Generative AI has seen its rapid progress slow significantly since the 2022 release of ChatGPT. Dr. Jennifer Golbeck, a computer scientist and AI researcher, explained why AI development seems to have hit a plateau and the challenges that lie ahead.
“We’ve built artificial intelligence already that on specific tasks performs better than humans,” said Golbeck during her recent TEDx Talk, but the promise of reaching Artificial General Intelligence (AGI) remains distant. Despite some industry claims that AGI is near, she noted that tools like Google’s AI Search tool, meant to revolutionize search, are “not going super well.”
Golbeck identifies reliability as a critical issue, pointing out that AI is “wrong all the time.” Even as she uses AI tools like ChatGPT for her work, she finds herself correcting it daily due to misunderstandings or lack of context. This problem is exacerbated by AI “hallucinations,” where the technology simply “makes stuff up.” Golbeck emphasizes that solving this isn’t straightforward, as the current AI models are fundamentally designed to generate content rather than retrieve accurate information.
The challenges of AI aren’t just technical. Golbeck has been building AI for over 20 years and notes that “if we train AI on human data, the AI adopts human biases.” However, efforts to mitigate these biases persist, leading to significant concerns about the widespread adoption of AI in decision-making roles.
Ultimately, Golbeck remains skeptical about AGI’s potential to replace human intelligence, stating: “Our human intelligence is not defined by our productivity at work, it’s defined by our ability to connect with other people.” While AI can mimic some aspects of human interaction, it lacks the core qualities that make us human.
