By: Oluwaseye Ogunsanya
Artificial Intelligence (AI) is revolutionising multiple facets of society, driving innovation, and reshaping industries. From healthcare to finance, AI’s ability to analyse vast amounts of data, make predictive decisions, model scenarios and generate innovative solutions to complex problems is enhancing efficiency and accuracy.
For instance in healthcare, AI aids in diagnosing diseases and personalising treatment plans, while in finance, it improves fraud detection and automates trading. In our everyday life, AI’s influences are becoming indispensable.
One of the subsets of artificial intelligence is Generative AI. It focuses on creating new content from learned data patterns. Unlike traditional AI, which processes existing data to make decisions, generative AI can produce original text, images, music, and more. This technology uses advanced neural networks like Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) to enable machines to create, innovate, and imagine, moving us into a realm where creativity knows no bounds. Generative AI also stands as a monumental leap in technology and it is altering the very fabric of our world. With its transformative potential, it is rewriting the rules of what’s possible.
Illustrating how it works in simple terms, Intelligent Block describes it like a computer artist that can make new things. It is not just following instructions, but it can create its own original stuff, like pictures, music, or even stories.
“This AI uses what it has learned from lots of examples to come up with something completely new and creative. It’s like a virtual imagination that can help us make cool new things we’ve never seen before.” It added.
As such, Generative AI is being used to create chatbots that can hold natural conversations with humans, it can be used to create realistic images from text descriptions, it can be used to create new music from scratch, and it can used to create realistic text, such as news articles, blog posts, and even novels.
AI Models and Applications
ChatGPT
OpenAI stunned the world in November of 2022 when the company fully launched ChatGPT. Generative pre-trained transformers, or GPTs, had been around for a while before OpenAI released ChatGPT.
The release of ChatGPT according to The Future of Being Human made the technology exceptionally accessible to anyone with a web browser and an internet connection. And it demonstrated — with meme-like virality — just how sophisticated and human-like natural language interfaces were becoming. It also added that the release of GPT4 took things to another level.
“The public success of ChatGPT paved the way for other models and companies to enter the fray at a speed and scale that was mind boggling. Meta’s Llama, Google’s BARD and Anthropic’s Claude, are just three examples from an exploding scene around large language models that further boosted an increasing tumultuous landscape.” it stated.
Stable Diffusion
In 2022, Stable Diffusion became widely available. It is essentially the “ChatGPT” of the image generation world. With just a few prompts, Stable Diffusion proved able to generate brand-new, contextualised, highly detailed images. It formed the second prong of the 2022 AI revolution, and like ChatGPT, it has enjoyed many upgrades since.
StyleGAN
StyleGAN predates both ChatGPT and Stable Diffusion. First released in 2018, it surprised people by producing photo-realistic faces. While less generalised than other image generators, several iterative upgrades made StyleGAN one of the most powerful and convincing face generators in the business.
The Use of Generative AI in Journalism and Fact-Checking
As earlier established, Generative AI is making its mark across various industries, pushing boundaries and fostering innovation in unprecedented ways. It is not a different case for journalism and fact-checking as it is proving to be an invaluable tool in the field due to its ability to generate coherent and engaging articles as well its huge role in the areas of news production, distribution and consumption.
Generative AI systems can also generate text tailored to the specific style and tone of news organisations, facilitating tasks such as transcription, translation, and daily news updates thereby saving time and effort for content creators.
At a time when the world is grappling with the challenge of mis/disinformation across all sectors, fact-checking organisations around the world are responding to the challenge positively as they are regularly infusing AI into various aspects of their work and also looking for ways to automate fact-checking.
Fact-checking organisations around the world are responding to this challenge positively as they are regularly infusing AI into various aspects of their work and also looking for ways to automate fact-checking.
The quest for this started in 2013 when the founder of the American fact-checking organisation Politifact, Bill Adair, first experimented with an instant verification tool called Squash at Duke University Reporters’ Lab in 2013. Squash is a system under development that fact checks video of politicians as they speak. The goal is to display related fact checks on viewers’ screens in seconds.
Squash listens to what politicians say and transcribes their words, making them searchable text. It then compares that text to previously published fact checks to look for matches but its utility was limited. It did not have access to a big enough library of fact-checked pieces to cross-reference claims against, and its transcriptions were full of errors that humans needed to double-check.
“Squash was an excellent first step that showed us the promise and challenges of live fact-checking,” Adair told WIRED. “Now, we need to marry what we’ve done with new advances in AI and develop the next generation.”
Another instance is Newtral’s multilingual AI language model, ClaimHunter, which was developed in 2020, and funded by the profits from its TV wing, which produces a shown fact-checking politicians, and documentaries for HBO and Netflix.
Using Microsoft’s BERT language model, ClaimHunter’s developers used 10,000 statements to train the system to recognize sentences that appear to include declarations of fact, such as data, numbers, or comparisons.
ClaimHunter automatically detects political claims made on Twitter, while another application transcribes video and audio coverage of politicians into text. Both identify and highlight statements that contain a claim relevant to public life that can be proved or disproved—as in, statements that aren’t ambiguous, questions, or opinions—and flag them to Newtral’s fact-checkers for review.
According to Newtral’s chief technology officer, Rubén Míguez, the system is not perfect and occasionally flags opinions as facts, but its mistakes help users to continually retrain the algorithm. It has cut the time it takes to identify statements worth checking by 70 to 80 percent.
