Monday, September 08, 2014

The problem with personalized education

Personalized education has had some spectacular failures lately, in large part due to how tone-deaf the backers have been to the needs of teachers, parents, and students.

The right way to do personalization is to prove you're useful first. Personalization is just a tool. If a new tool doesn't work better than the old tool, it's useless. There's no reason to use personalized education unless it works better than unpersonalized education. A tool needs to be useful.

Teachers are already overworked and, after having been burned too many times on supposedly exciting new technologies that fail to help, correctly are cynical about tech startups coming in and demanding something of them. If some tech startup isn't helping a teacher get something done they need to get done, it's a bad tool and it's useless.

Parents are leery of companies who say they only want to help and what corporations are doing with the data they have on their children, correctly so given all the marketing abuses that have happened in the past.

Kids don't want more boring busywork to do -- they get enough of that already -- and don't see why anything this company is talking about helps them or is useful to them.

If a company wants to succeed in personalized education, it should:
  1. Be useful, noticeably raise test scores
  2. Not require additional busy work
  3. Be optional
  4. Have no marketing whatsoever, only use data to help
I think there are plenty of examples of how this might work. I would like to see a company offer a free Duolingo-like pre-algebra and algebra app that jumps students ahead rapidly as they answer questions correctly and spends more time on similar problems after a question is wrong. The app would be completely optional for students to use, but, when students use it, their test scores increase.

I would like to see a company use the existing standardized tests required by several states, analyze the incorrect answers to identify concepts a student is not understanding, and then print short worksheets targeting only those missed concepts for teachers to hand out to each student. The worksheets would be free and arrive in teachers' mailboxes. If the teacher doesn't want to hand them out, that's not a problem, but test scores go up for the classrooms where the teachers do hand them out. So, even if most teachers don't hand them out at first and most students throw them away at first, over time, more and more teachers will start handing them out and more and more students will do them, as only helps those who do.

In both of these examples, a startup could set up from the beginning to run large scale experiments, showing different problems to different students, and learning what raises test scores, what designs and lesson lengths cause students to stop, what concepts are important and which matter less, what can be taught easily through this and what cannot, what people enjoy, and what works.

When a company comes in and says, "Give us your data, teachers, parents, and kids, and do all this work. Maybe we'll boost your test scores for you later," they're being arrogant and tone-deaf. Everyone responds, "I don't believe you. How about you prove you're useful first? I'm busy. Do something for me or go away." And they're right to do so.

There likely is a way to do personalized education that everyone would embrace. But that way probably requires proving you're useful first. After all, personalization is just a tool.

Wednesday, September 03, 2014

More quick links

More of what caught my attention lately:
  • The overwhelming majority of smartphone users set up their phone once, then barely ever download a new app again ([1] [2])

  • Cool and successful use of speculative execution in cloud computing for games, trading off extra CPU and bandwidth for the ability to hide network latency ([1])

  • Infrared vision on your phone ([1] [2])

  • How easy is it to get people to memorize hard-to-crack random 56-bit passwords, equivalent to about 12 random letters or 6 words? ([1] [2])

  • Desalination needs warm water, data centers need to be cooled, why not put them together? Clever idea. ([1])

  • It's easy to overhype this, but it's still pretty cool, transmitting data (0 and 1 bits) directly brain-to-brain without implants (using magnetic stimulation of the brain and EEG reading of the brain, both from the surface of the scalp) with relatively low error rates (5-15%). Data rates are extremely low at 2-3 bits/minute, but it's still interesting that it's possible at all. ([1])

  • Xiaomi's remarkable iPhone clone ([1])

  • Has Amazon sold less than 35k Fire phones? ([1] [2])

  • Facebook publishes a paper which details how its ad targeting works and suggests they will be doing more personalization in the future ([1] [2])

  • "Having a multiyear project with no checks along the way and the promise of one big outcome is not a highly successful approach, in or outside government" ([1] [2])

  • More evidence patent trolls cause real harm. Trolled firms "dramatically reduce R&D spending". ([1])

  • "Using nothing more than a laptop ... [they could] alter the normal timing pattern of the [traffic] lights, turning all the lights along a given route green, for instance, or freezing an intersection with all reds" ([1])

  • Interesting data visualization showing how CD took over in music sales, then got replaced by downloads, all over the last two decades or so ([1])

  • Neat charts on how the strike zone expands on 3 ball counts and contracts on 2 strike counts ([1])

  • Cute SMBC comic on "What is the fastest animal?" ([1])

  • Great SMBC comic on job interviews ([1])