Google, AI Announcements, and the Future of Learning

For real transformation, we'll need learning-focused solutions instead of tech-focused ones

Was this forwarded to you by a friend? Sign up, and get your own copy of the news that matters sent to your inbox every week. Sign up for the On EdTech newsletter. Interested in additional analysis? Try with our 30-day free trial and Upgrade to the On EdTech+ newsletter.

I didn’t really want to write this post. There are already so many things written about AI in EdTech, but we do get many questions here about how to interpret education-related news, and I thought it would be useful to lay out my thoughts, particularly from a lens of working with educators. Furthermore, Google has big footprint in education, and what they do with AI will help shape what EdTech-specific vendors do.  

Google I/O

Three weeks ago, Google held their annual developer-focused meeting, Google I/O, and rolled out many features coming (at some point) to a browser or device near you. At the heart of what Google announced (and what Open AI announced the day before) is the arrival of omnichannel AI. Instead of interacting with an AI-powered chatbot by means of text, you can now do so by voice or by pointing your camera at an object, among other methods, and importantly in parallel and with minimal latency.  

There were a lot of other new features and approaches - two hours’ worth - and education was an important focus. Additionally, some of the non-education specific features will likely have a big impact on education, at least eventually.

Here is the education section (starting at 1:45:14).

Many of the education-specific features are based on Google’s LearnLM – this is part of the Gemini’s family of large language models but fine-tuned for education. Google is working with Columbia Teachers College, MIT Raise, ASU, and Khan Academy in an attempt to make this as useful and engaging for learners as possible.  Google is also working to build tools and applications based on LearnLM, which you can read about in their fascinating (but dense) technical paper.

At the event, Google announced several tools based on LearnLM. Claire Zau from GSV did a great job of summarizing the key education features and use cases [emphasis original].

Circle to Search: Android users can now work through math or physics word problems directly from their phones and tablets. Later this year, users will be able to solve complex problems involving symbolic formulas, diagrams, graphs and more.

Gems for Education: Google is creating pre-made Gems (customized versions of Gemini, like a GPT), including one called Learning Coach. Learning Coach offers step-by-step study guidance and leverages practice and memory techniques to build understanding, rather than simply sharing the answer.

YouTube Integration: LearnLM will make educational videos more interactive. Users can ask clarifying questions, get helpful explanations, and take quizzes. Given Gemini model’s long-context capabilities, these features will even work for long lectures or seminars…

Google Classroom: LearnLM will help teachers simplify and improve the lesson planning process to meet the individual needs of students. A demo showcased a tool that can simplify or re-level content for a target grade level. Future features may help teachers discover new ideas and unique activities.

Google also announced NotebookLM, which takes content you provide and produces summaries and quizzes as well as audio and audio discussions to explain complex topics.

These announcements were impressive, and anyone who has been remotely near a classroom could think of many additional applications and ways that the new technologies could be used to great effect. On social media, many people were enthusiastic.

This sort of education coverage contrasted with some of the more general coverage that I sampled.  Several observers found the event to be low energy and a bit cringey, with Google seemingly more focused on establishing its chops as an agile AI company after some months of speculation that it was being left behind. And fitting in as many Taylor Swift jokes in as they could.

The fact that the education features were better received than general updates stems in part from the fact that the use cases made more sense. Contrast the education examples from Zau’s summary with the forced nature of general examples, such as pointing the camera at objects and asking what they were, and doing odd tasks like swapping lines and columns in an Excel spreadsheet.

But despite the excitement, some of the positive coverage of the Google announcements in education gloss over a flawed perspective that will need to be changed if we are to see significant impacts in the long term for education.

Announcement ≠ Delivery

Claire Zau reminded us in her notes that just because something was announced doesn’t mean that it is available. While a few features have been released, many others are slated for delivery at later (or unspecified) dates. NotebookLM is listed as “Experimental” and quite a few others are part of Labs or Project (i.e. we are going to need to wait a while before we can use them). To those of us more accustomed to LMS-style feature timelines, this is a tad frustrating, especially given Google’s history of being long on promises but short on delivery. But it is important to bear in mind that this is Google’s view of the next wave – or the near future -- of AI integration in EdTech

Fascinated but Underwhelmed

This latter fact is important, because I find a lot of what Google introduced somewhat underwhelming, once you get past the omnichannel part of it. A good part of what is in NotebookLM looks a lot like what every other small vendor demonstrating at the ASU+GSV AIR show – just without the voice. Similarly, the announcement of instructors being able to use LearnLM to support lesson planning is a nice feature, but not that much beyond what Anthology and increasingly others have been doing within LMSs.

Given that a lot of what Google announced is still a ways away, the fact that the features are so similar to tools we already have is frustrating. I fear we have been seduced by the flirty voices and interaction and overlooked the heart of what the tools are really doing. To be sure, the omnichannel aspect is compelling, but if it is doing the same task, how much of an innovation is it really?

