Seeing How Computers “Think” Helps Humans Stump Machines and Reveals Artificial Intelligence Weaknessescmns.umd.edu
- Innovative interface that shows how computers “think” enabled researchers to discover the limitations of AI.
- This led to developing 1,213 easy-for-humans questions, which even the strongest algorithms can’t answer correctly.
- Computers still can’t understand basic wordplay like paraphrasing, distracting language, or unexpected contexts.
- They’re also stumped when reasoning skills are required, like logic, calculation, or building complex conclusions.
- We’re very far from intelligent AI, but this lays out a comprehensive research roadmap for several years to come.