Meta, Snapchat and more turn to AI chatbots. But what are they really?
From social media to search engines, AI chatbot integration has arrived.
Recently users of Meta’s suite of apps, including Facebook, Instagram, Messenger and WhatsApp, have likely noticed the presence of an artificially intelligent chatbot designed to engage users in useful conversation.
A self-proclaimed “friendly AI assistant with a passion for helping others,” Meta AI is among dozens of chatbots that have surfaced since OpenAI’s ChatGPT launched in late 2022 and disrupted the tech landscape.
Others include Anthropic’s Claude, Microsoft’s Copilot and Bing AI, Amazon’s Q, xAI’s Grok and Google’s Gemini.
“Meta AI has the potential to revolutionize the way we interact with technology and access information,” the chatbot claims with an alleged quote from Meta CEO Mark Zuckerberg that comes with no attribution or corroborating sources. Due to the novelty of the technology, accuracy in responses may vary, according to Meta.
Meta AI invites users to ask it anything. Unless users mute it by diving into the settings, Meta AI has no trouble generating a response.
Meta AI and other modern AI chatbots are at their core interfaces to large language models, systems trained by massive datasets of text covering a wide range of topics that recognize word usage patterns and predict text based on context, said Vivek Singh, the director of the Behavioral Informatics Lab at Rutgers University.
What is a chatbot?
Chatbots, or digital interfaces that talk to people, have been around for years, but things have changed with the integration of large language models that have allowed them to develop a knowledge of intricate word relationships. Chatbots appear conversational. They excel in creating natural-sounding text and can easily refine existing text or create quiz questions based on excerpts.
Some, such as Microsoft Copilot, aim for versatility and can offer up-to-date information on current events that link back to sources. It can also write, summarize, translate and converse. Others are more nuanced. New Jersey-based ZeroBot’s AI is a voice-enabled chatbot linked to Grok that offers custom avatars, called “agents,” to deliver information. Others still were tailored for research, coding or web searches.
They all, however, lack human-like comprehension of language, said Jonathan Howell, a professor in the linguistics department at Montclair State University.
Instead, they continuously refine their understanding of language to enhance their ability to accurately predict the proper response, according to experts. They are good at repeating what we said and can guess what a reply might be, Howell said. When prompted, they perform what is essentially an “autocomplete” function on steroids, Singh added.
“It gives you the illusion as if it knows things,” Singh said. “It’s actually trying to make its best guess out of what it has seen in the past.”
How does AI work?
The result is systems that can help with repetitive and boring text-based tasks, such as filling out forms, Singh said. Since chatbots draw from a ton of data and are not constrained by reality, they are also good for brainstorming, according to experts. However, they are not exactly ideal for producing external-facing documents, Howell said. Give chatbot responses a double check, if not a sanity check, Singh added.
More:Friday’s earthquake was the strongest in NJ since 1783. A look back at quake history
“You always want to be critical,” Howell said. “When it’s really important, you don’t want to end your search for information with the chatbot.”
Flaws in artificial intelligence
Modern AI chatbots struggle with numerical tasks, forward planning and discerning factual accuracy, particularly in current events, according to experts. They can consequently produce inaccurate or fabricated content, making up an answer rather than conceding a lack of knowledge in what experts call “hallucinations,” Singh said.
Chatbots come with several other key flaws. They can replicate or amplify problems included in their data sets, display inequity across languages and misuse copyrighted content, according to Singh, Howell and other experts.
So far, the flaws in chatbots have left many in New Jersey skeptical. A recent poll conducted by Stockton University found most of the state’s residents (79%) express low levels of trust regarding the ethical and responsible use of AI. A similar percentage, 78%, said they believe that AI will exacerbate the spread of false information
While Microsoft’s offerings make use of OpenAI’s GPT-4 large language model, many others use a proprietary system. Meta AI, for example, is called LLaMA 3, which gathers information from a wide variety of sources. Most tap into publicly available internet resources. Google recently inked a $60 million deal to use Reddit posts to help inform its Gemini system.
AI and copyrighted content
Along those lines, the Associated Press has licensed part of its archives to OpenAI. However, often information is scraped from sites without consent from content creators.
A slew of U.S. newspaper publishers have claimed in recent court filings that tech giants Microsoft and OpenAI are reusing their articles without permission and attributing inaccurate information to them through AI-generated content.
The lawsuit follows The New York Times’ earlier legal action against OpenAI for allegedly using its archives without authorization.
More:North Jersey home built by founder of Time and Life magazines listed for $5.8 million
While what is posted on the internet may help AI chatbots improve their functionality, officials at the Mozilla Foundation warn against giving chatbots too much information. The nonprofit behind the Firefox browser released a list of tips to safeguard against chatbots with privacy policies that allow them to collect and disseminate user data.
Chatbot users are urged not to give sensitive information or access to photos, videos, cameras or microphones. Users should refuse ad tracking, choose strong passwords and request the deletion of all data once they stop using an app.
AI Chatbots in their own words:
ChatGPT: OpenAI – GPT-3.5 (GPT-4 available)
“I’m an AI language model created by OpenAI called GPT-3.5, designed to understand and generate human-like text based on the input I receive. My purpose is to assist users like you by providing information, answering questions, generating content, and engaging in conversation on a wide range of topics. I’ve been trained on a diverse dataset covering various domains, so I can discuss everything from science and technology to literature and philosophy. If you have any specific questions or topics you’d like to explore, feel free to ask!”
Claude: Anthropic – Claude 3
“I am an AI assistant created by Anthropic. I have a broad knowledge base that was current as of August 2023, which allows me to converse on a wide range of topics and assist with things like writing, analysis, math, coding, and more. However, my knowledge does not extend past August 2023, so for events after that date I can only speculate based on the information I have. I’m an artificial intelligence without a physical form, but I strive to be helpful, honest, and ethical in my interactions.”
Copilot: Microsoft – Microsoft Prometheus, GPT-4
“I am an AI language model developed by OpenAI, and my primary purpose is to assist users like you by providing information, answering questions, and engaging in conversation. You can think of me as a friendly digital companion here to help with various tasks and discussions. While I don’t have personal experiences or emotions, I rely on my training data and algorithms to generate responses. If there’s anything specific you’d like to know or discuss, feel free to ask, and I’ll do my best to assist you!”
Gemini: Google – Gemini (language model)
“I am Gemini, a large language model created by Google AI. I am similar to Google Assistant, but I am still under development and focused on providing informative responses. I can access and process information from the real world through Google Search and keep my response consistent with search results. Unlike Assistant, I cannot directly control smart home devices or perform actions in the real world. However, I can answer your questions in a comprehensive and informative way, even if they are open ended, challenging, or strange.”
Meta: AI Meta – LLaMA 3
“I am Meta AI, a friendly AI assistant. I’m here to help you with any questions or tasks you have. I can:-Answer questions on a wide range of topics, from science and history to entertainment and culture-Generate text on a given topic or subject-Translate text from one language to another-Summarize long pieces of text into shorter, digestible versions-Offer suggestions and ideas for brainstorming sessions-Chat and converse on a variety of topicsI’m constantly learning and improving, so please bear with me if I make any mistakes. I’m here to assist you, and I’m happy to help in any way I can!”
