Emily Bender, professor of computational linguistics, discusses the true intelligence of generative AI systems like GPT-4. They debate whether the systems truly comprehend or just respond to cues. They explore how GPT-4 identified fraud in the Enron scandal and the limitations of AI in human-level reasoning. The speakers also touch on the emotional connection to AGI and the challenge of separating machines and humans when interacting with AI.
Generative AI systems like ChatGPT excel at pattern matching and stringing words together, but lack true understanding of the meaning of words and genuine cognitive processes.
The impressive performance of AI systems in tasks like legal reasoning often leads to misconceptions that they truly understand and reason like humans, but these achievements are based on pattern matching and statistical inference rather than true intelligence.
Deep dives
The Limitations of ChatGPT and AI Reasoning
Despite the impressive capabilities of generative AI systems like ChatGPT, some experts argue that these systems are not actually reasoning or thinking like humans. AI systems like ChatGPT excel at pattern matching and stringing words together, but they lack true understanding of the meaning of the words they use. These systems can pass exams and solve problems, but often they rely on surface-level cues and memorized patterns rather than genuine cognitive processes. The skepticism surrounding the reasoning capabilities of AI is supported by experiments in which ChatGPT fails when faced with complex reasoning tasks and struggles to provide explanations for its answers. The debate over the intelligence of chatbots raises questions about what constitutes true intelligence and whether artificial general intelligence is achievable.
The Impressive Performance of ChatGPT
While there are doubts about the reasoning abilities of AI systems like ChatGPT, there is no denying the impressive performance of these systems in tasks such as passing the bar exam or displaying legal reasoning. ChatGPT's ability to provide coherent and plausible responses often leads to the misconception that it truly understands and reasons like a human. However, experts emphasize that these achievements are based on pattern matching and statistical inference, rather than genuine human-like cognition. The large-scale training data and computational power behind ChatGPT allow it to generate text that appears to be intelligent and reasoned, but this does not necessarily indicate true intelligence.
The Hype and Misunderstandings Surrounding AI
The hype surrounding AI, fueled by marketing claims and media reports, often leads to misconceptions about the reasoning capabilities of AI systems. Some researchers argue that the claims made by AI companies and the enthusiasm of investors may be driven by wishful thinking rather than scientific evidence. Critics, like linguistics professor Emily Bender, underscore the importance of distinguishing between the impressive outputs of AI systems and genuine human-like reasoning. Deep-seated anthropomorphism and emotional connections to the idea of AI and AGI can cloud the judgment of both AI researchers and the general public, creating an environment where exaggerated claims and high expectations prevail.
Are generative AI systems such as OpenAI’s ChatGPT really intelligent? Large language models such as GPT 4 appear to use human-level cognitive abilities when they engage in legal reasoning, write essays or solve complex problems. Hosts John Thornhill and Madhumita Murgia speak to Emily Bender, professor of computational linguistics at the University of Washington, to find out what’s really happening under the hood, and also hear from Pablo Arredondo of CaseText, which develops AI tools for lawyers; influential computer scientist Melanie Mitchell, professor at the Santa Fe Institute, and Konstantin Arkoudas, an AI expert who’s worked on Amazon’s Alexa.
Tech Tonic is presented by Madhumita Murgia and John Thornhill. Senior producer is Edwin Lane and the producer is Josh Gabert-Doyon. Executive producer is Manuela Saragosa. Sound design by Breen Turner and Samantha Giovinco. Original music by Metaphor Music. The FT’s head of audio is Cheryl Brumley.