Hash: Do we see a new player pushing simulation into the mainstream?
Let me introduce you to ?simulation town“. A small place on the outskirts of ?data science city“ and very close to ?Excel analysis megalopolis“. So far, it is a bit sleepy here. Several shops sell their tools to a small number of inhabitants… But:
There is a new kid in simulation town. This does not happen too often, ?simulation town“ is still a bit of a secret place. But with reinforcement-learning and other toys on the rise, more and more people seem to want to move here.
Current inhabitants include various commercial tools like AnyLogic, Simio, Simul8 and Arena. Recently, additional open-source settlers moved in, such as Salabim, a Python library.
However, I was surprised to hear that none other than Joel Spolsky is now chairing a new startup that plans to move into simulation town and turn it into a city…
Joel is the co-founder of StackOverflow, the first stop for coders around the globe to find solutions to their problems. So he knows his stuff. And he decided to become it’s chairman.
What is it?
So the new kid is called ?Hash“. It is an open-source tool to build simulation models. So far, the website does not give away too many details so let me summarize them before going into some commentary…
First, the tool will be build from the ground up using functional programming (as opposed to object-oriented programming). They promise speed improvements of 10-50 times ?over existing best-in-class tools“.
Second, it will likely be a mix of process modelling and agent-based modelling, as far as I could tell. They are not explicit about this but the suggested application areas hint to it (supply chains, warehouses, call centers). It might even include system dynamics as they also want to solve for macroeconomics, ?political risk“ and others.
Third, they will create a marketplace around the tool: Users can create packaged solutions (agents…) and share or even sell them to others. They also will generate income through selling computing time directly.
My 50 cents
There are still many unknowns but here are some thoughts I had while studying their promises.
The good
These folks seem to understand the huge potential of simulations for the coming decade. And they also see that the existing market of tools is not prepared to serve thousands of data scientists that ?just need answers“.
Moreover, I welcome the marketplace idea: they aim to become the platform that everyone uses. Similar to Amazon (or anybody, these days), they do not want to be a competitor in a market but they want to be the marketplace. Great idea. Why is there no build-in app store for AnyLogic or Simio models yet?
Also, their (likely big) funding will allow them to offer things for free and grab market share. If they achieve similar functional capabilities than existing commercial tools, simulation town might loose inhabitants rather fast…
The bad
Unfortunately, very few details are shared so far. This is not necessarily bad but the promises are *very* big. How much they can convert into capability remains to be seen, I am happy to give the benefit of the doubt here. However, remember that many of the incumbents in simulation town have invested in capabilities for years.
The ugly
Although they seem to hire developers, this seems to be a two-man show currently (edit: I was told that the team is "quite a bit larger"). While I trust these guys are able to pull of developing a full stack of cloud technologies in a short amount of time, I would be impressed if they can create a simulation platform that is intuitive, feature-rich and cloud-ready in a short space of time. But I would love to be proven wrong.
Moreover, their (very good) website does not mention one problem that plagues anyone who wants to live in simulation town: Experience.
Even with the best tools, you still need people who are good simulation modellers. Great tools help us overcome the ?science“ of simulation. However, there is an element of ?art“: How do I build a model that represents reality well?
Hardcore data scientists will likely not even ponder this question and simply go for a brute-force approach: ?I just model it 100% accurately and crunch the numbers“. But this is doomed to fail, and has done so many times in the past.
Verdict
This is an exciting move and I am happy to see that simulation town will likely be disrupted, if only a little bit. I am sure these guys will push the boundaries of simulation technology and their open-access approach might draw in a new wave of modellers into our small simulation town.
However, I am skeptical that a new tool is really enough. As a community, we need to think harder how to help casual modellers to develop very good models. Otherwise, this might fuel a simulation hype that will be followed by a lot of disappointment because many modellers did a bad job using great tools.
You can read more blog posts about the world of simulation on my website benjamin-schumann.com
President and Co-Founder, GoldSim Technology Group
5 年I was excited about this but then went to the website. Home page says “Build simulations in minutes not days”. That is where they lost me. No real defensible simulation model of a complex system used to make real-world decisions can be built in minutes (very few can be built in days). As you point out, simulation modeling is an art, and takes lots of effort (that has nothing to do with the software at all). Models used to design multi-million dollar systems or design systems that impact health and safety are not built in minutes and only in days for relatively simple systems. Have any of these guys actually built a real simulation of a complex system that was used to make decisions? Based on that headline, I doubt it...
Superforecaster? at Good Judgment Inc
5 年Thanks for sharing Ben. Indeed there seems to be some serious activity in this area lately - notice the case of Skymind, who renamed themselves as Pathmind and changed course from offering a deep learning framework (DeepLearning4j) to an online platform for optimizing AnyLogic simulation models with Reinforcement Learning: https://pathmind.com/ (still in early access). FWIW, I would not worry that serious data scientists (themselves kind of artists, too) will enter the brute force trap - I guess your characterization of "hardcore" ones was not unintentional ;)
Principal Consultant | Simulation and Analytics Expert | Associate Professor | Author
5 年A very nice article, thanks for it Benjamin and also for the news about the "Hash". It’s a move that must be followed up. I especially liked your point about simulation being an “art”. In deed, if two modellers work independently to build simulation models of a system, the resultant models will totally be different. This is due to the modellers see and abstract the world differently and furthermore apply the tools for simulation model building with different styles. I am sure the "Hash" will consider all these. I have a similar idea which has not progressed well since I took action in 2013. It is the www.simulationmodel.com . It is aimed at being “a platform to share simulation models” which are built using COTS simulation software. It is not only intended for academic and learning purposes but also for consultancy services, and in the future, it can be used for trading and for assurance. Due to the lack of funding though, this start up idea is crawling at the moment. Just because of that, its existence has not been publicly announced yet.
Transport Planning, Modelling and Economics Consultant | Commentator | Collaborator | Adviser | Recruiter
5 年Thanks for this Benjamin,? Do you see them wanting to enter the Cities / Transport Planning market? is the simulation they are creating applicable? Fred
Senior Manager Data Science at Asurion
5 年I think the problem with simulation for me as a rather new data scientist is more complicated than just the right tool. I have applied it in multiple of my research papers and had them successfully published. However, when it comes to the industry, I am hesitant to start investing heavily into simulation to develop a model that performs nearly as good as other tools that I can use (or ideally outperform them). This is mainly because when you mention simulation, people get super excited and want to see that cool small model with graphics as a digital twin in the production, while it takes a lot of time and effort to get to that point. That time and effort will be ten folded as there are very few resources that can be easily accessed out there, and the optimal modeling approach will need hours and hours of time if you are not very experienced. While I could enjoy the research aspect of it while I was developing such models during my Ph.D., as a data scientist I can avoid all these pains by just using another tool and get really good results.?