IT WAS hard to avoid artificial intelligence in 2023, with the vertiginous rise of chatbots powered by large language models (LLMs). By February, OpenAI鈥檚 ChatGPT had become . By the year鈥檚 end, it had become an everything machine: browsing the internet, interpreting pictures, generating any requested image and inserting itself into many existing tools and services 鈥� and it wasn鈥檛 the only AI to do so.
鈥淲here the technology is going next is clearly to be more multimodal,鈥� says at the University of Oxford. 鈥淚t will be to do with sound and voice and video and so on.鈥�
Sleeping giants like Google and Microsoft were jolted awake by OpenAI鈥檚 success. Microsoft, which is a large investor in OpenAI, has incorporated GPT-4, the latest LLM to power ChatGPT, into its search engine Bing and programs such as Word and Excel. Google released its Bard chatbot in February and, in December, unveiled Gemini, its most powerful LLM yet, which it claimed can exceed GPT-4鈥檚 capabilities on a range of tasks.
Advertisement
These AIs have now become part of daily life for a large chunk of the population, from students looking for help with homework to software developers. With the introduction of GPT-4 in March, ChatGPT became even more capable at generating responses. Some observers claimed it showed the first sparks of artificial general intelligence 鈥� the ability to understand and learn any intellectual task that humans are capable of 鈥� though many people disagreed.
OpenAI鈥檚 triumphant year almost came off the rails in November, when its board unexpectedly fired CEO Sam Altman. The vast majority of its staff threatened to quit en masse, and he was quickly reinstated.
AI鈥檚 renaissance in 2023 has been a seismic technological event, says Wooldridge. 鈥淓verybody wants to be the empire that鈥檚 built on the back of this technology, and what we鈥檙e seeing is a frenzy from big tech companies.鈥�
Some economists have estimated that millions of jobs could be at risk from automation, but much is still unclear. We don鈥檛 know how capable these models are, and legal questions remain around how they are trained.
In the US, there are at least seven copyright infringement lawsuits against AI companies. A group of programmers suing Microsoft for its code-generating Copilot program were the first to take action, and large corporations like Getty Images soon followed, suing Stability AI for allegedly using copyrighted images to train its Stable Diffusion image generator.
Though many people suspect that their online data has been scraped and used to train these models, it is hard to prove because the datasets and training methods have been kept secret, supposedly to
AI鈥檚 leap forward also caught the attention of national governments, who see the tech as both a threat and an opportunity. In November, the UK hosted a summit where 28 countries and the European Union signed a declaration agreeing that global action is needed to minimise the risks of AI. At the same time, US president Joe Biden signed an executive order requiring companies training AI models of a certain size to comply with government safety and national security procedures. The EU is also working on its own AI law aiming to regulate the largest models, which is expected to come into effect in 2025.
