The following essay is reprinted with permission fromThe ConversationThe Conversation, an online publication covering the latest research.
In 2023, AI's pivotal moment unfolded, thrusting generative AI into the spotlight. A headline-grabbing AI startup's boardroom saga captivated the public, while the Biden administration and the European Union responded by attempting to regulate the already galloping AI advancements through executive orders and laws.
Our panel of AI scholars anticipates 2024, highlighting challenges for developers, regulators, and the general public. They share hopes and recommendations for navigating the evolving landscape.
Casey Fiesler, Associate Professor of Information Science, University of Colorado Boulder
In 2023, AI hype dominated discussions, with divergent narratives predicting it as a savior or a threat. Amidst these visions, the line between futuristic aspirations and present realities blurred. While acknowledging the importance of anticipating future ethical concerns, it's crucial not to succumb to the enchantment of AI as a magical force. To regain control, a deeper understanding of the technology is imperative, as it remains a malleable entity shaped by deliberate choices.
In the AI discourse of 2023, a significant debate unfolded regarding the involvement of ChatGPT and analogous chatbots in educational settings. Initially, headlines primarily highlighted concerns about students exploiting these tools for academic dishonesty, leading to educators implementing restrictive measures with unintended consequences.
As the year progressed, a shift occurred, acknowledging that neglecting AI education placed students at a disadvantage. Many schools reconsidered their bans, realizing that failing to teach about AI hindered a comprehensive understanding. While advocating against overhauling education to prioritize AI, it became evident that imparting knowledge about AI's functioning was crucial. Without such education, individuals, including students, would lack an awareness of AI's limitations, affecting their discernment of its appropriate usage. This principle extends beyond students; a widespread understanding of AI empowers individuals to both leverage and scrutinize its applications.
For 2024, my anticipation, or perhaps my aspiration, is a substantial emphasis on education. Back in 1966, Joseph Weizenbaum, the mind behind the ELIZA chatbot, highlighted how machines, though initially captivating, lose their enchantment when their inner workings are explained in plain language. The challenge with today's generative AI lies in its complexity, making it harder to demystify.
I believe it's achievable. My hope is that universities, in their rush to recruit technical AI experts, equally prioritize the hiring of AI ethicists. I trust that media outlets play a pivotal role in dispelling exaggerated claims. I encourage everyone to reflect on their utilization of this technology and its repercussions. Moreover, I wish that tech companies heed well-informed criticisms when shaping the choices that will define the future.
Kentaro Toyama, Professor of Community Information, University of Michigan
In 1970, Marvin Minsky, an AI trailblazer and neural network skeptic, proclaimed to Life magazine that a machine with the general intelligence of an average human would emerge within three to eight years. Fast forward to the present, and the singularity, where AI matches and surpasses human intelligence, is still on the horizon – Minsky's prediction off by a magnitude.
However, forecasting for the next year seems less fraught with peril. What awaits AI in 2024? A fervent race is underway. Since ChatGPT's public release in 2022, a fierce competition for profit, acclaim, and global dominance has erupted, promising not only more potent AI but also a deluge of novel applications.
The pivotal technical query revolves around how swiftly and effectively AI engineers can tackle deep learning's current Achilles' heel – a deficiency in generalized hard reasoning, including deductive logic. Will adjustments to existing neural-net algorithms suffice, or does it demand a paradigm shift, as proposed by neuroscientist Gary Marcus? An army of AI scientists is diligently addressing this challenge, and I anticipate notable progress in 2024.
Simultaneously, as new AI applications emerge, so do novel challenges. Prepare to encounter AI chatbots and assistants engaging in conversations with each other, orchestrating entire dialogues on your behalf, unbeknownst to you. Some of these interactions may go awry, delivering a blend of comedy, tragedy, or both. Deepfakes, AI-crafted images and videos that defy easy detection, are poised to proliferate despite fledgling regulations, potentially wreaking havoc on individuals and democracies. Additionally, expect the rise of unprecedented classes of AI-induced calamities, unimaginable just half a decade ago.
On the front of those raising concerns about AI, figures like Elon Musk and Sam Altman paradoxically persist in developing increasingly potent AI. It's akin to arsonists igniting a fire they then report, imploring authorities to intervene. In alignment with these considerations, my foremost hope for 2024 – albeit progressing slowly – is the establishment of more robust AI regulations on national and international scales.
Anjana Susarla, Professor of Information Systems, Michigan State University
In the year following the introduction of ChatGPT, the evolution of generative AI models is progressing rapidly. Unlike the ChatGPT of yesteryear, which operated on textual prompts for input and generated textual output, the latest breed of generative AI models is designed to be multi-modal. This means their training data extends beyond text from sources like Wikipedia and Reddit to include information from videos on YouTube, songs on Spotify, and other audio and visual content. With the advent of these new-generation multi-modal Large Language Models (LLMs), applications now allow text inputs not only to generate images and text but also audio and video.
Companies are in a frenzied race to develop LLMs adaptable to various hardware and applications, even to run on smartphones. The rise of lightweight LLMs and open-source alternatives might lead to a realm of autonomous AI agents, a prospect for which society may be unprepared.
While these advanced AI capabilities hold transformative potential across domains like business and precision medicine, my primary apprehension lies in the challenges they pose in distinguishing between human-generated and AI-generated content. Additionally, they bring forth new varieties of algorithmic harms that need careful consideration.
The surge of synthetic content produced by generative AI raises concerns about a potential landscape where nefarious actors and organizations exploit this technology to craft synthetic identities and propagate widespread misinformation. The inundation of AI-generated material, strategically designed to circumvent algorithmic filters and recommendation systems, threatens to overwhelm crucial functionalities, including information verification, digital literacy, and the serendipity offered by search engines, social media platforms, and digital services.
The Federal Trade Commission has issued warnings about the potential for fraud, deception, privacy infringements, and other unfair practices facilitated by the ease of AI-assisted content creation. While platforms like YouTube have implemented guidelines for disclosing AI-generated content, there's a pressing need for heightened scrutiny regarding algorithmic harms from entities such as the FTC and lawmakers engaged in crafting privacy protections, like the American Data Privacy & Protection Act.
A bipartisan bill recently introduced in Congress seeks to institutionalize algorithmic literacy as an integral aspect of digital literacy. As AI becomes increasingly intertwined with various aspects of daily life, it becomes evident that the focus should shift beyond algorithms as mere technological components. Instead, a holistic consideration of the contexts in which algorithms operate— encompassing people, processes, and society—is imperative.
