OpenAI released its Transformer-Based Large Language Model (LLM) called “ChatGPT” in late 2022. Contrary to OpenAI staff’s expectations, ChatGPT became the fastest-growing web-based application in history, rising to 100 million active users in two months (surpassed only by Meta’s Threads). Early public impressions of ChatGPT had both sublime qualities and signs of disaster. In February 2023, Henry Kissinger, Eric Schmidt and Daniel Huttenlocher wrote that generative artificial intelligence (AI) is comparable to the intellectual revolution ushered in by the printing press, this time consolidating and ‘distilling’ the repository of human knowledge. In March 2023, foreseeing extinction-level risks, Eliezer Yudkowsky implored the world’s governments and militaries to halt the AI project and be “willing to destroy a rogue data center with an airstrike.”
These first impressions represent two ends of a spectrum, but the logic that bridges the gap between them is commonplace in technology policy analysis: Personal impressions of generative AI seep into the background assumptions against which policy analyzes are conducted. When fundamental assumptions are not questioned, it is easy to fall into the trap of extrapolating future technological marvels from current technological conditions. Technology policy analysts of all stripes are doing excellent work, but it’s time to identify the gaps in our reasoning and aim higher individually and collectively.
An example shows the general trend. Paul Scharre of the Center for a New American Security avoids the future of AI in his book “Four Battlegrounds,” which is generally a treasure trove of insights, but says, “Building larger, more diverse data sets yields more robust models. Multimodal data sets, text “can help create models that can relate concepts represented in multiple formats such as images, video, and audio.” This expectation is based on the idea that scaling up AI systems (increasing their internal capabilities and training datasets) will lead to new capabilities, a positive reference to Richard Sutton’s famous argument about the benefits of such techniques in “The Bitter Lesson”.
Shortly thereafter, Microsoft’s researchers helped set the tone for overly optimistic claims about the future of LLMs with their provocatively titled “Sparks of Artificial General Intelligence” paper on GPT-4. It’s not hard to see how one’s personal impression of GPT-4 could lead to an equivalent feeling of “We’re on the verge of something big here.” But this is no justification for allowing these emotion-related assumptions to fester in one’s analysis.
Extensive research highlights the limitations of Master and other Transformer-based systems. Hallucinations—credible but factually inaccurate statements—continue to plague graduate students; some researchers argue that these are innate features of this technology. Voters using chatbots for basic information about the 2024 elections can easily be misinformed about hallucinated polling places and other inaccurate or outdated information, according to a recent study. Other research shows that graduate students’ abilities to construct abstractions and generalize them lag behind humans; The reasoning abilities of multimodal systems are a similar story. While OpenAI’s latest development, the text-to-video generator “Sora,” is notable for its realism, it invents objects and people out of thin air and fails to adhere to real-world physics.
So much for the idea that new methods like images and video will lead to the reliable, robust, and explainable AI systems we desire.
None of these show that there is Only excitement in the world of technology. Carnegie’s Matt O’Shaughnessy correctly notes that talk of “superintelligence” is likely to negatively impact policymaking Because Fundamental limitations of machine learning. Additionally, the Biden administration’s sweeping executive order on AI in October 2023 was more varied in tone than expected, while dramatically invoking the Defense Production Act to allow the monitoring of certain computationally powerful AI systems.
However, the problem we describe here is not an exaggerated problem. by himself. Hype is a conclusion Getting stuck in analytical frameworks that are easily ignored in favor of rapid publications and individual or organizational self-promotion. Lest we mistakenly believe that this is a weird graduate school trend, the disappointment of AI-powered and autonomous drones on the battlefield in Ukraine should raise eyebrows about the alleged pace of fundamental breakthroughs occurring in 2023. Moreover, it is easier to find nuances. In the field of quantum information science, however, at the same time that the crown jewel of quantum computing begins to see its future downgraded, little individual or collective reflection seems to be occurring.
However, generative AI today is starting to look like a parody of Mao’s Permanent Revolution; The transformation of this technology into a human-like “general” intelligence or some other technological wonder of imagination is always one model upgrade away and cannot be allowed. Bowing to challenges from regulatory agencies or popular movements.
The takeaway here is that policy analysts make choices when evaluating technology. Preferring certain assumptions over others gives the analyst a range of possible policy options at the expense of others. It is inevitable that individuals will have a first impression of new technologies and can be a source of diversity of opinion. The problem of policy analysis arises when practitioners fail to pour their first (or second, or third, etc.) impressions into a common melting pot that exposes unstable ideas to high-temperature intellectual critique, thus leading them to the articulation of specific policy challenges and solutions. wholesale solutions without unnecessarily neglecting other possibilities.
Policy analysis is often a mix of components from industry, domestic politics and international relations. Merely identifying that a policy problem exists is not enough new but it arises from the intuitive connection between the needs and values of a society and the expected or actual effects of developments within or outside its borders. This intuition (we all have it) should be the focus of our honest and collective examination.
Vincent J. Carchidi He is a Non-Resident Researcher in the Strategic Technologies and Cybersecurity Program at the Middle East Institute. He is also a Foreign Policy member of America’s Next Generation Initiative 2024 Cohort.
Head to The Hill for the latest news, weather, sports and streaming video.