Advertisement
X

Economic Survey 2025 Lists Challenges in Implementation of AI

The Economic Survey 2025 highlighted the requirement of large investment in research and development of a model and its application

Economic Survey 2025 Lists Challenges in Implementation of AI

The Economic Survey Report 2025 acknowledged the significant impact of artificial intelligence (AI) in the tech industry, considering it a breakthrough technology. However, the report undermined the viability of this technology, listing a set of challenges that are hampering its impact and utility.

Advertisement

In the context of the large language models (LLMs), the report stated that while the LLMs are capable of acing exams and achieving high test scores, the field is still far from having a model to come up with original, publishable research. The current standards of LLMs have decent small-scale utility but are not practically applicable in large-scale and high-impact roles.

“At its current stage of development, AI is more experimental as it is still finding its footing. This is not inherently negative, as it signifies innovation's curious and exploratory nature. However, from a practical standpoint, its experimental nature makes its real-world utility unclear despite the technology demonstrating impressive capabilities,” the report added.

Challenges in AI Implementation

Reliability Challenge- Giving the reference of a publication by McKendrick and Thurai (2022) named the non-negotiable nature of reliability, the report highlighted how AI is yet not reliable on high stake applications. It stated that having a 10% error rate on an LLM result for personal use would not have a high scale impact as it can be edited easily by the user. However, a self-driving car with the same error rate can be fatal and is not reliable for the application.

Advertisement

Infrastructure Challenge- The report stated that once the practicality of a technology is proven, a large infrastructure development is followed to implement it at a large scale. Since infrastructure development is a time-consuming process, the full potential of the technology cannot be harnessed till the support infrastructure is widely available. The development of AI and the rate of adoption is similarly going to depend on the availability of quality infrastructure and the pace of its creation.

Resource Challenge- The Economic Survey highlighted the requirement of large investment in research and development of a model and its application. It also highlighted the cost of training the AI models as it is becoming increasingly expensive as the availability of data is saturating and high-quality data acquisition costs are rising.

The report stated that “Training the first ‘Transformer’ model developed by Google, which laid the foundation for ChatGPT, cost around $930. In stark contrast, training OpenAI’s GPT-4 cost the company $78.4mn, while the costs incurred by Google for training Gemini Ultra stood at $191.4mn.”

Advertisement

Secondly, It is significantly costly to develop new advanced models. It is because processing user queries requires vast computational resources, that AI firms incur running costs for the model.

The report stated that “Practicality refers to feasibility, effectiveness, and usefulness in addressing real-world problems or needs. This also encompasses ease of implementation, cost-effectiveness, scalability, user access, and the ability to deliver measurable benefits. Achieving this stage is the most challenging part since many innovations that emerged over the years have been clear breakthroughs but failed to find the mass acceptance that comes to characterize a General Purpose Technology.”

Show comments