Tutoring über alles?

A second problem I have with the OpenAI, but especially the Google announcements, is the overwhelming emphasis on tutoring as the killer app. This emphasis bothers me for several reasons.

First, Google Gemini applications for education are currently only available for students over the age of 18. Given the centrality of tutoring in the discourse about AI, this is odd. You can use AI in all these cool ways, but you can’t actually do it yet unless you are a university student (or have the Google equivalent of Sal Khan sitting next to you).

Second, I think the tutor emphasis is based on a misreading of Bloom’s Two-Sigma Problem, which we see a lot of in EdTech. Benjamin Bloom argued in 1984 that tutoring was able to improve student achievement by two standard deviations (or two sigmas in statistics speak), which is enormous and bigger than almost any other teaching intervention. But too often people look at Bloom’s work and say hey, tutoring produces this big effect and has been unreachable to most people because of cost, and AI changes that.

This is based on a superficial reading of Bloom and, as Paul von Hippel argues, it ignores some of the evidence underlying Bloom’s research which paints a far more complex picture.

Two sigmas is an enormous effect size. As Bloom explained, a two-sigma improvement would take a student from the 50th to the 98th percentile of the achievement distribution. If a tutor could raise, say, SAT scores by that amount, they could turn an average student into a potential Rhodes Scholar.

Focusing too much on tutoring has not replicated the two sigma results.

More recently, a 2020 meta-analysis of randomized studies by Andre Nickow, Philip Oreopoulos, and Vincent Quan found that the average effect of tutoring was 0.37 standard deviations, or 14 percentile points—“impressive,” as the authors wrote, but far from two sigmas. Among 96 tutoring studies the authors reviewed, none produced a two-sigma effect.

Long and short, tutoring can be overemphasized while missing the specific context of instruction and testing and feedback.

But even beyond the misreading Bloom problem, thus far I am unconvinced that the kinds of tutoring currently offered via AI matches the concept of watching a student’s thought processes and identifying the core issues they aren’t understanding. Instead, AI tutoring today seems to consist of breaking down problems into component parts and explaining the components. This is no doubt helpful, but it is not tutoring in the true sense of the word.

I also suspect that AI tutors could be used by different types of students in different ways. Good students might use these tools to do even better (much as with other EdTech), but the improvements will be small because they are good students and room for improvement is limited. Poor students might use these tools because they are seen as a route to the answer, and it won’t help with core understanding. The improvements will be small and transitory.

The issue at the heart of the matter

This all stems from an issue that Phil identified in a recent podcast (much as it pains me to admit that). Phil argued that Google:

is stepping too far outside of their swim lane into areas where you have to say, I'm not saying Google can't do education, but they should be providing…tools so that people much closer to the educational problem can craft solutions… [it] shouldn't be Google and OpenAI answering that, that should be, should be higher education institutions. It should be edtech vendors steeped in the field understanding what's possible and crafting effective solutions that help student learning.

At first when Phil said this I thought, hold on, Google is an EdTech company, simply measured in numbers of students using tools in education. But Google is an EdTech company in spite of its true, primary nature as a tech company.

The education-specific AI solutions showcased at Google I/O were tech-focused solutions looking for learning examples, not learning-focused solutions looking for effective technology usage, and this focus is reflected in the kinds of use cases that are emphasized. These use cases are long on technology and short on pedagogy, especially pedagogically innovative uses of technology.

I think Michael Feldstein’s commentary on the research paper captured this problem and is well worth reading.

As I read some of the early literature, I see an all-too-familiar pattern: technologists build the platforms, data scientists decide which data are important to capture, and they consult learning designers and researchers. However, all too often, the research design clearly originates from a technologist’s perspective, showing relatively little knowledge of detailed learning science methods or findings. A good example of this mindset’s strengths and weaknesses is Google’s recent paper, “Towards Responsible Development of Generative AI for Education: An Evaluation-Driven Approach“. It reads like a paper largely concieved by technologists who work on improving generative AI and sharpened up by educational research specialists they consulted with after they already had the research project largely defined.

Parting thoughts

The closest thing I have to a truism in EdTech is that we tend to overstate the impact of a new tool in the short term and understate it in the long term. What we are seeing now with generative AI are short-term solutions. In the longer term, educators and EdTech vendors will take the tools developed by Google, OpenAI, Microsoft, and others and apply them in ways that make more sense and address a real “job to be done” in interesting and unexpected ways. I just hope that some of the mistakes in focus of this early phase don’t delay that, or send us down some unproductive roads.

The main On EdTech newsletter is free to share in part or in whole. All we ask is attribution.

Thanks for being a subscriber.