(GIST OF YOJANA) UNLOCKING THE POTENTIAL AND CHALLENGES OF GENERATIVE AI

(FEBRUARY-2024)

UNLOCKING THE POTENTIAL AND CHALLENGES OF GENERATIVE AI

Context:

ChatGPT’s release, Artificial Intelligence (AI) specifically Generative AI has caught the attention of many governments, corporations, and businesses. AI is already pervasive in our lives, and many of us use it tonnes of times a day without even thinking about it. Every time one does a web search on Google, that’s AI. Every time one goes to a website like Amazon or Netflix, it recommends products according to the history of one’s preferences.

Generative AI

Generative Al is a powerful technology with a wide range of applications across various industries. Here are some of the key areas where it’s making a significant impact:

1. Writing: Generative Al can be used as a brainstorming companion. For instance, if one is trying to name a product, one can ask it to brainstorm some names, and it will come up with some creative suggestions. LLMs can also be good at answering questions, and if given access to information specific to a company, they can help employees find information like the availability of parking at the office. They can also be useful for writing press releases. However, by providing them with details of the event, Generative Al creates a detailed and insightful press release specific to the event. In fact, some of the LLMs are even better at language translation than the dedicated machine translation engines.

2. Reading: In addition to writing, Generative Al is also good at reading tasks. For example, an online shopping e-commerce company gets a lot of different customer emails. Generative Al can read customer emails and help quickly figure out whether an email has a complaint or not. The complaints can then be routed to the appropriate department. LLMs are also being used for summarising long articles and proofreading them for grammatical errors.

3. Chatting: Lastly, Generative Al is also used for many special-purpose chatbot tasks, like government chatbots, can be used to help citizens and visitors get access to the right information on various schemes and policies.

Problems with Generative AI

Gender-Bias: One widely held concern about Al is whether it might amplify humanity’s worst impulses. LLMs are trained on text from the internet, which reflects some of humanity’s best qualities but also some of its worst, including some of our prejudices,
hatreds, and misconceptions. If one asks an LLM after its initial training to fill in the blank.
Job Losses: A second major concern is who will be able to make a living when Al can do our jobs faster and cheaper than any human can? To understand whether this is likely to happen, let’s look at radiology.
Hallucinations and Misinformation: Another concern is that it can sometimes ‘hallucinate’ inaccurate information with complete confidence. It can even invent its own references, sources, and deep fakes that are non-existent.
Plagiarised Content: LLMs sometimes output plagiarised content. If any enterprise uses that in their operations, only they are held accountable when the plagiarism is discovered, not the Generative Al model.
Transparency and User Explainability: Generative Al models give a disclaimer that the
data they have presented may be inaccurate.

Thus it seems such models obey transparency rules, but the reality is that many end users do not read the terms and conditions and do not understand how the technology works. This results in the perception that anything that LLMs say is accurate.

Key dimensions of implementing responsible AI:

Fairness of information to ensure that Al doesn’t perpetuate or amplify gender biases.
Transparency of information is vital to ensuring ethical decision-making. Users should have accessible, non-technical explanations of Generative Al, its limitations, and the risks it creates.
Privacy is another dimension for implementing responsible Al by protecting user data and ensuring confidentiality.
Safeguarding the Al systems from malicious attacks.
Ethical use of data, ensuring that Al is used only for beneficial purposes.

Conclusion:

NITI Aayog publishes discussion papers on ‘Responsible Al for All’, presenting a unique framework for implementing Al responsibly. It’s important to build a culture that encourages discussions and debates on ethical issues.
Brainstorming with a broader group of stakeholders on how things could go wrong helps identify potential problems and allows the technical team to mitigate them in advance. A checklist for brainstorming could be the five dimensions of fairness, transparency, privacy, security, and ethical use.
Generative Al has the potential to give society intelligent guidance on how to approach some of the biggest problems, like climate change and pandemics. In the coming times, Al will contribute to longer, healthier, and more fulfilling lives worldwide if used responsibly.