Five Things to Understand About AI in 2024
By Neil Brady
8th February 2024
This article first appeared in the Hilary edition of Trinity Today.
As the world enters 2024, technology analysts everywhere are trying to anticipate the likely next instalment in the generative artificial intelligence story.
In January, for example, Accenture CEO Julie Sweet mused that ‘most companies do not have mature data capabilities and if you can’t use your data, you can’t use AI.’
While there’s much truth to this, in the wake of the foreseeable decision by The New York Times to file suit against OpenAI and Microsoft, claiming ‘billions of dollars in statutory and actual damages’ for the unlawful use of its data to train large language models like ChatGPT, a more prescient choice of phrasing might have been ‘others’ data’.
This question of data quality, ownership, legality and provenance, will be the defining issue of artificial intelligence in 2024 and beyond, perhaps second only to compute cost.
At CaliberAI, we have been working with Large Language Models (LLMs) for almost five years, and as a team, we have a collective experience in the implementation of multifarious disruptive technologies, from the internet to social media. My co-founder and father, Conor, oversaw the creation of The Irish Times’ original web domain (www. ireland.com), while CaliberAI’s Chief Technology Officer, Paul Watson, oversaw the creation of Storyful, one of the world’s first social media newswires and later acquired by Newscorp.
Thanks in no small part to the help of an Advisory Panel of eminent minds from the worlds of journalism, law, and philosophy, we have built our technology in anticipation of the kinds of problems now facing OpenAI and others in relation to copyright, and other data integrity imperatives.
For example, while right now most policy and technology observers are rightly focused on The AI Act and its implications, several provisions of The General Data Protection Regulation already regulate AI to a great extent, including Recital 71 and Articles 13, 14, 15 and 22, all of which mandate the kind of algorithmic explainability we have baked into our products.
So, here are five things to understand about artificial intelligence going into 2024.
1. It’s all about the data
Whether viewed from the perspective of a data scientist, machine learning engineer, copyright lawyer or writer, it is clear that data integrity and quality is what will differentiate and reduce risk for AI services and products going forward. Relatedly, in the wake of The AI Act’s tiered approach to risk management, it is also clear that the use of safeguards and guardrails, to mitigate reasonably foreseeable risk, will be crucial.
2. The European Union (EU) will remain the world’s preeminent data regulator, but regulation is splintering
The EU has led the way on data protection since 2016, and it will continue to do so through The AI Act. This was affirmed by OpenAI’s recent decision to shift responsibility for data control to OpenAI Ireland Limited. At the same time, other companies, such as Facebook, are increasingly suggesting that it won’t be possible to offer their products and services within the EU, supposedly due to the high regulatory bar. Meanwhile, as the EU agrees to place an outright ban on things like biometric categorisation systems, other jurisdictions, such as China, have done the opposite.
Splintered regulation is unavoidable, but the EU is as close as the world can get right now to supranational enforcement of standards.
3. Journalism won’t make the same mistake twice
In the early 2000s, as search engines began to index the internet, they came into conflict with content creators, most notably news publishers, over the legality of this, and whether it constituted fair use at law. And as the public’s eyeballs shifted from news publishers to search engines as a result, so did the advertising. The legalities and the revenue collapse that followed has never really been fully resolved, with battles to make technology companies pay for this material (the ‘snippet tax’) still ongoing.
Now, as demonstrated by The New York Times’ decision to sue OpenAI and Microsoft, generative AI threatens to destroy what commercial viability remains of the ad-traffic model, because it runs the risk of complete disintermediation of news delivery. As The Economist put it recently, ‘for years the complaints of publishers against platforms have rung somewhat hollow. Now they have a real story on their hands.’ News publishers are unlikely to back down.
4. Generative AI hype is peaking... there are many kinds of AI out there
Much of the world has been entranced by the capabilities of generative AI over the last year, but in 2024 there will be a realisation of the importance and capabilities of other kinds of AI, such as classificatory. Sure, most people know how effective AI can be in classifying imagery or faces. Fewer people understand how effective it can be in classifying dangerous language of the kind regulators now have in their crosshairs.
5. Yes, you *do* need to upskill!
Generative AI hype is peaking, but the technology and artificial intelligence more generally is here to stay.
In order to understand its likely effect on employment prospects, look to the field of radiology. For years it has been the go-to example of a sector where AI will replace entire roles, and yet the numbers of radiologists and employment have remained remarkably stable over the past decade. Why? Although AI outperforms humans in narrow ways like visual diagnosis, it doesn’t hold a candle in others. As one of the authors of a study published in the oncology journal, The Lancet, noted in August 2023, ‘the greatest potential of AI right now is that it could allow radiologists to be less burdened by the excessive amount of reading.’
This difference, between an employee using AI to do their job, to manage workload more effectively, and the productivity gains that result, and an employee who does not, will be the difference between employment and unemployment.