Artificial intelligence has been the buzzword on everyone’s lips over the past few years, but what will the next 12 months of development bring?

In a constantly evolving field, there is much to look forward to, including the integration of AI into your doctor's clinic, more work towards AI diagnostic tools across healthcare, and broader development in the small tech sector rather than just big companies.

At the same time, some AI-related issues have begun to surface, such as exaggerated flaws like bias and misinformation, a severe lack of oversight or regulation, and increasing debate around ethics, job security, and the real environmental harms of AI.

Are we facing a year of AI discovery or a clearing of AI accounts?

Here’s what experts think the world of AI development might have in store in 2024...

Not just the big tech world

Main AI headlines over the past few years have largely centered on a handful of big names, but that is something that has started to change, according to researcher Sasha Luchioni, who specializes in societal and environmental impacts of AI.

She told CTVNews.ca in a December phone interview: "I definitely think there will be more democracy or at least distributed deployment of AI."

"Right now, we've seen a kind of concentration of power with regard to big tech and OpenAI doing most of the deployments, especially in generative AI models, but I really see that shift changing with more and more startups and smaller companies overall catching up and doing their cool things with AI."

OpenAI is the team behind the huge brand of ChatGPT, one of the most famous generative AI tools.

One issue that has kept AI technology in the realm of tech giants is the massive computing power necessary to train and run AI models. This is easy for established companies like Google or Microsoft, but for the average person, it is virtually impossible, according to Luchioni.

She said, "With generative models, or large language models, they are really huge," adding that large language models require "from a thousand up to several thousand GPUs, which are specialized devices. And that rapidly adds up if you have to rent it on the cloud or build your own cluster. In most data models, you can no longer train on one computer; you need a huge amount of infrastructure."

"But actually, if you share models, you can build progressively, and you don't have to start from scratch."

Luchioni said she sees the impact of this collaboration through her work at HuggingFace, an open-source machine learning platform for both AI experts and enthusiasts to share and develop models.

A major move toward more collaboration occurred earlier this year when Meta publicly released Llama (Large Language Model Meta AI) to help workers in the AI research community. Meta was followed in the summer by the public launch of Llama 2.

"You don’t have to say, 'Okay, I'm just going to train this massive model for a million GPU hours,' you can take an existing model like Llama, or any open-source models, and fine-tune it — basically adjust it to fit your use case or your data set. That way, you can leverage that huge model without having to spend a million dollars on computing."

If a bank, for example, wanted a customer service chatbot, those coders could use an existing general language model and input data related to common customer inquiries or interactions with bank staff and tailor the model to be specific to the bank's needs, Luchioni explained.

"So we see a lot, especially startups that can’t access that huge amount of computing power, turning to more adapted training."

She expects we will see more sharing of tools and information in the AI research community in 2024, which could stimulate broader applications of this technology.

The growing field of medical AI

Medicine is one of the most exciting fields for practical AI applications, according to Oishi Banerjee, a PhD student in computer science at Harvard University.

She works in the Rajpurkar lab, where they focus on developing medical AI, and said AI in medical imaging is poised to take off soon.

Banerjee said, "I personally hope that 2024 in medicine will be the year we start seeing really specialized imaging models in medicine." "For example, you could just do a CT scan for an (AI image model), and it will automatically say, 'Oh, this is the liver, and this is the tumor.' 'This is the pancreas.' That would be amazing."

Banerjee added that while there are existing models now for very specific healthcare tasks, such as isolating kidneys in CT scans, expanding AI’s image generation capabilities hasn't yet reached healthcare — but it’s on its way.

She said, "We’ve already seen amazing progress on these models becoming more diverse and adaptable in the natural imaging field, and there is a lot of active work to bring that capability and diversity into medicine."

"Diagnostic tools are likely to become more diverse and better during 2024, just because of how diverse and powerful the underlying AI technologies are."

Many other potential applications are also on the horizon, ranging from chatbots that deal with patients to AI models that scan the latest medical literature to language models that help radiologists in reporting.

An AI model trained on a database of patients’ medical histories and their reactions to certain treatments might be able to assemble how some patients respond to a specific therapy.

If the model can capture those patterns, sometimes not yet discovered by humans, you could use this model to see a brand new patient, look at their scans, look at their history, and say, "Oh, I think the best possible drug for this person is drug A and not drug B," noting this is not her area of expertise.

In the future, an AI model trained to understand the full range of chemical properties could also suggest new molecules to test as part of drug discovery based on specific pathological presentations, Banerjee said.

“That’s exciting,” she said, but those advanced applications are still years away.

Banerjee is confident we will see a significant increase in AI on the administrative side of healthcare in 2024.

She said, "That seems less glamorous or exciting, but I can’t overstate how much time doctors spend on administrative paperwork."

"I would say in healthcare, 2024 might be the year AI gives your doctor more time to spend with you."

This expectation was echoed by Timothy Chan, a professor in Health Narrative and Analytics research at the University of Toronto and Canada.

He told CTVNews.ca via email: “Less exciting applications, such as improving operations or streamlining office or routine tasks, are likely to have a big impact without the average patient noticing,” adding that AI “will become more and more important in care planning and delivery.”

However, things are not smooth in medical AI.

A recent U.S. study found that doctors, when given a good AI model, saw a slight increase in their diagnostic accuracy. However, when researchers gave doctors a biased AI model, doctors’ treatment recommendations accuracy dropped by 11 percentage points. The study’s authors said this means doctors were not able to recognize and adapt to the AI model's bias.

Banerjee said weak AI models that replicate bias are an issue experts are working to combat.

“I see a push within the medical AI community that says when you evaluate a new medical AI tool, try to get a diverse population, and don’t just test it in one hospital in one wealthy city, but try to get many different groups of patients, do subgroup analyses to make sure you don't mess up one historically marginalized group,” she said.

"Many people are fully aware of this problem in medical AI and take it seriously.