Shape memory, AI for knowledge, Satellite Internet and what snake is that?

September 10

This week we will look at a material that will remember its’ shape. We also encounter Satellite Internet, an AI that is scanning the entire web for knowledge and an AI that is used to automatically grade exams. Finally we will have a quick look at an App that is being developed to allow us to identify all those spiders and snakes that we encounter in every day life in Australia.

Materials that remember their shape

Researchers at Harvard University have developed a material that will always return to its’ original shape no matter what stimuli is applied to it. The material can be 3D printed into any shape and then it is programmed with reversible shape memory.

The material is made from keratin extracted from leftover Agora wool, a waste product from textile manufacturing. To achieve the shape memory, the material relies on keratin’s hierarchical structure of spring like alpha-helix coils that twist together to form a structure known as a coiled-coil. The combination of the alpha-helix and the connective chemical bonds give the material its’ strength and shape memory.

When a fibre is altered by an external stimulus, the coiled-coils uncoil, allowing the fibre to remain in its’ new position until it is triggered back into its’ original form. The researchers printed a origami star. When it was dunked in water it unfolded and became malleable enough to be transformed into a tight tube. The material remained in this shape even when dry. To return to the original origami star shape, all the researchers did was put the tube into water where it transformed back into the star.

The ability for any shape to be 3D printed and then permanently set, allows complex shapes with structural features to be created down to the micron level. This capability opens up a huge range of uses from textile to tissue engineering.

Satellite Internet

SpaceX is in the early stages of launching up to 27,000 micro satellites into orbit. Those satellites will eventually give the whole world internet access. No matter where you are on the planet you will have access to low latency internet with up to 100Megabit download speed (same as the standard NBN in Australia). Each launch puts 60 satellites into orbit. To date over 700 satellites have been put into orbit. Almost enough to launch a commercial service.

Putting the satellites into orbit is half the equation. We also need a way to receive the data on the ground. For satellite TV, dishes are required. Those dishes are expensive and not always practical. Kymeta has developed the next generation of satellite broadband technology antennas.

Kymeta was founded in 2012 as an outgrowth of brainstorming session with a former Microsoft Chief Technology Officer and a crew of investors (the lead investor being Bill Gates). The company developed the u8 flat panel beam steering antenna for satellite and cellular internet. Kymeta uses metamaterials to steer its antennas electronically to find a signal thus removing the moving parts of traditional satellite dishes.

Early subscription rates won’t be cheap. 1 Gigabyte of data is likely to cost US$999 however as with all technological development, this cost will rapidly fall as uptake and competition intensifies. Early adopters will be military, mining and maritime users that don’t have any alternatives.

Diffbot

There are only 3 companies that scan the entire web for information and data. Those companies are Google, Microsoft and Diffbot. Google and Microsoft scan for their search engines. Diffbot has a different purpose.

Currently language engines used by AI have little sense of what they are actually saying. The recently released GPT-3 is probably the most advanced AI for writing natural language. However Twitter employee Paul Kaisen built a spreadsheet for a range of data for US states (e.g. population, date the state joined the union etc) and used GPT-3 to populate the spreadsheet. Unfortunately GPT-3 made up all of the data, nothing was accurate.

Diffbot wants to change this. Their approach is to read every page on the entire public internet, in multiple languages and extract as many facts as possible from those pages. Diffbot turns what it learns into three part factoids that relate one thing to another: subject, verb and object. Each of these factoids is joined up with billions of others in an interconnected network of facts that is known collectively as a “Knowledge Graph”.

Google uses knowledge graphs to source the information in the boxes shown on the top right hand side of many popular searches. That box contains the basic information about the subject that you have searched.

Knowledge Graphs are difficult to build and maintain. Despite being around for decades they have typically been built by hand. Diffbot is aiming to automate that process.

The Diffbot engine browses the web like a human and thus can see the same facts that we can see. The AI had to learn to scroll down, switch between tabs and click away popups to replicate the human browsing experience. Diffbot crawls continuously and updates its’ knowledge graph every 4 to 5 days. Each month 100 to 150 million new entities are added as new companies are formed and products launched. Out of date data is deleted and new information added.

There are already multiple uses for Diffbot. Researchers can use the Diffbot knowledge graph for free however companies need to pay. For example Adidas and Nike use Diffbot to search the web for people selling fake shoes. The current interactions with Diffbot are via code however a natural language interface is planned. That will allow us to interact with a know-it-all bot with a human like front end.

AI Exam Grading

The dream of every teacher or lecturer is to be able to simply, quickly and easily grade exams. Sounds like a job for AI. Edgenuity, an online platform for virtual learning (lots of learning has become virtual this year) does exactly this. Complete the exam and you receive your grade almost instantaneously. Sounds like a perfect system. However for every perfect system there is a perfect solution.

Dana Simmons, a history professor, was watching her son complete his 7th grade history exam online. He received 50 out of 100. It wasn’t a practice test but a real exam. The thing that intrigued Dana was that her son received the grade instantaneously. She now knew that the system was using an AI to grade the papers. Watching her son compete a few more assignments she developed the hypothesis that the system was looking for keywords in the answers.

Like every parent she wanted to help her son do well so she decided to try and game the system. For every question she entered two short coherent sentences to start and then a laundry list of words that might have some relevance to the answer. That word salad was the key to Dana’s (son’s) grade immediately going from 50% to 100% for every test.

Again, being a helpful parent, Dana put this solution on twitter for all to see.

More than 20,000 schools in the US use the platform for some or all of their tuition in certain subjects. The solution doesn’t work for all subjects (maths for example requires specific correct answers) and all types of questions (e.g. multiple choice questions). The Verge was first to report this exploit and during their research they found multiple students who had used a similar method over the past few years. These students will no doubt, be America’s politicians of the future.

Critterpedia

Living in Australia, we are used to being around the planet’s deadliest spiders and snakes. Most of the time we live in our homes and the spiders and snakes live in the garden. However, occasionally those deadly creatures invite themselves inside and take up residence.

When encountering one of these deadly creatures it is probably wise that we know what sort of snake that you are dealing with prior to reaching around behind the toilet to pull, what you hope is, a harmless python out of its’ hiding place. The Critterpedia App may help.

(FYI this is a highly venomous Tiger Snake)

Critterpedia in collaboration with CSIRO’s Data61, is developing a machine learning engine for automated species identification that will be embedded in an App. Take a photo of the snake or spider, upload it and the app will identify it. The hope is that the app will be educative as well as a useful way to find out which snake or spider you are dealing with. You can sign up to become a phase 1 tester here.

Paying it Forward

If you have a start-up or know of a start-up that has a product ready for market please let me know. I would be happy to have a look and feature the startup in this newsletter. Also if any startups need introductions please get in touch and I will help where I can.

If you have any questions or comments please email me via my website craigcarlyon.com or comment below. 

I would also appreciate it if you could forward this newsletter to anyone that you think might be interested.