Google Alpha Earth

Google Alpha Earth is a new geospatial AI model that acts as a virtual satellite: it assimilates all kinds of geospatial data & does the processing for you. Then, you just ask & get ready-to-use AI-generated 'maps' of whatever you want. No more working with dozens of raw datasets!

Example: glacier mapping. In thed old days you would have spent considerable time finding useful cloud-free images, do some kind of processing, and then run some kind of index or band-ratio calculations. Now you can just ask: give me the size of the Humboldt Glacier over the last 10 years.

For now the focus seems to be on physical data collected by sensors, but I would assume that human data from will be included in the future.

More here: AlphaEarth Foundations helps map our planet in unprecedented detail

Deep Research Tools from OpenAI & Google

These tools can basically synthesize the information from 100s of websites into nice and tidy cited literature reviews. And that's great for many basic science & technology topics where there is enough information available out there on reputable online sources.

But the notion that this can replace a classic scientific literture review of meta-analysis is incorrect. First, these AIs do not have access to the actual scientific literature (which tends to sit behind paywalls). Second, there is a difference between a summary, a review, and a synthesis. Sure, a competent & trained amateur such as an AI can summarize, but reviewing & synthesizing requires a familiarity with the subject matter that can only be gained by years and years of deep immersion (aka. 10,000 hours...).

XR, VR, AR in scientific research

Good article here: Why scientists are delving into the virtual world. But, it is clear tha XR, VR, AR, etc. are not really used (yet) for actual scientific research, but other ancillary functions such as:

• Virtual meetings.
• Training students.
• Educational outreach.
• Controlling your lab machines / robots remotely.

Here's a good quote: "VR offers a repeatable, standardized clinical training method that is not reliant on the quality of the trainer." Or, being able to access an actual simulation lab or medical dummy.

Now all you need is a set of VR goggles & a super-fast Internet connection!

AI-Generated Maps

Not sure I would call these 'maps' - to me maps really need to represent reality.

• Latent Diffusion (converts text to images)
• AI-generated maps: Copyright, ownership, metadata (from Spatial Reserves)

The Memex

In his essay 'As We May Think' (from July 1945!), Vannevar Bush describes a future device called 'the memex' and thereby (conceptually) invented Google and smart gizmos. Here, humans create their own associative joins bewteen items of information to build associative trails that can be shared from one memex to another.

Great read, especially Sections 6 and 7.

Machine Learning / AI

Or as we also call it: voodoo. Something that some people do somehow using this thing called Big Data that is out there, but nobody can see (= dark matter). Here are a couple of days to do this voodoo yourself and makes a lot of money.

• Ludwig (by Uber) available on GitHub.
• Northstar by DSAIL at MIT. 

How Machines Learn

Some interesting and scary similarities to the trends we see in Higher Education right now!

AMOS

Need a webcam images? AMOS, the Archive of Many Outdoor Scenes, can help with that. AMOS is a collection of long-term time lapse imagery from publicly accessible outdoor webcams around the world. Images like that are increasingly used, in combination with AI/machine learning, to derive quantitative data.

Smart Mapping

Or is it Artificial Intelligence? First Draft GIS creates a draft map for you - all you need to do is provide the data. That does not sound all that impressive, but it can map from news articles, Word Documents, PDFs, Excel Spreadsheets, or any type of webpage - just copy/paste the URL. Try this data set for example.

Mapping Cities with AI

Here are three interesting examples of using Google Street View and crowdsourcing to train AI algorithms to 'map' conditions in cities.

• Seen on Maps Mania: Project Sidewalk is a classic example of crowdsourcing, now used for training machine learning algorithms. Cool and also somewhat scary.
• Place Pulse has been around for a while and aims to quantitatively recognize which areas of a city are perceived as wealthy, modern, safe, lively, active, unique, central, adaptable or family friendly.
• StreetChange takes a similar approach, albeit focused to detecting urban change from Street View photos taken several years part. 

What is machine learning?

Here's a nice visual introduction, aptly called A Visual Introduction to Machine Learning. This is nothing new to those familiar with remote sensing and image analysis - machine learning is really just a new fancy word for image classification (both supervised and unsupervised).