An OpenAI gym-based, economic simulation, with a welfare-aware AI policy designer
Despite the phenomenal developments which economics and related fields have undergone since the fundamentals were ideated, there is still some level of reluctancy by society to fully reflect this progression, sometimes resulting in being blinkered by GDP (1, 2) or the bottom line (3), though these are becoming significantly dated - especially when discussed in the context of climate change or sociological evolution.
The impact of focusing on these areas is starting to be observed by organisations, though there is significant challenges still to overcome (4), due to indications from the market that this is becoming increasingly important - however due to the imperfect information the population (and therefore the free market) operates on and the difficult of empathising with future generations (5, 6), there is potential that this action is lagging behind the requirements (especially ecologically) of the world.
As will be widely acknowledged this can and will have monumental impacts on almost all of life as we know it in years to come; changing ecosystems, impacting farmers, forcing migration, and having severe economic impacts (7 - 15). Not only this, but if this is not avoided economists, sociologists, and policy makers, amongst other individuals in positions of leadership and power, will have fundamentally failed in their honourably spirited endeavours of making the world a better place, though of course the diligent work by these communities will not let this happen.
In combination with this investment, we are in a period of opportunity due to the partial temporary reduction in pollution the pandemic has created (16 - 18) and with the new initiatives of building back better (19 - 21), its possible that we could use this unique coalescence to create impactful changes that will make the world more sustainable. However, undertaking this transformation successfully will require the use of all resources at our disposal to overcome one of the biggest long term challenges which our species will ever face in order to avoid being annihilated by our own success and productivity.
This will mean coming up with solutions which we have previously been unable to think of, using tools we previously have never used, and collaborating across disciplines in ways which so far have not been seen. This will especially include the implementation and use of techniques on the leading edge of fields with immense potential, such as artificial intelligence, or more specifically reinforcement learning using transformer models.
Although the field is relatively new and impossibility brilliant, current accessible economic simulations fall fowl of the limited orientation discussed in paragraph 1 (22 - 27). Though due to the open source nature, and supportive, active coding community, there is significant potential to build on previous work, combining it with leading research from other fields (15) and creating something which can help to us to identify how to solve these problems at hand.
And hopefully in our life times.
1 https://ec.europa.eu/environment/beyond_gdp/index_en.html
3 https://www.ft.com/content/30b3b8d2-f014-11e9-ad1e-4367d8281195
5 https://brill.com/view/book/edcoll/9789004398122/BP000014.xml
6 https://policy.friendsoftheearth.uk/sites/files/policy/documents/2019-02/empathy-effect.pdf
7 Warming-driven migration of core microbiota indicates soil property changes at continental scale - https://www.sciencedirect.com/science/article/abs/pii/S2095927321000578
8 The diel vertical migration of the nuisance alga Gonyostomum semen is controlled by temperature and by a circadian clock - https://www.sciencedirect.com/science/article/pii/S007595111930194X
9 The relationship between algal blooms, fish kill incidents, and oxygen depletions along the omani coast - https://www.researchgate.net/publication/286197049_The_relationship_between_algal_blooms_fish_kill_incidents_and_oxygen_depletions_along_the_omani_coast
10 Southward migration of the austral limit of mangroves in South America - https://www.sciencedirect.com/science/article/abs/pii/S0341816220303258
11 Global Warming Prediction Model of Fish Migration Based on ArcGIS - https://www.e3s-conferences.org/articles/e3sconf/abs/2020/64/e3sconf_icrteg2020_01004/e3sconf_icrteg2020_01004.html
12 Global warming modifies long-distance migration of an agricultural insect pest - https://link.springer.com/article/10.1007/s10340-019-01187-5
13 Natue Climate Migration - https://www.nature.com/collections/dagebcjjai
14 International Journal of Climate Change Strategies and Management - Special Issue: Climate change, migration and displacement of populations - https://www.emerald.com/insight/publication/issn/1756-8692/vol/10/iss/1
15 Virtual Seminar on climate economics - https://www.frbsf.org/economic-research/files/Rossi-HansbergSlides-VirtualClimate.pdf
16 https://www.nature.com/articles/s41558-020-0883-0
18 Near-real-time monitoring of global CO2 emissions reveals the effects of the COVID-19 pandemic - https://www.nature.com/articles/s41467-020-18922-7
20 https://www.buildbackbetteruk.org/
21 https://www.who.int/mental_health/emergencies/building_back_better/en/
22 https://www.tandfonline.com/doi/abs/10.1080/00220480309595209
23 https://sim4edu.com/sims/20/Lengnick-Baseline-Economy-1 The above journal describing and following web implementation of a basic Macro Economic Simulation, though the measures are only average price, average wage and % employment.
24 https://github.com/salesforce/ai-economist
25 https://arxiv.org/abs/2004.13332
26 https://www.brookings.edu/research/can-ai-model-economic-choices/
27 https://www.technologyreview.com/2020/05/05/1001142/ai-reinforcement-learning-simulate-economy-fairer-tax-policy-income-inequality-recession-pandemic/ The above describes one of the pioneering uses of AI to model economic policy. However, there are significant limitations in the economic reality of their code, since they almost entirely use just static wealth (rather than wealth distributed over time) as an indicator of not only utility, but also of welfare and equality. In addition, the size of the model, only 4 agents, is quite limiting, and prevents tangible insights from being derived.