Osama Feyyad, senior vice director of AI and data strategies in the Northeast, says companies should start doing small things when introducing AI into their businesses.
He will speak at Lou Research Institute in Northeastern, Portland, Maine. Photo: Matthew Moderno/Northeastern University
With all the hype around chatbots like ChatGpt, Gemini and Claude, many business leaders feel pressured to use generated AI to quickly overhaul their operations.
That would be a mistake, says Usama Fayyad, senior vice-president of AI and data strategy at Northeastern University.
Instead, it's better for businesses to slow down, focus and start small.
“This is the opposite philosophy of what everything else in ChatGpt and Generating AI is doing,” Fayyad said last week at the first AI Action Business Summit held at the Northeastern Campus in Portland, Maine.
“They basically say, 'We want a single model, which is all models. It can solve any problem. It speaks every language. It knows every science field.'' It's not in the right direction,” he said. “I think it's actually the opposite direction from where we should be heading.”
The most popular and most promoted chatbots used today are large language models. That is, they are trained with a large amount of data. These models are not ideal for the business world, Fayyad said.
It is far more practical to look at smaller tailor-made models where companies are gaining increasing traction in the field of AI research.
“The small language model offers an incredible solution,” Feiyad added. “They are efficient, fast, private. They don't send data anywhere. And they can customize them. This is the most important part for me.”
“What is the biggest model I can afford to build? I should ask, 'What is the smallest LLM I can get away with?' ”
But even smaller, the Taylor model is as good as the data being trained. And Feiyad emphasized that this has been true since researchers began collaborating with machine learning technology in the 1940s.
In fact, many of the core technologies behind today's generative AI tools are similar to those used decades ago. The biggest difference is the significant increase in the data available in the training model, not the algorithm.
“It's a completely different world between the internet and digital transformation,” he said. “That's why it made the difference.”
Therefore, Fayyad said the companies that are most successful in integrating AI into their businesses are experts who acquire and act on data. They understand that their dataset is their secret source.
“A lot of companies, I don't care how small you are. You have the kind of data that you kill so Openai, Microsoft, or Google can access,” he said.
For many workers, generative AI can find it confusing, fast moving, and difficult to apply effectively to work. But to stay competitive, both businesses and employees must invest in AI literacy and luxury, Fayyad said.
“Is AI replacing my job?” That's a lasting question these days,” he said. “The answer is no, but people who use AI will replace your job. …It's important to use something like this.”
However, it is important to understand that these technologies can be used unauthorizedly. Fayyad pointed out a surge in deepfakes as an example.
“Today, we are creating a world where you can't trust or believe in what you read, see or hear,” he said. “In this world, where you rely on digital, how can you function when what you see is no longer reliable?”
To address these issues, Fayyad highlighted the importance of safety regulations and responsible AI frameworks being developed at universities such as Northeastern.
“I think it's a huge area to think about. How do you need to use this technology and where do you need to take it?”
