Reimagine Everything

Toward the end of 2023, CME and Siren held an end-of-year event to reflect on the work done by the two companies and discuss cross-cutting themes. The event included a panel on AI, moderated by Siren’s Head of Technology and Operations, Dany Mezher, addressing many questions that have been on everyone’s minds. It began with the simple question, “What exactly does AI do?”

For Kamal Youssef, a Senior Solution Architect at CME, “AI is not magic. It’s basically statistics on steroids. You give it a lot of data, and it starts to estimate stuff based on that data. It learns how to detect some patterns; it makes some predictions. It collects insights and helps us make better decisions.” What is making it stand out now is that more computational resources are available, enabling AI to process and interpret more data and generate increasingly accurate results, he said.

AI’s not-so-recent beginnings

AI is not new – the term ‘artificial intelligence’ was coined at a conference in the 1950s, where the early pioneers envisioned creating machines that could mimic human intelligence and perform tasks. It wasn’t until recently that image and speech recognition improved, as well as natural language processing. This takes us to the present day, where AI is at its current peak of sophistication with recommendation systems, virtual assistants, and AI integrated into many applications.

When asked why AI is so popular, Zahi Lahham, a Senior Software Engineer at CME, said it is partly due to everyone wanting to have a hand in the game. “We all have smart devices now where we store data. There comes a point where a company says, ‘we have all this data – what can we do with it?’”

He explained that AI can help companies leverage that data to keep ahead of competition, so more and more companies are getting involved. For him, three factors have lowered the barriers to entry: the abundance of data, better collaboration tools and cheaper computational resources.

Shortcomings to address

AI can be particularly useful in business, but first, it must be correctly understood, including its security and implications. Kamal continued, “We don’t have AGI (Artificial Generative Intelligence) yet, which has the capacity of self-learning and problem solving across domains, without the interference of expert knowledge and analysis. Understanding capabilities and limitations enables us to choose the best methodology for the given problem and account for its shortcomings using typical software engineering approaches.”

In the meantime, AI needs to be reliable and best practices should be created. There must be a very tight cross-functional connection between the AI and software engineering teams for proper business production, improvement and assessment, leading to successful and manageable AI enabled solutions.

As far as the future is concerned, Amer Mouawad, AUB Assistant Professor of Computer Science and Siren AI Consultant, said it all changed when AI became generative.

“The problem is that the answer is often different when you submit a query. The answer you receive depends on the model’s weights and is unpredictable. The way Large Language Models (LLMs) are used is also going to change. Why? Because currently these models cannot verify any of the answers they give, and this is a big issue.” He explained that when you ask a question and request a link from some LLMs, they will give you one, despite the link being completely fabricated.

Amer thinks that “over the next few years, LLMs will no longer be used as black boxes but as components within RAG (Retrieval Augmented Generation) systems and therefore, will shrink in size so that almost anyone can run them on their own machines, providing better accessibility.”

Ethical concerns persist

Ethically speaking, there are still a lot of issues that need to be addressed as technology becomes more integrated into various aspects of our daily lives. There is a lack of transparency and dilemmas with privacy, security, and societal impact. What is the best way to handle these challenges?

Kamal concluded, “the first step is using data that is validated, unbiased and clean to train your models. If you want to control your AI, you need to feed it the proper data.” He added that worldwide regulations also need to be uniformly established.

Initiatives and regulations are gaining momentum, and several countries are working on a framework to address legal and ethical concerns associated with AI technologies. Policy makers are grappling with the challenge of striking a balance between fostering innovation and ensuring responsible use. But in the current climate, two things are undoubtedly sure – there will be an ongoing focus on creating adaptive regulations that will keep up with the rapidly evolving nature of AI, and it will continue to drive massive innovation across multiple industries, pushing the envelope of what’s technologically possible.