ChatGPT is the fastest-growing app in history, with over 1m users amassed in just a few days. The ascent of artificial intelligence chat bots into the public consciousness represents a watershed moment in over six years of development in neural networks and machine learning. But progress has not always been linear. AI winters – where progress stagnates – have defined much of the young technology’s 70-year history.
The development of large language models has been instrumental in revitalising AI. LLMs are AI models trained on a diverse range of internet text, making them capable of generating human-like text based on the input they receive. They operate on a machine learning architecture known as the transformer decoder – a complex system designed for comprehending context in input data. These models learn to generate text by predicting the next word in a sentence. Through extensive training on large text datasets, they acquire knowledge about grammar, facts about the world, reasoning abilities and even pick up biases inherent in their training data.
LLMs bring significant value to various sectors, excelling in both content generation and retrieval. They can analyse large volumes of data – like PDF documents – and distil complex information into concise summaries. This capability puts detailed information at users’ fingertips within seconds. As content generators, they offer a wealth of AI-produced material, potentially increasing productivity and efficiency.
Security challenges and ethical concerns
However, these capabilities come with challenges. When retrieving content, LLMs risk overgeneralising data. In content generation, accuracy isn’t guaranteed and can be subject to a phenomenon known as ‘hallucinations’, where an AI produces information it deems correct but is erroneous. It is therefore essential to verify and cross-reference the work produced by AI.
Nevertheless, as LLMs consume more data, their performance and accuracy are expected to improve, reinforcing their role in boosting productivity. With these enhancements, however, ethical questions arise, particularly concerning privacy and the handling of data.
First, privacy concerns are a significant issue. When LLMs are trained on public data, they can inadvertently learn and generate text based on potentially sensitive information. While the models don’t have the capability to access personal data about individuals, unless explicitly provided during the conversation, there remains a concern that they could generate information that appears privacy-violating. Mitigating this risk requires advanced techniques to scrub and anonymise data, but these are not always fool proof.
Data security is another area of concern. As businesses and organisations increasingly rely on AI and LLMs to handle their data, these systems become attractive targets for malicious actors seeking to acquire valuable intellectual property. Cybersecurity measures need to evolve to protect these advanced systems from data breaches or manipulation attempts. It’s also important to consider the risk of AI models being reverse-engineered to reveal sensitive information from their training data.
Navigating these issues effectively will require multi-faceted strategies and a collaborative effort from AI developers, policy-makers and society at large. Regulatory frameworks will need to be updated or established, keeping pace with the rapid development of technology. Businesses must enforce stricter data management protocols and invest in advanced cybersecurity measures. As individuals, a heightened awareness of digital privacy and a better understanding of AI and its implications are essential.
A threat to job markets?
But even allowing for such significant challenges, it is hard to imagine a future without AI and its threats and opportunities. Its impact on job markets is already considerable and set to grow. Goldman Sachs estimates that a quarter of work tasks could be automated by AI in Europe and the US, with office and administrative support work hit the hardest. Recent announcements from IBM and BT would seem to confirm this prediction, potentially replacing roles that focus on data analysis and content creation.
Yet AI’s emergence isn’t solely a harbinger of job displacement. It also paves the way for new job opportunities and industries will most likely be built around AI expertise. Investment is flowing into the sector, especially following Nvidia’s price spike in May driven by profits attributed to AI.
In the business tech world, AI provides the ability to analyse data at an unprecedented rate, enhancing customer experiences and revealing innovative revenue streams. It’s anticipated that many proprietary algorithms, including search engines, will become obsolete as AI becomes proficient at understanding users’ needs with minimal input.
Consequently, data integrity and security will emerge as prime concerns as businesses strive to maintain their competitive edge. Samsung is reported to have banned its employees from using AI tools, such as ChatGPT, after discovering staff uploaded sensitive code to the platform. Samsung is now preparing its own internal AI tools.
Societal implications are inevitable
The rise of AI and LLMs marks a turning point in technology history. Some changes are already affecting our day-to-day lives without us realising it, through AI-enhanced communication and interactions. European regulators have banned Google’s Bard with ChatGPT now also in their sights over data privacy concerns, most likely prompting users to access AI tools through virtual private networks.
Businesses and policy-makers need to prepare for these challenges. As we step into this new era, it’s crucial to be aware, adapt and adopt if we wish to harness potential and navigate challenges responsibly. Today’s decisions will shape an AI-augmented future.
Alex Allen is Full Stack Developer and Chief Executive Officer of TeckBuff.
These themes will explored in further detail as part of OMFIF’s new series, ‘AI and central banks: redefining monetary policy and finance’. Register to attend the first event, ‘AI and implications for central banking’.