A fundamental change in K-12 science education has created a huge opportunity for large language models. Not today’s models but reasonable adaptations of them.
Students now mostly do research investigations so they need to look a lot of stuff up, stuff they can understand. This is hard because today’s search engine cannot do grade level search, but LLM’s should be able to if properly designed.
First a bit of background. What is taught in American public school science education from kindergarten through high school (K-12) is specified by what are called science standards. The standards say what topics will be taught at each grade level, as it is called. These levels are typically kindergarten, grades 1 through 5, plus middle and high school, making 8 levels.
Standards used to be knowledge based, specifying in detail what basic scientific facts the student should learn. But over the last 10 years a new set of standards have taken over called the Next Generation Science Standards (NGSS). These standards are activity based so they instead specify investigations that the student must undertake. What they learn depends on what they find.
The National Science Teachers Association says 20 states have adopted the full NGSS while 29 others have adopted standards that are based on the Framework of the NGSS. A map showing this by state is here: https://www.nsta.org/science-standards.
By way of example here is how NGSS teaches about electricity in grades 3 and 4 (which is all that is taught on this topic until high school):
NGSS Thirds grade physical science – forces and interactions (3-PS2):
“3-PS2-3. Ask questions to determine cause and effect relationships of electric or magnetic interactions between two objects not in contact with each other.
[Clarification Statement: Examples of an electric force could include the force on hair from an electrically charged balloon and the electrical forces between a charged rod and pieces of paper; examples of a magnetic force could include the force between two permanent magnets, the force between an electromagnet and steel paperclips, and the force exerted by one magnet versus the force exerted by two magnets. Examples of cause and effect relationships could include how the distance between objects affects strength of the force and how the orientation of magnets affects the direction of the magnetic force.] [Assessment Boundary: Assessment is limited to forces produced by objects that can be manipulated by students, and electrical interactions are limited to static electricity.]”
NGSS Fourth grade physical science – energy (4-PS3):
“4-PS3-2. Make observations to provide evidence that energy can be transferred from place to place by sound, light, heat, and electric currents.
[Assessment Boundary: Assessment does not include quantitative measurements of energy.]
4- PS3-4. Apply scientific ideas to design, test, and refine a device that converts energy from one form to another.
[Clarification Statement: Examples of devices could include electric circuits that convert electrical energy into motion energy of a vehicle, light, or sound; and, a passive solar heater that converts light into heat. Examples of constraints could include the materials, cost, or time to design the device.] [Assessment Boundary: Devices should be limited to those that convert motion energy to electric energy or use stored energy to cause motion or produce light or sound.]”
Each of these tasks typically involves looking up a lot of stuff. But how does, say, a third grader find information on static electricity that is written at their grade level? Pretty much everything that a regular search on static electricity will return will assume a lot more knowledge than a typical third grader has.
Here is where AI large language models could be a great help because each grade level is also a language level. AI can be trained to recognize and deliver material written at a given grade level.
I used this fact some years ago when I developed a grade level search system for the US Energy Department. We were building ScienceEducation.gov, a central repository for federal agency educational materials along the lines of the successful science.gov. It was actually fielded but then the funding was pulled and it went away.
My Teacher Team cataloged all the technical language taught at each grade level. The computer ranked each document based on the highest grade language used. It worked well.
Grade levels are specific technical language communities. Students know the technical terms they have been taught but not the ones they have yet to be taught.
Mind you we would like to do without political bias in searches and that is something the States might be able to insist on, as they have done pretty well with textbooks.
There are roughly 50 million K-12 students in America which is a lot of potential users for an AI grade level search system. It would sure help them out under NGSS.