Newtral is also working with the London School of Economics and the broadcaster ABC Australia to develop a claim “matching” tool that identifies repeated false statements made by politicians, saving fact checkers time by recycling existing clarifications and articles debunking the claims.
Similarly, Full Fact, a media company founded in 2009 is offering several fact-checking tools, including ones that are automated through the use of artificial intelligence having won the Google.org AI Impact Challenge in May 2019 alongside Africa Check, Chequeado and the Open Data Institute.
With the support of Google, the organisation is using machine learning to improve and scale fact-checking by working with international experts to define how artificial intelligence could transform the work, develop new tools and deploy and evaluate them.
It is also building AI tools to help fact-checkers understand what is the most important, and check-worthy, information of the day. It also aims to design an algorithm that can identify when somebody knowingly repeats something they know to be false.
It is also worthy of note to mention that in the run-up to the 2023 General elections in Nigeria, Full Fact offered its artificial intelligence suite — consisting of three tools that work in unison to automate lengthy fact-checking processes — to greatly expand fact-checking capacity in Nigeria.
The three tools from Full Fact — search, alerts and live functions — work in real-time to detect claims, alert fact checkers when false claims are repeated, and instantly transcribe television or radio interviews (cross-referencing things said with existing fact checks).
Introducing FactCheckAfrica’s MyAIFactchecker Tool
Equally, FactCheckAfrica is also developing innovative tools to further enhance the work of fact-checkers and aid the general public in inculcating the habit of fact-checking. One of such tools is our award-winning AI AI-powered chatbot known as MyAIFactchecker.
MyAIFactchecker harnesses a synergy of artificial intelligence and reputable news sources. Combining Google’s fact-checking API with the GPT-4 model, it also has a nice user experience interface which incorporates French, Swahili and Nigerian local languages to break the language barrier as well as a voice option for fact-checking.
Additionally, it uses the Llama model via Groq’s services. Groq provides the infrastructure to efficiently run LLaMA 3, which helps the tool to analyse and verify claims quickly. This setup improves the accuracy of our fact-checking and enhances the overall speed of delivering reliable information to our users.
According to deepsense.ai Llama stands for Large Language Model Meta AI, which is an autoregressive language model that relies on a transformer architecture (similar to many of the recently developed alternatives). While the first iteration of Llama (presented in late February 2023) was generously made available for non-commercial use, the second version, Llama 2, takes a leap forward, by not only being open to the public but also offering itself for commercial usage.
It added that the Llama 2 licence permits any commercial use of the model with one small exception – if you had a user count of over 700 million per month at the time of the model’s launch, obligatory permission must be sought from Meta. This licence exception was implemented due to Meta AI’s desire to prevent their current competitors from utilising the model. Anyone else can make unlimited use of it, and even if applications based on it reach that kind of scale in the future, it will still be license-compliant.
Llama 3, the latest version of Meta’s large language model, has been introduced in two models, boasting 8 billion and 70 billion parameters, designed to redefine processing power, versatility and accessibility. Unlike its predecessors, Llama 3 is open source. According to builtin.com, it handles a more extensive array of tasks, including text, image and video processing. It was trained on more than 15 trillion tokens, a dataset seven times larger than that used for Llama 2, allowing for more nuanced understanding and generation of content. Here are some of its key features and capabilities.
BBYDI’s AI Engineer Speaks On the Use of Llama 3 For MyAIFactchecker
In an exclusive interview, Alamin Musa Magaga, a data scientist and AI engineer who also heads the technology department of Brain Builders Youth Development Initiative, discussed the rationale behind using Llama 3 models for myaifactchecker, the integration process, challenges faced as well as future development.
According to him the team initially used the OpenAI GPT model for the fact-checking platform’s generative AI. However, they switched to Llama models due to its reliability and cost-effectiveness.
Magaga explained that the advanced search APIs fetch claims and information online, which the Llama model uses to generate accurate fact-checking responses.
He said that the integration involved using advanced search APIs like Calgary API and site API combined with Groq framework which allowed the team to use the Llama model without downloading the large and computationally expensive models, this according to him made the process more efficient and cost-effective.
“The Llama models provide nuanced and precise responses, making them crucial for the platform’s success. The model delivered accurate fact-checking output” he added.
Speaking on the challenges faced, he said: “One significant challenge was the large size and computational expense of downloading the Llama model. The team overcame this by using the Grok framework, which provides fast AI inference without the need to download the entire model, thus reducing costs and resource requirements.”
He further highlighted that there is a development of a mobile app and an offline messaging fact-checking platform and the team is continuously updating the platform with the latest technology. They plan to also enhance its capabilities further, ensuring it remains at the forefront of AI-driven fact-checking.
According to him, the tool ensures accuracy through a robust fact-checking process and has garnered over 15,000 users. Also feedback from the users, especially journalists, has been overwhelmingly positive with no issues related to misinformation. These efforts have helped it gain wide visibility.
Speaking on the ethical considerations and data privacy, Magaga said “The platform prioritises users’ data privacy and security ensuring that all fact-checked information remains anonymous.” He added that the team employs measures to prevent misuse and addresses potential biases by incorporating multiple advanced search engines and AI models.
He went on to advise developers to engage in continuous learning and project-based learning to gain experience emphasising the importance of integrating AI into various activities, not just fact-checking, to leverage its full potential.
AI is fundamentally changing the world, with generative AI leading the charge in creativity and innovation. The concerted efforts by tech companies to democratise this technology are paving the way for a future where AI-driven creativity and efficiency are within everyone’s reach.
Edited by Habeeb Adisa
