Responsible and Trustworthy AI

Introduction

Recently, AI technologies have seen explosive growth. With more genAI applications being built it is important to understand the impact and consequences of AI. I saw an interesting quote recently: "We wanted AI to do our mundane tasks so that we could focus on creativity and not that AI to do creative tasks so that we can do our mundane tasks”.

There is also a risk of using AI as GenAI is exciting and will open up many opportunities to add to the global economy. But it also opens up a lot of risks and dangers to corporations and individuals alike. Responsible GenAI is not only the government's responsibility, but we all are collectively responsible for generating more benefits with genAI. I built a genAI-powered application to provide educators with a platform to develop personalized lesson plans. That experience gave me a first-hand perspective of the power of genAI and the responsibilities that come to us as technology leaders to ensure we use AI safely and that it is trustworthy.

Problems with AI.

As AI’s power increases, it also opens up several risks and problems. This is in no way a comprehensive list but I am just scratching the surface.

• Inappropriate use: A technology can be used beneficially or maliciously. We have seen a lot of issues come up with deep fakes and misinformation that are known. There are several which are unknown.
• AI Race: We have seen the negative consequences of the arms race and if governments get into an AI race who’s to say where it will end? The impact of the arms race was visible and I am afraid that with AI technology, we won’t even see the impact as it can be highly targeted and obfuscated.
• Rogue AIs: Humans have been trying to build an autonomously intelligent robot for several decades. While the objective is to benefit the human race and planet Earth, the risk comes from if and when the AI becomes truly autonomous and humans lose control over them.
• Misalignment: For now, the AI is a robot that does what it is told. If the instructions provided are not accurate and aligned with the desired objective, the results could be unexpected to catastrophic.
• Hallucinations: GenAI LLMs are compared to parrots. Sometimes they listen / read other people’s work and parrot it back without any validation even if it is incorrect. There was a very famous and hilarious incident where the chat bot declared that Australia didn’t exist.
• Conformism: Even today, the AI LLMs conform to human intelligence. I asked the bot what 20+20 was and it correctly responded that it is 40 and confirmed it when I asked if it was sure about it. However, when I said that I think 20+20 was 42, it corrected itself and said that it made a mistake. If the bots can do this with absolute truths, then I am worried about the veracity of the results in subjective situations.

Conclusion:

AI and Machine Learning (ML) has made a lot of progress in the past few years, but it still has a long way to go. Especially the LLMs. There are other key and more impactful applications of AI, but the risks of using AI at this time can be extremely dangerous for humans. AI and ML can help in accelerating and eliminating mundane tasks, but it requires strong human oversight to ensure the results are not only accurate and actually helpful for people but also solving the intended problems and not creating a cobra effect.