The US state of Texas Education Department will use natural language processing systems in the coming weeks to assess students' answers to standardized tests. The department can save at least $15 million per year.
According to the editorial staff of The Texas Tribune, the underlying machine learning technology is comparable to GPT technology from companies such as OpenAI and uses natural language systems. The system was trained on 3,000 responses to the open-ended questions from the State of Texas Assessment of Academic Readiness test battery, as well as the associated human assessments.
The AI system would be based on this data to learn the properties of the given answers and also to be able to award scores as a human corrector would do. If the system has doubts about its own assessment, the answer to a question will automatically be passed on to a human corrector. This also happens if the system cannot decipher an answer at a certain point. The reporting also shows that parents and students can have the tests reassessed if they do not agree with the result. This procedure will cost $50.
The Texas Education Department will be able to save between $15 and $20 million per year through the measure. The savings are possible because the department has to hire far fewer staff thanks to the AI system. Last year, the department hired 6,000 temporary improvers. This year fewer than 2,000 workers would be needed.
Update, 7:50 PM: 'improve' in the title, introduction and body text changed to 'assess'. Remember the reaction of – peter -.