ChatGPT has become one of the hottest topics in recent times, experiencing tremendous growth over just a couple of months.

Gaining an impressive 1 million users within just five days of its launch, ChatGPT has undoubtedly established itself as a major player in the field. Additionally, with recent unveiling of GPT-4, ChatGPT’s presence is expected to continue for the long term. However, concerns have been raised by experts and members of the public regarding the potential negative impacts of chatbot-AI technologies like ChatGPT, including the threats to privacy and the future of work. However, in this article we will explore how to embrace the opportunities presented by this new AI technology, when integrating it into your daily activities.

As model development revolves around conceptualizing, validating, and building statistical models, the possibility of AI assisting in these areas are very tempting. As a test case, we investigated how to create a Probability of Default (PD) model in Python. A PD model assesses the probability that a customer will experience a default based on a variety of risk drivers. To evaluate the advantages of using AI in constructing a PD model, we have explored two scenarios.

Using AI from a modelers’ perspective

The first scenario identified is where the modeler had a deep understanding of modeling, yet is less familiar with Python programming. In this situation, we investigated how AI could serve as a useful tool to quickly generate code to bring models to life. By asking clear and direct modeling questions, the AI was able to provide equally clear, concise, and executable code in response for the modeler to build a PD model without the need for extensive coding knowledge. We did find that:

• Some code suggestions were quite generic and still had to be adjusted to fit the specific dataset being used, such as input data formats that needed to be adapted, or adding an additional step to adjust the dataset to work with the suggested code.
• Correctly phrasing questions was crucial to receiving the desired outcome.
• In some cases, it would have been much quicker and easier if there had been prior knowledge of Python, or if one had asked a colleague for help. For example, when encountering an error that could not be solved using the AI, a colleague who experienced the same problem before could quickly diagnose the issue and help to resolve it.

Nonetheless, the advantages outweighed the following negatives:

• This technology provides a unique solution for modelers who are seeking to expand their coding skills by making the transition to a new language much smoother and faster (in this case from MATLAB/SAS to Python).
• The technology has the potential to level the playing field for individuals who are looking to enter the field of data science, regardless of their programming background.
• The efficiency and effectiveness of using AI can save valuable time and effort compared to traditional methods of learning a new programming language.
• With its ability to quickly generate executable code, it eliminates the need for extensive reading and studying, as well as the time commitment required for taking courses and attending training sessions.
• The AI could help to perform mundane tasks such as writing code comments, explaining the code functionality, creating test cases, rewriting code, etc.
• Arguably, this technology can also be a very useful tool for banks or companies looking to replace old legacy code where they lack the in-house knowledge or expertise to maintain the outdated code. However, it is important to note that replacing old legacy code can be a daunting task, and may require careful planning and execution to ensure success.

Using AI from a developers’ perspective

When the user is an experienced developer with extensive Python knowledge, AI could also be used for guidance in the methodological setup of the PD model. Having a solid understanding of Python ensures that one could leverage on the AI for its ability to quickly come up with a basic step-by-step guide on how to create the PD methodology, the necessary steps regarding data cleaning, transformation, and investigation and subsequently the model itself. However, we found that:

• When using AI for this purpose, it is especially important to have the necessary knowledge to make informed decisions about the answers provided. In other words, a background in modeling helps you to get the most out of using the AI, as it enables you to effectively navigate the model's output and make any necessary adjustments to ensure that the results meet your needs.
• We were missing out on the human interaction that often yields the most creative solution to the problem at hand. The back-and-forth discussions and brainstorming sessions with other people brings a dynamic energy to problem-solving and decision-making that is simply not present in (current) AI models. We missed the personal touch of having someone to ask: “Have you thought of this?” or “Have you tried that?” The human connection and collaboration are integral components of the creative/development process, and it is important to consider this trade-off when deciding whether or not to use AI in a particular scenario.
• AI models have certain constraints that need to be taken into account when using it. Its answer generation capabilities are extensive but still limited by the input it receives and the conditions under which it was trained. It is a probabilistic language model, answering in a way that has the highest probability of sounding correct, but it certainly does not imply that what the model says is factually correct.

Having observed this, using the initial step-by-step guide on ‘how do I create a Probability of Default model’ resulted in some quick and elementary modeling steps:

• Building further on the broad blueprint it had provided, with typical questions on the specifics, it could quickly elaborate on each individual step. It provided multiple methods to achieve some basic data cleaning, investigating, and transformation required for model calibration.
• In most situations, it provided multiple options or recommendations. However, it was still up to the user to make the final decision based on, for example, our specific dataset with its particular characteristics and challenges. In this way, the modeler’s capabilities still played a crucial role in the modeling and decision-making process. It is always a good idea to consult with colleagues who have experience and expertise in the relevant area. Complex problem-solving, critical thinking, creativity or understanding context may require a human touch and common sense that a machine simply cannot provide (yet).

Our vision

Our vision for ChatGPT and other Chatbot-AI in the field of modeling and Python is to bring a new level of efficiency and productivity. It is revolutionizing the way we work and can help experts in making it easier and faster to get their job done. When the user knows how to ask the right questions, and knows what type of answers it should expect, the AI proves to be a valuable asset. In some cases, however, the insights and feedback provided by a colleague may be more useful than those offered by AI. This is because personal experience, expertise and common sense can provide a deeper understanding of the problem at hand, allowing the colleague to provide more accurate feedback and answers to the questions.

As the use of AI continues to grow, it is becoming an increasingly popular tool for coding purposes. The key advantage is the ability to provide tailored solutions to specific coding questions, saving time and effort otherwise spent searching the web for answers. However, it is important to note that more complex and innovative solutions will still require the expertise and creativity of individuals.

Recently, there has been an outcry from tech leaders to put a pause in the training of powerful AI systems due to the potential risks to society and humanity, and the European Union also took an important step towards a new law that aims to regulate AI. It is important to remember that, although AI can be quite a handy tool, it should be used in a responsible and ethical way.

To conclude, while the new AI technology can certainly enhance learning and productivity, it is important to be aware of its limitations and potential drawbacks before using it. Nonetheless, when utilized properly, this technology can definitely assist to accelerate your learning-curve or boost your productivity!

How can Zanders help your organization?

Did you find this article interesting and do you have questions or need additional assistance? Our team of experts is ready to assist you in finding the solutions you need.

Please feel free to reach out to us to discuss your needs in more detail. Whether you’re looking for advice on a specific project or just need someone to exchange ideas with, we are here to assist.