Kolker
ge healthcare
MILWAUKEE WI