Artificial intelligence (AI) should not be solely relied upon to determine what content people should read, according to Arjun Narayan, head of trust, safety, and customer experience at SmartNews. Narayan emphasized the need for human oversight and guardrails to ensure that the right content is pushed to users. SmartNews is a news aggregator platform that curates articles from 3,000 news sources worldwide. Its mission is to provide reliable and relevant news to its users. The platform uses machine learning and natural language processing technologies to identify and prioritize news based on users’ personal preferences. It also evaluates the trustworthiness and accuracy of news sources.
Narayan highlighted the importance of human oversight in the use of AI, as it still lacks key traits needed to decipher information. He mentioned that large language models, while efficient in analyzing evergreen content, struggle with processing breaking or topical news. These AI models are effective in summarizing large chunks of content and supporting functions such as content distribution. Narayan’s team is currently evaluating the effectiveness of using large language models to determine if content meets the company’s editorial policies. However, he stressed that AI models are not “conscious” and lack contextual comprehension, and these issues can be resolved over time with more data and training. It is crucial to invest equal effort in ensuring training data is unbiased and free of inconsistencies. Narayan expressed concerns about the lack of transparency in the industry regarding the data used to train AI models and called for more disclosure. He also highlighted the risk of AI generating false information so realistic that people believe it to be true, emphasizing the need for human oversight and governance. Organizations should audit the output of their AI models and implement necessary guardrails to prevent issues like providing instructions on building a bomb or plagiarizing articles. Narayan concluded that AI is not at a stage where it can run on its own, and investment in human capabilities and oversight is necessary to ensure content is proofread and fact-checked.
While AI has limitations in handling breaking news and real-time data, it has proven effective in powering SmartNews’ recommendation engine. The platform uses AI to prioritize articles based on users’ reading patterns. Narayan hinted at the potential of generative AI in facilitating human interaction with machines, allowing anyone to obtain the information they need by asking the right questions. However, there are ongoing discussions with publishers about how articles written entirely by AI or augmented with AI should be managed and enforced. The level of disclosure for these variations is also a topic of debate.
