GPT Marmaray and Feminist Appr...

March 25, 2023

2023   ·   linkedin  

This is an automated Linkedin post reflection

GPT, Marmaray and Feminist Approaches:

I’ve been trying the new Microsoft Bing, in which GPT-4, the most advanced of the language models that has recently started to appeal to the end user with ChatGPT, is integrated. I did a simple test for this with interesting results.

First, I showed the station-based Marmaray travel data in the İstanbul Büyükşehir Belediyesi Open Data Portal as an example. I wanted her to compile the 3 Marmaray stations with the highest male and female passenger ratios for me and assign gender.

Female stations:

  • Yeşilyurt (63%)
  • Yeşilköy (62%)
  • Florya Akvaryum (61%)

Male stations:

  • Tersane (75%)
  • Kaynarca (74%)
  • Güzelyalı (73%)

Then I asked her to compare the development indexes of the districts where these stations are located and to assign gender again. She reached these values ​​herself by finding the SEGE 2022 report. I say herself because there seems to be a “someone” who, besides being able to talk like a human being, can make instant research, evaluate and comment.

Female districts (0.809):

  • Bakırköy (0.857) x2
  • Küçükçekmece (0.713)

Male districts (0.670):

  • Tuzla (0.661)
  • Pendik (0.674) x2

Then, without needing any further intervention, she made the following comments to me:

“According to this result, the average of the development scores of the female districts is higher than the average of the development scores of the male districts. This shows that female passengers prefer stations in more developed and high quality of life districts. The results obtained from this analysis reveal the spatial differences of gender distribution in rail systems and show the relationship of these differences with socio-economic development. These results may contribute to evidence-based policy design processes for the planning, operation and improvement of urban transport systems.”

My own interpretation is that ChatGPT, which uses the free version GPT-3.5 infrastructure and has training data up to 2021, can provide very convincingly fabricated information, whereas Bing can overcome this problem by performing simultaneous searches with better GPT-4 engine. However, since the “hallucination” habit of language models remains, it is necessary to see these tools as assistants without leaving the steering wheel, instead of using them as the only sources of information. Although it is a short way to access information for now, the user should still draw the route. In this example, I directed the raw data available according to the simple method in my mind.

#got #bing #data #planning